lab2/lab_2.ipynb

{
   "cells": [
      {
         "cell_type": "markdown",
         "metadata": {
            "pycharm": {
               "name": "#%% md\n"
            }
         },
         "source": [
            "# Klasyfikacja niezbalansowana, klasyfikatory zespołowe i wyjaśnialna AI"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "pycharm": {
               "name": "#%% md\n"
            }
         },
         "source": [
            "## Wykorzystanie Google Colab\n",
            "\n",
            "Jeśli korzystasz z Google Colab skopiuj plik `feature_names.json` do katalogu głównego projektu.\n",
            "\n",
            "[![Open in Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/apohllo/sztuczna-inteligencja/blob/master/lab2/lab_2.ipynb)"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "source": [
            "## Ładowanie i eksploracja danych"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "source": [
            "Na tym laboratorium wykorzystamy zbiór danych [Polish companies bankruptcy](https://archive.ics.uci.edu/ml/datasets/Polish+companies+bankruptcy+data). Dotyczy on klasyfikacji, na podstawie danych z raportów finansowych, czy firma zbankrutuje w ciągu najbliższych kilku lat. Jest to zadanie szczególnie istotne dla banków, funduszy inwestycyjnych, firm ubezpieczeniowych itp., które z tego powodu zatrudniają licznie data scientistów. Zbiór zawiera 64 cechy, obliczone przez ekonomistów, którzy stworzyli ten zbiór, są one opisane na podlinkowanej wcześniej stronie. Dotyczą one zysków, posiadanych zasobów oraz długów firm.\n",
            "\n",
            "Ściągnij i rozpakuj dane (`Data Folder` -> `data.zip`) do katalogu `data` obok tego notebooka. Znajduje się tam 5 plików w formacie `.arff`, wykorzystywanym głównie przez oprogramowanie Weka. Jest to program do \"klikania\" ML w interfejsie graficznym, jakiś czas temu popularny wśród mniej technicznych data scientistów. W Pythonie ładuje się je za pomocą bibliotek SciPy i Pandas.\n"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {},
         "source": [
            "Jeśli korzystasz z Linuksa możesz skorzystać z poniższych poleceń do pobrania i rozpakowania tych plików."
         ]
      },
      {
         "cell_type": "code",
         "execution_count": 2,
         "metadata": {
            "ExecuteTime": {
               "end_time": "2023-10-17T12:09:50.173669Z",
               "start_time": "2023-10-17T12:09:24.311939Z"
            },
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "outputs": [],
         "source": [
            "#!mkdir -p data\n",
            "#!wget https://archive.ics.uci.edu/static/public/365/polish+companies+bankruptcy+data.zip -O data/data.zip"
         ]
      },
      {
         "cell_type": "code",
         "execution_count": 3,
         "metadata": {
            "ExecuteTime": {
               "end_time": "2023-10-17T12:10:04.017260Z",
               "start_time": "2023-10-17T12:09:57.831756Z"
            },
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "outputs": [],
         "source": [
            "#!unzip data/data.zip -d data"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "source": [
            "\n",
            "W dalszej części laboratorium wykorzystamy plik `3year.arff`, w którym na podstawie finansowych firmy po 3 latach monitorowania chcemy przewidywać, czy firma zbankrutuje w ciągu najbliższych 3 lat. Jest to dość realistyczny horyzont czasowy.\n",
            "\n",
            "Dodatkowo w pliku `feature_names.json` znajdują się nazwy cech. Są bardzo długie, więc póki co nie będziemy z nich korzystać."
         ]
      },
      {
         "cell_type": "code",
         "execution_count": 4,
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%%\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "outputs": [],
         "source": [
            "import json\n",
            "import os\n",
            "\n",
            "from scipy.io import arff\n",
            "import pandas as pd\n",
            "\n",
            "data = arff.loadarff(os.path.join(\"data\", \"3year.arff\"))\n",
            "\n",
            "with open(\"feature_names.json\") as file:\n",
            "    feature_names = json.load(file)\n",
            "\n",
            "X = pd.DataFrame(data[0])"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "pycharm": {
               "name": "#%% md\n"
            }
         },
         "source": [
            "Przyjrzyjmy się teraz naszym danym."
         ]
      },
      {
         "cell_type": "code",
         "execution_count": 5,
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%%\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "outputs": [
            {
               "data": {
                  "text/html": [
                     "<div>\n",
                     "<style scoped>\n",
                     "    .dataframe tbody tr th:only-of-type {\n",
                     "        vertical-align: middle;\n",
                     "    }\n",
                     "\n",
                     "    .dataframe tbody tr th {\n",
                     "        vertical-align: top;\n",
                     "    }\n",
                     "\n",
                     "    .dataframe thead th {\n",
                     "        text-align: right;\n",
                     "    }\n",
                     "</style>\n",
                     "<table border=\"1\" class=\"dataframe\">\n",
                     "  <thead>\n",
                     "    <tr style=\"text-align: right;\">\n",
                     "      <th></th>\n",
                     "      <th>Attr1</th>\n",
                     "      <th>Attr2</th>\n",
                     "      <th>Attr3</th>\n",
                     "      <th>Attr4</th>\n",
                     "      <th>Attr5</th>\n",
                     "      <th>Attr6</th>\n",
                     "      <th>Attr7</th>\n",
                     "      <th>Attr8</th>\n",
                     "      <th>Attr9</th>\n",
                     "      <th>Attr10</th>\n",
                     "      <th>...</th>\n",
                     "      <th>Attr56</th>\n",
                     "      <th>Attr57</th>\n",
                     "      <th>Attr58</th>\n",
                     "      <th>Attr59</th>\n",
                     "      <th>Attr60</th>\n",
                     "      <th>Attr61</th>\n",
                     "      <th>Attr62</th>\n",
                     "      <th>Attr63</th>\n",
                     "      <th>Attr64</th>\n",
                     "      <th>class</th>\n",
                     "    </tr>\n",
                     "  </thead>\n",
                     "  <tbody>\n",
                     "    <tr>\n",
                     "      <th>0</th>\n",
                     "      <td>0.174190</td>\n",
                     "      <td>0.41299</td>\n",
                     "      <td>0.14371</td>\n",
                     "      <td>1.3480</td>\n",
                     "      <td>-28.9820</td>\n",
                     "      <td>0.60383</td>\n",
                     "      <td>0.219460</td>\n",
                     "      <td>1.1225</td>\n",
                     "      <td>1.1961</td>\n",
                     "      <td>0.46359</td>\n",
                     "      <td>...</td>\n",
                     "      <td>0.163960</td>\n",
                     "      <td>0.375740</td>\n",
                     "      <td>0.83604</td>\n",
                     "      <td>0.000007</td>\n",
                     "      <td>9.7145</td>\n",
                     "      <td>6.2813</td>\n",
                     "      <td>84.291</td>\n",
                     "      <td>4.3303</td>\n",
                     "      <td>4.0341</td>\n",
                     "      <td>b'0'</td>\n",
                     "    </tr>\n",
                     "    <tr>\n",
                     "      <th>1</th>\n",
                     "      <td>0.146240</td>\n",
                     "      <td>0.46038</td>\n",
                     "      <td>0.28230</td>\n",
                     "      <td>1.6294</td>\n",
                     "      <td>2.5952</td>\n",
                     "      <td>0.00000</td>\n",
                     "      <td>0.171850</td>\n",
                     "      <td>1.1721</td>\n",
                     "      <td>1.6018</td>\n",
                     "      <td>0.53962</td>\n",
                     "      <td>...</td>\n",
                     "      <td>0.027516</td>\n",
                     "      <td>0.271000</td>\n",
                     "      <td>0.90108</td>\n",
                     "      <td>0.000000</td>\n",
                     "      <td>5.9882</td>\n",
                     "      <td>4.1103</td>\n",
                     "      <td>102.190</td>\n",
                     "      <td>3.5716</td>\n",
                     "      <td>5.9500</td>\n",
                     "      <td>b'0'</td>\n",
                     "    </tr>\n",
                     "    <tr>\n",
                     "      <th>2</th>\n",
                     "      <td>0.000595</td>\n",
                     "      <td>0.22612</td>\n",
                     "      <td>0.48839</td>\n",
                     "      <td>3.1599</td>\n",
                     "      <td>84.8740</td>\n",
                     "      <td>0.19114</td>\n",
                     "      <td>0.004572</td>\n",
                     "      <td>2.9881</td>\n",
                     "      <td>1.0077</td>\n",
                     "      <td>0.67566</td>\n",
                     "      <td>...</td>\n",
                     "      <td>0.007639</td>\n",
                     "      <td>0.000881</td>\n",
                     "      <td>0.99236</td>\n",
                     "      <td>0.000000</td>\n",
                     "      <td>6.7742</td>\n",
                     "      <td>3.7922</td>\n",
                     "      <td>64.846</td>\n",
                     "      <td>5.6287</td>\n",
                     "      <td>4.4581</td>\n",
                     "      <td>b'0'</td>\n",
                     "    </tr>\n",
                     "    <tr>\n",
                     "      <th>3</th>\n",
                     "      <td>0.024526</td>\n",
                     "      <td>0.43236</td>\n",
                     "      <td>0.27546</td>\n",
                     "      <td>1.7833</td>\n",
                     "      <td>-10.1050</td>\n",
                     "      <td>0.56944</td>\n",
                     "      <td>0.024526</td>\n",
                     "      <td>1.3057</td>\n",
                     "      <td>1.0509</td>\n",
                     "      <td>0.56453</td>\n",
                     "      <td>...</td>\n",
                     "      <td>0.048398</td>\n",
                     "      <td>0.043445</td>\n",
                     "      <td>0.95160</td>\n",
                     "      <td>0.142980</td>\n",
                     "      <td>4.2286</td>\n",
                     "      <td>5.0528</td>\n",
                     "      <td>98.783</td>\n",
                     "      <td>3.6950</td>\n",
                     "      <td>3.4844</td>\n",
                     "      <td>b'0'</td>\n",
                     "    </tr>\n",
                     "    <tr>\n",
                     "      <th>4</th>\n",
                     "      <td>0.188290</td>\n",
                     "      <td>0.41504</td>\n",
                     "      <td>0.34231</td>\n",
                     "      <td>1.9279</td>\n",
                     "      <td>-58.2740</td>\n",
                     "      <td>0.00000</td>\n",
                     "      <td>0.233580</td>\n",
                     "      <td>1.4094</td>\n",
                     "      <td>1.3393</td>\n",
                     "      <td>0.58496</td>\n",
                     "      <td>...</td>\n",
                     "      <td>0.176480</td>\n",
                     "      <td>0.321880</td>\n",
                     "      <td>0.82635</td>\n",
                     "      <td>0.073039</td>\n",
                     "      <td>2.5912</td>\n",
                     "      <td>7.0756</td>\n",
                     "      <td>100.540</td>\n",
                     "      <td>3.6303</td>\n",
                     "      <td>4.6375</td>\n",
                     "      <td>b'0'</td>\n",
                     "    </tr>\n",
                     "  </tbody>\n",
                     "</table>\n",
                     "<p>5 rows × 65 columns</p>\n",
                     "</div>"
                  ],
                  "text/plain": [
                     "      Attr1    Attr2    Attr3   Attr4    Attr5    Attr6     Attr7   Attr8  \\\n",
                     "0  0.174190  0.41299  0.14371  1.3480 -28.9820  0.60383  0.219460  1.1225   \n",
                     "1  0.146240  0.46038  0.28230  1.6294   2.5952  0.00000  0.171850  1.1721   \n",
                     "2  0.000595  0.22612  0.48839  3.1599  84.8740  0.19114  0.004572  2.9881   \n",
                     "3  0.024526  0.43236  0.27546  1.7833 -10.1050  0.56944  0.024526  1.3057   \n",
                     "4  0.188290  0.41504  0.34231  1.9279 -58.2740  0.00000  0.233580  1.4094   \n",
                     "\n",
                     "    Attr9   Attr10  ...    Attr56    Attr57   Attr58    Attr59  Attr60  \\\n",
                     "0  1.1961  0.46359  ...  0.163960  0.375740  0.83604  0.000007  9.7145   \n",
                     "1  1.6018  0.53962  ...  0.027516  0.271000  0.90108  0.000000  5.9882   \n",
                     "2  1.0077  0.67566  ...  0.007639  0.000881  0.99236  0.000000  6.7742   \n",
                     "3  1.0509  0.56453  ...  0.048398  0.043445  0.95160  0.142980  4.2286   \n",
                     "4  1.3393  0.58496  ...  0.176480  0.321880  0.82635  0.073039  2.5912   \n",
                     "\n",
                     "   Attr61   Attr62  Attr63  Attr64  class  \n",
                     "0  6.2813   84.291  4.3303  4.0341   b'0'  \n",
                     "1  4.1103  102.190  3.5716  5.9500   b'0'  \n",
                     "2  3.7922   64.846  5.6287  4.4581   b'0'  \n",
                     "3  5.0528   98.783  3.6950  3.4844   b'0'  \n",
                     "4  7.0756  100.540  3.6303  4.6375   b'0'  \n",
                     "\n",
                     "[5 rows x 65 columns]"
                  ]
               },
               "execution_count": 5,
               "metadata": {},
               "output_type": "execute_result"
            }
         ],
         "source": [
            "X.head()"
         ]
      },
      {
         "cell_type": "code",
         "execution_count": 6,
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%%\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "outputs": [
            {
               "data": {
                  "text/plain": [
                     "Attr1     float64\n",
                     "Attr2     float64\n",
                     "Attr3     float64\n",
                     "Attr4     float64\n",
                     "Attr5     float64\n",
                     "           ...   \n",
                     "Attr61    float64\n",
                     "Attr62    float64\n",
                     "Attr63    float64\n",
                     "Attr64    float64\n",
                     "class      object\n",
                     "Length: 65, dtype: object"
                  ]
               },
               "execution_count": 6,
               "metadata": {},
               "output_type": "execute_result"
            }
         ],
         "source": [
            "X.dtypes"
         ]
      },
      {
         "cell_type": "code",
         "execution_count": 7,
         "metadata": {
            "pycharm": {
               "name": "#%%\n"
            }
         },
         "outputs": [
            {
               "data": {
                  "text/html": [
                     "<div>\n",
                     "<style scoped>\n",
                     "    .dataframe tbody tr th:only-of-type {\n",
                     "        vertical-align: middle;\n",
                     "    }\n",
                     "\n",
                     "    .dataframe tbody tr th {\n",
                     "        vertical-align: top;\n",
                     "    }\n",
                     "\n",
                     "    .dataframe thead th {\n",
                     "        text-align: right;\n",
                     "    }\n",
                     "</style>\n",
                     "<table border=\"1\" class=\"dataframe\">\n",
                     "  <thead>\n",
                     "    <tr style=\"text-align: right;\">\n",
                     "      <th></th>\n",
                     "      <th>Attr1</th>\n",
                     "      <th>Attr2</th>\n",
                     "      <th>Attr3</th>\n",
                     "      <th>Attr4</th>\n",
                     "      <th>Attr5</th>\n",
                     "      <th>Attr6</th>\n",
                     "      <th>Attr7</th>\n",
                     "      <th>Attr8</th>\n",
                     "      <th>Attr9</th>\n",
                     "      <th>Attr10</th>\n",
                     "      <th>...</th>\n",
                     "      <th>Attr55</th>\n",
                     "      <th>Attr56</th>\n",
                     "      <th>Attr57</th>\n",
                     "      <th>Attr58</th>\n",
                     "      <th>Attr59</th>\n",
                     "      <th>Attr60</th>\n",
                     "      <th>Attr61</th>\n",
                     "      <th>Attr62</th>\n",
                     "      <th>Attr63</th>\n",
                     "      <th>Attr64</th>\n",
                     "    </tr>\n",
                     "  </thead>\n",
                     "  <tbody>\n",
                     "    <tr>\n",
                     "      <th>count</th>\n",
                     "      <td>10503.000000</td>\n",
                     "      <td>10503.000000</td>\n",
                     "      <td>10503.000000</td>\n",
                     "      <td>10485.000000</td>\n",
                     "      <td>1.047800e+04</td>\n",
                     "      <td>10503.000000</td>\n",
                     "      <td>10503.000000</td>\n",
                     "      <td>10489.000000</td>\n",
                     "      <td>10500.000000</td>\n",
                     "      <td>10503.000000</td>\n",
                     "      <td>...</td>\n",
                     "      <td>1.050300e+04</td>\n",
                     "      <td>10460.000000</td>\n",
                     "      <td>10503.000000</td>\n",
                     "      <td>10474.000000</td>\n",
                     "      <td>10503.000000</td>\n",
                     "      <td>9.911000e+03</td>\n",
                     "      <td>10486.000000</td>\n",
                     "      <td>1.046000e+04</td>\n",
                     "      <td>10485.000000</td>\n",
                     "      <td>10275.000000</td>\n",
                     "    </tr>\n",
                     "    <tr>\n",
                     "      <th>mean</th>\n",
                     "      <td>0.052844</td>\n",
                     "      <td>0.619911</td>\n",
                     "      <td>0.095490</td>\n",
                     "      <td>9.980499</td>\n",
                     "      <td>-1.347662e+03</td>\n",
                     "      <td>-0.121159</td>\n",
                     "      <td>0.065624</td>\n",
                     "      <td>19.140113</td>\n",
                     "      <td>1.819254</td>\n",
                     "      <td>0.366093</td>\n",
                     "      <td>...</td>\n",
                     "      <td>6.638549e+03</td>\n",
                     "      <td>-0.530082</td>\n",
                     "      <td>-0.014817</td>\n",
                     "      <td>3.848794</td>\n",
                     "      <td>1.429319</td>\n",
                     "      <td>5.713363e+02</td>\n",
                     "      <td>13.935361</td>\n",
                     "      <td>1.355370e+02</td>\n",
                     "      <td>9.095149</td>\n",
                     "      <td>35.766800</td>\n",
                     "    </tr>\n",
                     "    <tr>\n",
                     "      <th>std</th>\n",
                     "      <td>0.647797</td>\n",
                     "      <td>6.427041</td>\n",
                     "      <td>6.420056</td>\n",
                     "      <td>523.691951</td>\n",
                     "      <td>1.185806e+05</td>\n",
                     "      <td>6.970625</td>\n",
                     "      <td>0.651152</td>\n",
                     "      <td>717.756745</td>\n",
                     "      <td>7.581659</td>\n",
                     "      <td>6.428603</td>\n",
                     "      <td>...</td>\n",
                     "      <td>5.989196e+04</td>\n",
                     "      <td>55.978608</td>\n",
                     "      <td>18.684047</td>\n",
                     "      <td>190.201224</td>\n",
                     "      <td>77.273270</td>\n",
                     "      <td>3.715967e+04</td>\n",
                     "      <td>83.704103</td>\n",
                     "      <td>2.599116e+04</td>\n",
                     "      <td>31.419096</td>\n",
                     "      <td>428.298315</td>\n",
                     "    </tr>\n",
                     "    <tr>\n",
                     "      <th>min</th>\n",
                     "      <td>-17.692000</td>\n",
                     "      <td>0.000000</td>\n",
                     "      <td>-479.730000</td>\n",
                     "      <td>0.002080</td>\n",
                     "      <td>-1.190300e+07</td>\n",
                     "      <td>-508.120000</td>\n",
                     "      <td>-17.692000</td>\n",
                     "      <td>-2.081800</td>\n",
                     "      <td>-1.215700</td>\n",
                     "      <td>-479.730000</td>\n",
                     "      <td>...</td>\n",
                     "      <td>-7.513800e+05</td>\n",
                     "      <td>-5691.700000</td>\n",
                     "      <td>-1667.300000</td>\n",
                     "      <td>-198.690000</td>\n",
                     "      <td>-172.070000</td>\n",
                     "      <td>0.000000e+00</td>\n",
                     "      <td>-6.590300</td>\n",
                     "      <td>-2.336500e+06</td>\n",
                     "      <td>-0.000156</td>\n",
                     "      <td>-0.000102</td>\n",
                     "    </tr>\n",
                     "    <tr>\n",
                     "      <th>25%</th>\n",
                     "      <td>0.000686</td>\n",
                     "      <td>0.253955</td>\n",
                     "      <td>0.017461</td>\n",
                     "      <td>1.040100</td>\n",
                     "      <td>-5.207075e+01</td>\n",
                     "      <td>0.000000</td>\n",
                     "      <td>0.002118</td>\n",
                     "      <td>0.431270</td>\n",
                     "      <td>1.011275</td>\n",
                     "      <td>0.297340</td>\n",
                     "      <td>...</td>\n",
                     "      <td>1.462100e+01</td>\n",
                     "      <td>0.005137</td>\n",
                     "      <td>0.006796</td>\n",
                     "      <td>0.875560</td>\n",
                     "      <td>0.000000</td>\n",
                     "      <td>5.533150e+00</td>\n",
                     "      <td>4.486075</td>\n",
                     "      <td>4.073700e+01</td>\n",
                     "      <td>3.062800</td>\n",
                     "      <td>2.023350</td>\n",
                     "    </tr>\n",
                     "    <tr>\n",
                     "      <th>50%</th>\n",
                     "      <td>0.043034</td>\n",
                     "      <td>0.464140</td>\n",
                     "      <td>0.198560</td>\n",
                     "      <td>1.605600</td>\n",
                     "      <td>1.579300e+00</td>\n",
                     "      <td>0.000000</td>\n",
                     "      <td>0.050945</td>\n",
                     "      <td>1.111000</td>\n",
                     "      <td>1.199000</td>\n",
                     "      <td>0.515500</td>\n",
                     "      <td>...</td>\n",
                     "      <td>8.822900e+02</td>\n",
                     "      <td>0.051765</td>\n",
                     "      <td>0.106880</td>\n",
                     "      <td>0.953060</td>\n",
                     "      <td>0.002976</td>\n",
                     "      <td>9.952100e+00</td>\n",
                     "      <td>6.677300</td>\n",
                     "      <td>7.066400e+01</td>\n",
                     "      <td>5.139200</td>\n",
                     "      <td>4.059300</td>\n",
                     "    </tr>\n",
                     "    <tr>\n",
                     "      <th>75%</th>\n",
                     "      <td>0.123805</td>\n",
                     "      <td>0.689330</td>\n",
                     "      <td>0.419545</td>\n",
                     "      <td>2.959500</td>\n",
                     "      <td>5.608400e+01</td>\n",
                     "      <td>0.072584</td>\n",
                     "      <td>0.142275</td>\n",
                     "      <td>2.857100</td>\n",
                     "      <td>2.059100</td>\n",
                     "      <td>0.725635</td>\n",
                     "      <td>...</td>\n",
                     "      <td>4.348900e+03</td>\n",
                     "      <td>0.130010</td>\n",
                     "      <td>0.271310</td>\n",
                     "      <td>0.995927</td>\n",
                     "      <td>0.240320</td>\n",
                     "      <td>2.093600e+01</td>\n",
                     "      <td>10.587500</td>\n",
                     "      <td>1.182200e+02</td>\n",
                     "      <td>8.882600</td>\n",
                     "      <td>9.682750</td>\n",
                     "    </tr>\n",
                     "    <tr>\n",
                     "      <th>max</th>\n",
                     "      <td>52.652000</td>\n",
                     "      <td>480.730000</td>\n",
                     "      <td>17.708000</td>\n",
                     "      <td>53433.000000</td>\n",
                     "      <td>6.854400e+05</td>\n",
                     "      <td>45.533000</td>\n",
                     "      <td>52.652000</td>\n",
                     "      <td>53432.000000</td>\n",
                     "      <td>740.440000</td>\n",
                     "      <td>11.837000</td>\n",
                     "      <td>...</td>\n",
                     "      <td>3.380500e+06</td>\n",
                     "      <td>293.150000</td>\n",
                     "      <td>552.640000</td>\n",
                     "      <td>18118.000000</td>\n",
                     "      <td>7617.300000</td>\n",
                     "      <td>3.660200e+06</td>\n",
                     "      <td>4470.400000</td>\n",
                     "      <td>1.073500e+06</td>\n",
                     "      <td>1974.500000</td>\n",
                     "      <td>21499.000000</td>\n",
                     "    </tr>\n",
                     "  </tbody>\n",
                     "</table>\n",
                     "<p>8 rows × 64 columns</p>\n",
                     "</div>"
                  ],
                  "text/plain": [
                     "              Attr1         Attr2         Attr3         Attr4         Attr5  \\\n",
                     "count  10503.000000  10503.000000  10503.000000  10485.000000  1.047800e+04   \n",
                     "mean       0.052844      0.619911      0.095490      9.980499 -1.347662e+03   \n",
                     "std        0.647797      6.427041      6.420056    523.691951  1.185806e+05   \n",
                     "min      -17.692000      0.000000   -479.730000      0.002080 -1.190300e+07   \n",
                     "25%        0.000686      0.253955      0.017461      1.040100 -5.207075e+01   \n",
                     "50%        0.043034      0.464140      0.198560      1.605600  1.579300e+00   \n",
                     "75%        0.123805      0.689330      0.419545      2.959500  5.608400e+01   \n",
                     "max       52.652000    480.730000     17.708000  53433.000000  6.854400e+05   \n",
                     "\n",
                     "              Attr6         Attr7         Attr8         Attr9        Attr10  \\\n",
                     "count  10503.000000  10503.000000  10489.000000  10500.000000  10503.000000   \n",
                     "mean      -0.121159      0.065624     19.140113      1.819254      0.366093   \n",
                     "std        6.970625      0.651152    717.756745      7.581659      6.428603   \n",
                     "min     -508.120000    -17.692000     -2.081800     -1.215700   -479.730000   \n",
                     "25%        0.000000      0.002118      0.431270      1.011275      0.297340   \n",
                     "50%        0.000000      0.050945      1.111000      1.199000      0.515500   \n",
                     "75%        0.072584      0.142275      2.857100      2.059100      0.725635   \n",
                     "max       45.533000     52.652000  53432.000000    740.440000     11.837000   \n",
                     "\n",
                     "       ...        Attr55        Attr56        Attr57        Attr58  \\\n",
                     "count  ...  1.050300e+04  10460.000000  10503.000000  10474.000000   \n",
                     "mean   ...  6.638549e+03     -0.530082     -0.014817      3.848794   \n",
                     "std    ...  5.989196e+04     55.978608     18.684047    190.201224   \n",
                     "min    ... -7.513800e+05  -5691.700000  -1667.300000   -198.690000   \n",
                     "25%    ...  1.462100e+01      0.005137      0.006796      0.875560   \n",
                     "50%    ...  8.822900e+02      0.051765      0.106880      0.953060   \n",
                     "75%    ...  4.348900e+03      0.130010      0.271310      0.995927   \n",
                     "max    ...  3.380500e+06    293.150000    552.640000  18118.000000   \n",
                     "\n",
                     "             Attr59        Attr60        Attr61        Attr62        Attr63  \\\n",
                     "count  10503.000000  9.911000e+03  10486.000000  1.046000e+04  10485.000000   \n",
                     "mean       1.429319  5.713363e+02     13.935361  1.355370e+02      9.095149   \n",
                     "std       77.273270  3.715967e+04     83.704103  2.599116e+04     31.419096   \n",
                     "min     -172.070000  0.000000e+00     -6.590300 -2.336500e+06     -0.000156   \n",
                     "25%        0.000000  5.533150e+00      4.486075  4.073700e+01      3.062800   \n",
                     "50%        0.002976  9.952100e+00      6.677300  7.066400e+01      5.139200   \n",
                     "75%        0.240320  2.093600e+01     10.587500  1.182200e+02      8.882600   \n",
                     "max     7617.300000  3.660200e+06   4470.400000  1.073500e+06   1974.500000   \n",
                     "\n",
                     "             Attr64  \n",
                     "count  10275.000000  \n",
                     "mean      35.766800  \n",
                     "std      428.298315  \n",
                     "min       -0.000102  \n",
                     "25%        2.023350  \n",
                     "50%        4.059300  \n",
                     "75%        9.682750  \n",
                     "max    21499.000000  \n",
                     "\n",
                     "[8 rows x 64 columns]"
                  ]
               },
               "execution_count": 7,
               "metadata": {},
               "output_type": "execute_result"
            }
         ],
         "source": [
            "X.describe()"
         ]
      },
      {
         "cell_type": "code",
         "execution_count": 8,
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "outputs": [
            {
               "data": {
                  "text/plain": [
                     "['net profit / total assets',\n",
                     " 'total liabilities / total assets',\n",
                     " 'working capital / total assets',\n",
                     " 'current assets / short-term liabilities',\n",
                     " '[(cash + short-term securities + receivables - short-term liabilities) / (operating expenses - depreciation)] * 365',\n",
                     " 'retained earnings / total assets',\n",
                     " 'EBIT / total assets',\n",
                     " 'book value of equity / total liabilities',\n",
                     " 'sales / total assets',\n",
                     " 'equity / total assets',\n",
                     " '(gross profit + extraordinary items + financial expenses) / total assets',\n",
                     " 'gross profit / short-term liabilities',\n",
                     " '(gross profit + depreciation) / sales',\n",
                     " '(gross profit + interest) / total assets',\n",
                     " '(total liabilities * 365) / (gross profit + depreciation)',\n",
                     " '(gross profit + depreciation) / total liabilities',\n",
                     " 'total assets / total liabilities',\n",
                     " 'gross profit / total assets',\n",
                     " 'gross profit / sales',\n",
                     " '(inventory * 365) / sales',\n",
                     " 'sales (n) / sales (n-1)',\n",
                     " 'profit on operating activities / total assets',\n",
                     " 'net profit / sales',\n",
                     " 'gross profit (in 3 years) / total assets',\n",
                     " '(equity - share capital) / total assets',\n",
                     " '(net profit + depreciation) / total liabilities',\n",
                     " 'profit on operating activities / financial expenses',\n",
                     " 'working capital / fixed assets',\n",
                     " 'logarithm of total assets',\n",
                     " '(total liabilities - cash) / sales',\n",
                     " '(gross profit + interest) / sales',\n",
                     " '(current liabilities * 365) / cost of products sold',\n",
                     " 'operating expenses / short-term liabilities',\n",
                     " 'operating expenses / total liabilities',\n",
                     " 'profit on sales / total assets',\n",
                     " 'total sales / total assets',\n",
                     " 'constant capital / total assets',\n",
                     " 'profit on sales / sales',\n",
                     " '(current assets - inventory - receivables) / short-term liabilities',\n",
                     " 'total liabilities / ((profit on operating activities + depreciation) * (12/365))',\n",
                     " 'profit on operating activities / sales',\n",
                     " 'rotation receivables + inventory turnover in days',\n",
                     " '(receivables * 365) / sales',\n",
                     " 'net profit / inventory',\n",
                     " '(current assets - inventory) / short-term liabilities',\n",
                     " '(inventory * 365) / cost of products sold',\n",
                     " 'EBITDA (profit on operating activities - depreciation) / total assets',\n",
                     " 'EBITDA (profit on operating activities - depreciation) / sales',\n",
                     " 'current assets / total liabilities',\n",
                     " 'short-term liabilities / total assets',\n",
                     " '(short-term liabilities * 365) / cost of products sold)',\n",
                     " 'equity / fixed assets',\n",
                     " 'constant capital / fixed assets',\n",
                     " 'working capital',\n",
                     " '(sales - cost of products sold) / sales',\n",
                     " '(current assets - inventory - short-term liabilities) / (sales - gross profit - depreciation)',\n",
                     " 'total costs / total sales',\n",
                     " 'long-term liabilities / equity',\n",
                     " 'sales / inventory',\n",
                     " 'sales / receivables',\n",
                     " '(short-term liabilities * 365) / sales',\n",
                     " 'sales / short-term liabilities',\n",
                     " 'sales / fixed assets']"
                  ]
               },
               "execution_count": 8,
               "metadata": {},
               "output_type": "execute_result"
            }
         ],
         "source": [
            "feature_names"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "source": [
            "DataFrame zawiera 64 atrybuty numeryczne o zróżnicowanych rozkładach wartości oraz kolumnę `\"class\"` typu `bytes` z klasami 0 i 1. Wiemy, że mamy do czynienia z klasyfikacją binarną - klasa 0 to brak bankructwa, klasa 1 to bankructwo w ciągu najbliższych 3 lat. Przyjrzyjmy się dokładniej naszym danym."
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "### Zadanie 1 (0.5 punktu)"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "1. Wyodrębnij klasy jako osobną zmienną typu `pd.Series`, usuwając je z macierzy `X`. Przekonwertuj go na liczby całkowite.\n",
            "2. Narysuj wykres słupkowy (bar plot) częstotliwości obu klas w całym zbiorze. Upewnij się, że na osi X są numery lub nazwy klas, a oś Y ma wartości w procentach.\n",
            "\n",
            "**Uwaga:** sugerowane jest użycie `if` w podpunkcie 1, żeby można było tę komórkę bezpiecznie odpalić kilka razy."
         ]
      },
      {
         "cell_type": "code",
         "execution_count": 9,
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%%\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "outputs": [
            {
               "data": {
                  "image/png": "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",
                  "text/plain": [
                     "<Figure size 800x500 with 1 Axes>"
                  ]
               },
               "metadata": {},
               "output_type": "display_data"
            }
         ],
         "source": [
            "import matplotlib.pyplot as plt\n",
            "\n",
            "if (\"class\" in X.columns):\n",
            "    y = X.pop(\"class\")  \n",
            "\n",
            "y = y.astype(int)\n",
            "\n",
            "class_counts = y.value_counts(normalize=True) * 100\n",
            "\n",
            "# Plotting\n",
            "plt.figure(figsize=(8, 5))\n",
            "class_counts.sort_index().plot(kind='bar')\n",
            "plt.title('Bankruptcy Class Distribution')\n",
            "plt.xlabel('0: Not Bankrupt, 1: Bankrupt')\n",
            "plt.ylabel('Frequency (%)')\n",
            "plt.xticks(rotation=0)\n",
            "plt.show()\n",
            "\n"
         ]
      },
      {
         "cell_type": "code",
         "execution_count": 10,
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "outputs": [
            {
               "name": "stdout",
               "output_type": "stream",
               "text": [
                  "Solution is correct!\n"
               ]
            }
         ],
         "source": [
            "assert \"class\" not in X.columns\n",
            "\n",
            "print(\"Solution is correct!\")"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "source": [
            "Jak widać, klasa pozytywna jest w znacznej mniejszości, stanowi poniżej 5% zbioru. Taki problem nazywamy **klasyfikacją niezbalansowaną (imbalanced classification)**. Mamy tu **klasę dominującą (majority class)** oraz **klasę mniejszościową (minority class)**. Pechowo prawie zawsze interesuje nas ta druga, bo klasa większościowa jest trywialna. Przykładowo, 99% badanych jest zdrowych, a 1% ma niewykryty nowotwór - z oczywistych przyczyn chcemy wykrywać właśnie sytuację rzadką (problem diagnozy jako klasyfikacji jest zasadniczo zawsze niezbalansowany). W dalszej części laboratorium poznamy szereg konsekwencji tego zjawiska i metody na radzenie sobie z nim.\n",
            "\n",
            "Mamy sporo cech, wszystkie numeryczne. Ciekawe, czy mają wartości brakujące, a jeśli tak, to ile. Policzymy to z pomocą biblioteki Pandas i metody `.isna()`. Domyślnie operuje ona na kolumnach, jak większość metod w Pandasie. Sumę wartości per kolumna zwróci nam metoda `.sum()`. Jeżeli podzielimy to przez liczbę wierszy `len(X)`, to otrzymamy ułamek wartości brakujących w każdej kolumnie.\n",
            "\n",
            "Pandas potrafi też stworzyć wykres, z pomocą funkcji np. `.plot.hist()` czy `.plot.bar()`. Przyjmują one opcje formatowania wykresu, z których korzysta pod spodem biblioteka matplotlib."
         ]
      },
      {
         "cell_type": "code",
         "execution_count": 11,
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%%\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "outputs": [
            {
               "data": {
                  "text/plain": [
                     "<Axes: title={'center': 'Fraction of missing values per column'}>"
                  ]
               },
               "execution_count": 11,
               "metadata": {},
               "output_type": "execute_result"
            },
            {
               "data": {
                  "image/png": "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",
                  "text/plain": [
                     "<Figure size 1500x500 with 1 Axes>"
                  ]
               },
               "metadata": {},
               "output_type": "display_data"
            }
         ],
         "source": [
            "na_perc = X.isna().sum() / len(X)\n",
            "na_perc.plot.bar(title=\"Fraction of missing values per column\", figsize=(15, 5))"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "Jak widać, cecha 37 ma bardzo dużo wartości brakujących, podczas gdy pozostałe cechy mają raczej niewielką ich liczbę. W takiej sytuacji najlepiej usunąć tę cechę, a pozostałe wartości brakujące **uzupełnić / imputować (impute)**. Typowo wykorzystuje się do tego wartość średnią lub medianę z danej kolumny. Ale uwaga - imputacji dokonuje się dopiero po podziale na zbiór treningowy i testowy! W przeciwnym wypadku wykorzystywalibyśmy dane ze zbioru testowego, co sztucznie zawyżyłoby wyniki. Jest to błąd metodologiczny - **wyciek danych (data leakage)**.\n",
            "\n",
            "Podział na zbiór treningowy i testowy to pierwszy moment, kiedy niezbalansowanie danych nam przeszkadza. Jeżeli zrobimy to czysto losowo, to są spore szanse, że w zbiorze testowym będzie tylko klasa negatywna - w końcu jest jej aż >95%. Dlatego wykorzystuje się **próbkowanie ze stratyfikacją (stratified sampling)**, dzięki któremu proporcje klas w zbiorze przed podziałem oraz obu zbiorach po podziale są takie same."
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "### Zadanie 2 (0.75 punktu)"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "1. Usuń kolumnę `\"Attr37\"` ze zbioru danych.\n",
            "2. Dokonaj podziału zbioru na treningowy i testowy w proporcjach 80%-20%, z przemieszaniem (`shuffle`), ze stratyfikacją, wykorzystując funkcję `train_test_split` ze Scikit-learn'a.\n",
            "3. Uzupełnij wartości brakujące średnią wartością cechy z pomocą klasy `SimpleImputer`.\n",
            "\n",
            "**Uwaga:**\n",
            "- jak wcześniej, sugerowane jest użycie `if` w podpunkcie 1\n",
            "- pamiętaj o uwzględnieniu stałego `random_state=0`, aby wyniki były **reprodukowalne (reproducible)**\n",
            "- `stratify` oczekuje wektora klas\n",
            "- wartości do imputacji trzeba wyestymować na zbiorze treningowym (`.fit()`), a potem zastosować te nauczone wartości na obu podzbiorach (treningowym i testowym)"
         ]
      },
      {
         "cell_type": "code",
         "execution_count": 12,
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%%\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "outputs": [],
         "source": [
            "from sklearn.model_selection import train_test_split\n",
            "from sklearn.compose import ColumnTransformer\n",
            "from sklearn.impute import SimpleImputer\n",
            "\n",
            "if (\"Attr37\" in X.columns):\n",
            "    X.pop(\"Attr37\")\n",
            "\n",
            "X_train, X_test, y_train, y_test = train_test_split(\n",
            "    X, y, test_size=0.20, random_state=0, \n",
            "    stratify=y, shuffle=True\n",
            ")\n",
            "\n",
            "column_transformer = ColumnTransformer([\n",
            "    (\"imputer\", SimpleImputer(strategy=\"mean\"), X.columns)\n",
            "])\n",
            "\n",
            "column_transformer.fit(X_train)\n",
            "\n",
            "X_train = column_transformer.transform(X_train)\n",
            "X_test = column_transformer.transform(X_test)\n"
         ]
      },
      {
         "cell_type": "code",
         "execution_count": 13,
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "outputs": [
            {
               "name": "stdout",
               "output_type": "stream",
               "text": [
                  "Solution is correct!\n"
               ]
            }
         ],
         "source": [
            "import numpy as np\n",
            "\n",
            "assert \"Attr37\" not in X.columns\n",
            "assert not np.any(np.isnan(X_train))\n",
            "assert not np.any(np.isnan(X_test))\n",
            "\n",
            "print(\"Solution is correct!\")"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "source": [
            "## Prosta klasyfikacja"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "source": [
            "Zanim przejdzie się do modeli bardziej złożonych, trzeba najpierw wypróbować coś prostego, żeby mieć punkt odniesienia. Tworzy się dlatego **modele bazowe (baselines)**.\n",
            "\n",
            "W naszym przypadku będzie to **drzewo decyzyjne (decision tree)**. Jest to drzewo binarne z decyzjami if-else, prowadzącymi do klasyfikacji danego przykładu w liściu. Każdy podział w drzewie to pytanie postaci \"Czy wartość cechy X jest większa lub równa Y?\". Trening takiego drzewa to prosty algorytm zachłanny, bardzo przypomina budowę zwykłego drzewa binarnego. W każdym węźle wykonujemy:\n",
            "1. Sprawdź po kolei wszystkie możliwe punkty podziału, czyli każdą (unikalną) wartość każdej cechy, po kolei.\n",
            "2. Dla każdego przypadku podziel zbiór na 2 kawałki: niespełniający warunku (lewe dziecko) i spełniający warunek (prawe dziecko).\n",
            "3. Oblicz jakość podziału według pewnej wybranej funkcji jakości. Im lepiej nasz if/else rozdziela klasy od siebie (im \"czystsze\" są węzły-dzieci), tym wyższa jakość. Innymi słowy, chcemy, żeby do jednego dziecka poszła jedna klasa, a do drugiego druga.\n",
            "4. Wybierz podział o najwyższej jakości.\n",
            "\n",
            "Taki algorytm wykonuje się rekurencyjnie, aż otrzymamy węzeł czysty (pure leaf), czyli taki, w którym są przykłady z tylko jednej klasy. Typowo wykorzystywaną funkcją jakości (kryterium podziału) jest entropia Shannona - im niższa entropia, tym bardziej jednolite są klasy w węźle (czyli wybieramy podział o najniższej entropii).\n",
            "\n",
            "Powyższe wytłumaczenie algorytmu jest oczywiście nieformalne i dość skrótowe. Doskonałe tłumaczenie, z interaktywnymi wizualizacjami, dostępne jest [tutaj](https://mlu-explain.github.io/decision-tree/). W formie filmów - [tutaj](https://www.youtube.com/watch?v=ZVR2Way4nwQ) oraz [tutaj](https://www.youtube.com/watch?v=_L39rN6gz7Y). Dla drzew do regresji - [ten film](https://www.youtube.com/watch?v=g9c66TUylZ4).\n",
            "\n",
            "<img src = https://miro.medium.com/max/1838/1*WyTsLwcAXivFCgNtF0OPqA.png width = \"642\" height = \"451\" >\n",
            "\n",
            "Warto zauważyć, że taka konstrukcja prowadzi zawsze do overfittingu. Otrzymanie liści czystych oznacza, że mamy 100% dokładności na zbiorze treningowym, czyli perfekcyjnie przeuczony klasyfikator. W związku z tym nasze predykcje mają bardzo niski bias, ale bardzo dużą wariancję. Pomimo tego drzewa potrafią dać bardzo przyzwoite wyniki, a w celu ich poprawy można je regularyzować, aby mieć mniej \"rozrośnięte\" drzewo. [Film dla zainteresowanych](https://www.youtube.com/watch?v=D0efHEJsfHo).\n",
            "\n",
            "W tym wypadku AI to naprawdę tylko zbiór if'ów ;)"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "source": [
            "Mając wytrenowany klasyfikator, trzeba oczywiście sprawdzić, jak dobrze on sobie radzi. Tu natrafiamy na kolejny problem z klasyfikacją niezbalansowaną - zwykła celność (accuracy) na pewno nie zadziała! Typowo wykorzystuje się AUC, nazywane też AUROC (Area Under Receiver Operating Characteristic), bo metryka ta \"widzi\" i uwzględnia niezbalansowanie klas. Wymaga ona przekazania prawdopodobieństwa klasy pozytywnej, a nie tylko binarnej decyzji.\n",
            "\n",
            "Bardzo dobre i bardziej szczegółowe wytłumaczenie, z interktywnymi wizualizacjami, można znaleć [tutaj](https://mlu-explain.github.io/roc-auc/). Dla preferujących filmy - [tutaj](https://www.youtube.com/watch?v=4jRBRDbJemM).\n",
            "\n",
            "Co ważne, z definicji AUROC, trzeba tam użyć prawdopodobieństw klasy pozytywnej (klasy 1). W Scikit-learn'ie zwraca je metoda `.predict_proba()`, która w kolejnych kolumnach zwraca prawdopodobieństwa poszczególnych klas."
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "### Zadanie 3 (0.75 punktu)"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "1. Wytrenuj klasyfikator drzewa decyzyjnego (klasa `DecisionTreeClassifier`). Użyj entropii jako kryterium podziału.\n",
            "2. Oblicz i wypisz AUROC na zbiorze testowym dla drzewa decyzyjnego (funkcja `roc_auc_score`).\n",
            "3. Skomentuj wynik - czy twoim zdaniem osiągnięty AUROC to dużo czy mało, biorąc pod uwagę możliwy zakres wartości tej metryki?\n",
            "\n",
            "**Uwaga:**\n",
            "- pamiętaj o użyciu stałego `random_state=0`\n",
            "- jeżeli drzewo nie wyświetli się samo, użyj `plt.show()` z Matplotliba\n",
            "- pamiętaj o tym, żeby przekazać do metryki AUROC prawdopodobieństwa klasy pozytywnej, a nie binarne predykcje!"
         ]
      },
      {
         "cell_type": "code",
         "execution_count": 14,
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%%\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "outputs": [
            {
               "name": "stdout",
               "output_type": "stream",
               "text": [
                  "AUROC: 0.7264\n"
               ]
            },
            {
               "data": {
                  "text/html": [
                     "<style>#sk-container-id-1 {\n",
                     "  /* Definition of color scheme common for light and dark mode */\n",
                     "  --sklearn-color-text: black;\n",
                     "  --sklearn-color-line: gray;\n",
                     "  /* Definition of color scheme for unfitted estimators */\n",
                     "  --sklearn-color-unfitted-level-0: #fff5e6;\n",
                     "  --sklearn-color-unfitted-level-1: #f6e4d2;\n",
                     "  --sklearn-color-unfitted-level-2: #ffe0b3;\n",
                     "  --sklearn-color-unfitted-level-3: chocolate;\n",
                     "  /* Definition of color scheme for fitted estimators */\n",
                     "  --sklearn-color-fitted-level-0: #f0f8ff;\n",
                     "  --sklearn-color-fitted-level-1: #d4ebff;\n",
                     "  --sklearn-color-fitted-level-2: #b3dbfd;\n",
                     "  --sklearn-color-fitted-level-3: cornflowerblue;\n",
                     "\n",
                     "  /* Specific color for light theme */\n",
                     "  --sklearn-color-text-on-default-background: var(--sg-text-color, var(--theme-code-foreground, var(--jp-content-font-color1, black)));\n",
                     "  --sklearn-color-background: var(--sg-background-color, var(--theme-background, var(--jp-layout-color0, white)));\n",
                     "  --sklearn-color-border-box: var(--sg-text-color, var(--theme-code-foreground, var(--jp-content-font-color1, black)));\n",
                     "  --sklearn-color-icon: #696969;\n",
                     "\n",
                     "  @media (prefers-color-scheme: dark) {\n",
                     "    /* Redefinition of color scheme for dark theme */\n",
                     "    --sklearn-color-text-on-default-background: var(--sg-text-color, var(--theme-code-foreground, var(--jp-content-font-color1, white)));\n",
                     "    --sklearn-color-background: var(--sg-background-color, var(--theme-background, var(--jp-layout-color0, #111)));\n",
                     "    --sklearn-color-border-box: var(--sg-text-color, var(--theme-code-foreground, var(--jp-content-font-color1, white)));\n",
                     "    --sklearn-color-icon: #878787;\n",
                     "  }\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-1 {\n",
                     "  color: var(--sklearn-color-text);\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-1 pre {\n",
                     "  padding: 0;\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-1 input.sk-hidden--visually {\n",
                     "  border: 0;\n",
                     "  clip: rect(1px 1px 1px 1px);\n",
                     "  clip: rect(1px, 1px, 1px, 1px);\n",
                     "  height: 1px;\n",
                     "  margin: -1px;\n",
                     "  overflow: hidden;\n",
                     "  padding: 0;\n",
                     "  position: absolute;\n",
                     "  width: 1px;\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-1 div.sk-dashed-wrapped {\n",
                     "  border: 1px dashed var(--sklearn-color-line);\n",
                     "  margin: 0 0.4em 0.5em 0.4em;\n",
                     "  box-sizing: border-box;\n",
                     "  padding-bottom: 0.4em;\n",
                     "  background-color: var(--sklearn-color-background);\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-1 div.sk-container {\n",
                     "  /* jupyter's `normalize.less` sets `[hidden] { display: none; }`\n",
                     "     but bootstrap.min.css set `[hidden] { display: none !important; }`\n",
                     "     so we also need the `!important` here to be able to override the\n",
                     "     default hidden behavior on the sphinx rendered scikit-learn.org.\n",
                     "     See: https://github.com/scikit-learn/scikit-learn/issues/21755 */\n",
                     "  display: inline-block !important;\n",
                     "  position: relative;\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-1 div.sk-text-repr-fallback {\n",
                     "  display: none;\n",
                     "}\n",
                     "\n",
                     "div.sk-parallel-item,\n",
                     "div.sk-serial,\n",
                     "div.sk-item {\n",
                     "  /* draw centered vertical line to link estimators */\n",
                     "  background-image: linear-gradient(var(--sklearn-color-text-on-default-background), var(--sklearn-color-text-on-default-background));\n",
                     "  background-size: 2px 100%;\n",
                     "  background-repeat: no-repeat;\n",
                     "  background-position: center center;\n",
                     "}\n",
                     "\n",
                     "/* Parallel-specific style estimator block */\n",
                     "\n",
                     "#sk-container-id-1 div.sk-parallel-item::after {\n",
                     "  content: \"\";\n",
                     "  width: 100%;\n",
                     "  border-bottom: 2px solid var(--sklearn-color-text-on-default-background);\n",
                     "  flex-grow: 1;\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-1 div.sk-parallel {\n",
                     "  display: flex;\n",
                     "  align-items: stretch;\n",
                     "  justify-content: center;\n",
                     "  background-color: var(--sklearn-color-background);\n",
                     "  position: relative;\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-1 div.sk-parallel-item {\n",
                     "  display: flex;\n",
                     "  flex-direction: column;\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-1 div.sk-parallel-item:first-child::after {\n",
                     "  align-self: flex-end;\n",
                     "  width: 50%;\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-1 div.sk-parallel-item:last-child::after {\n",
                     "  align-self: flex-start;\n",
                     "  width: 50%;\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-1 div.sk-parallel-item:only-child::after {\n",
                     "  width: 0;\n",
                     "}\n",
                     "\n",
                     "/* Serial-specific style estimator block */\n",
                     "\n",
                     "#sk-container-id-1 div.sk-serial {\n",
                     "  display: flex;\n",
                     "  flex-direction: column;\n",
                     "  align-items: center;\n",
                     "  background-color: var(--sklearn-color-background);\n",
                     "  padding-right: 1em;\n",
                     "  padding-left: 1em;\n",
                     "}\n",
                     "\n",
                     "\n",
                     "/* Toggleable style: style used for estimator/Pipeline/ColumnTransformer box that is\n",
                     "clickable and can be expanded/collapsed.\n",
                     "- Pipeline and ColumnTransformer use this feature and define the default style\n",
                     "- Estimators will overwrite some part of the style using the `sk-estimator` class\n",
                     "*/\n",
                     "\n",
                     "/* Pipeline and ColumnTransformer style (default) */\n",
                     "\n",
                     "#sk-container-id-1 div.sk-toggleable {\n",
                     "  /* Default theme specific background. It is overwritten whether we have a\n",
                     "  specific estimator or a Pipeline/ColumnTransformer */\n",
                     "  background-color: var(--sklearn-color-background);\n",
                     "}\n",
                     "\n",
                     "/* Toggleable label */\n",
                     "#sk-container-id-1 label.sk-toggleable__label {\n",
                     "  cursor: pointer;\n",
                     "  display: block;\n",
                     "  width: 100%;\n",
                     "  margin-bottom: 0;\n",
                     "  padding: 0.5em;\n",
                     "  box-sizing: border-box;\n",
                     "  text-align: center;\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-1 label.sk-toggleable__label-arrow:before {\n",
                     "  /* Arrow on the left of the label */\n",
                     "  content: \"▸\";\n",
                     "  float: left;\n",
                     "  margin-right: 0.25em;\n",
                     "  color: var(--sklearn-color-icon);\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-1 label.sk-toggleable__label-arrow:hover:before {\n",
                     "  color: var(--sklearn-color-text);\n",
                     "}\n",
                     "\n",
                     "/* Toggleable content - dropdown */\n",
                     "\n",
                     "#sk-container-id-1 div.sk-toggleable__content {\n",
                     "  max-height: 0;\n",
                     "  max-width: 0;\n",
                     "  overflow: hidden;\n",
                     "  text-align: left;\n",
                     "  /* unfitted */\n",
                     "  background-color: var(--sklearn-color-unfitted-level-0);\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-1 div.sk-toggleable__content.fitted {\n",
                     "  /* fitted */\n",
                     "  background-color: var(--sklearn-color-fitted-level-0);\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-1 div.sk-toggleable__content pre {\n",
                     "  margin: 0.2em;\n",
                     "  border-radius: 0.25em;\n",
                     "  color: var(--sklearn-color-text);\n",
                     "  /* unfitted */\n",
                     "  background-color: var(--sklearn-color-unfitted-level-0);\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-1 div.sk-toggleable__content.fitted pre {\n",
                     "  /* unfitted */\n",
                     "  background-color: var(--sklearn-color-fitted-level-0);\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-1 input.sk-toggleable__control:checked~div.sk-toggleable__content {\n",
                     "  /* Expand drop-down */\n",
                     "  max-height: 200px;\n",
                     "  max-width: 100%;\n",
                     "  overflow: auto;\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-1 input.sk-toggleable__control:checked~label.sk-toggleable__label-arrow:before {\n",
                     "  content: \"▾\";\n",
                     "}\n",
                     "\n",
                     "/* Pipeline/ColumnTransformer-specific style */\n",
                     "\n",
                     "#sk-container-id-1 div.sk-label input.sk-toggleable__control:checked~label.sk-toggleable__label {\n",
                     "  color: var(--sklearn-color-text);\n",
                     "  background-color: var(--sklearn-color-unfitted-level-2);\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-1 div.sk-label.fitted input.sk-toggleable__control:checked~label.sk-toggleable__label {\n",
                     "  background-color: var(--sklearn-color-fitted-level-2);\n",
                     "}\n",
                     "\n",
                     "/* Estimator-specific style */\n",
                     "\n",
                     "/* Colorize estimator box */\n",
                     "#sk-container-id-1 div.sk-estimator input.sk-toggleable__control:checked~label.sk-toggleable__label {\n",
                     "  /* unfitted */\n",
                     "  background-color: var(--sklearn-color-unfitted-level-2);\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-1 div.sk-estimator.fitted input.sk-toggleable__control:checked~label.sk-toggleable__label {\n",
                     "  /* fitted */\n",
                     "  background-color: var(--sklearn-color-fitted-level-2);\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-1 div.sk-label label.sk-toggleable__label,\n",
                     "#sk-container-id-1 div.sk-label label {\n",
                     "  /* The background is the default theme color */\n",
                     "  color: var(--sklearn-color-text-on-default-background);\n",
                     "}\n",
                     "\n",
                     "/* On hover, darken the color of the background */\n",
                     "#sk-container-id-1 div.sk-label:hover label.sk-toggleable__label {\n",
                     "  color: var(--sklearn-color-text);\n",
                     "  background-color: var(--sklearn-color-unfitted-level-2);\n",
                     "}\n",
                     "\n",
                     "/* Label box, darken color on hover, fitted */\n",
                     "#sk-container-id-1 div.sk-label.fitted:hover label.sk-toggleable__label.fitted {\n",
                     "  color: var(--sklearn-color-text);\n",
                     "  background-color: var(--sklearn-color-fitted-level-2);\n",
                     "}\n",
                     "\n",
                     "/* Estimator label */\n",
                     "\n",
                     "#sk-container-id-1 div.sk-label label {\n",
                     "  font-family: monospace;\n",
                     "  font-weight: bold;\n",
                     "  display: inline-block;\n",
                     "  line-height: 1.2em;\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-1 div.sk-label-container {\n",
                     "  text-align: center;\n",
                     "}\n",
                     "\n",
                     "/* Estimator-specific */\n",
                     "#sk-container-id-1 div.sk-estimator {\n",
                     "  font-family: monospace;\n",
                     "  border: 1px dotted var(--sklearn-color-border-box);\n",
                     "  border-radius: 0.25em;\n",
                     "  box-sizing: border-box;\n",
                     "  margin-bottom: 0.5em;\n",
                     "  /* unfitted */\n",
                     "  background-color: var(--sklearn-color-unfitted-level-0);\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-1 div.sk-estimator.fitted {\n",
                     "  /* fitted */\n",
                     "  background-color: var(--sklearn-color-fitted-level-0);\n",
                     "}\n",
                     "\n",
                     "/* on hover */\n",
                     "#sk-container-id-1 div.sk-estimator:hover {\n",
                     "  /* unfitted */\n",
                     "  background-color: var(--sklearn-color-unfitted-level-2);\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-1 div.sk-estimator.fitted:hover {\n",
                     "  /* fitted */\n",
                     "  background-color: var(--sklearn-color-fitted-level-2);\n",
                     "}\n",
                     "\n",
                     "/* Specification for estimator info (e.g. \"i\" and \"?\") */\n",
                     "\n",
                     "/* Common style for \"i\" and \"?\" */\n",
                     "\n",
                     ".sk-estimator-doc-link,\n",
                     "a:link.sk-estimator-doc-link,\n",
                     "a:visited.sk-estimator-doc-link {\n",
                     "  float: right;\n",
                     "  font-size: smaller;\n",
                     "  line-height: 1em;\n",
                     "  font-family: monospace;\n",
                     "  background-color: var(--sklearn-color-background);\n",
                     "  border-radius: 1em;\n",
                     "  height: 1em;\n",
                     "  width: 1em;\n",
                     "  text-decoration: none !important;\n",
                     "  margin-left: 1ex;\n",
                     "  /* unfitted */\n",
                     "  border: var(--sklearn-color-unfitted-level-1) 1pt solid;\n",
                     "  color: var(--sklearn-color-unfitted-level-1);\n",
                     "}\n",
                     "\n",
                     ".sk-estimator-doc-link.fitted,\n",
                     "a:link.sk-estimator-doc-link.fitted,\n",
                     "a:visited.sk-estimator-doc-link.fitted {\n",
                     "  /* fitted */\n",
                     "  border: var(--sklearn-color-fitted-level-1) 1pt solid;\n",
                     "  color: var(--sklearn-color-fitted-level-1);\n",
                     "}\n",
                     "\n",
                     "/* On hover */\n",
                     "div.sk-estimator:hover .sk-estimator-doc-link:hover,\n",
                     ".sk-estimator-doc-link:hover,\n",
                     "div.sk-label-container:hover .sk-estimator-doc-link:hover,\n",
                     ".sk-estimator-doc-link:hover {\n",
                     "  /* unfitted */\n",
                     "  background-color: var(--sklearn-color-unfitted-level-3);\n",
                     "  color: var(--sklearn-color-background);\n",
                     "  text-decoration: none;\n",
                     "}\n",
                     "\n",
                     "div.sk-estimator.fitted:hover .sk-estimator-doc-link.fitted:hover,\n",
                     ".sk-estimator-doc-link.fitted:hover,\n",
                     "div.sk-label-container:hover .sk-estimator-doc-link.fitted:hover,\n",
                     ".sk-estimator-doc-link.fitted:hover {\n",
                     "  /* fitted */\n",
                     "  background-color: var(--sklearn-color-fitted-level-3);\n",
                     "  color: var(--sklearn-color-background);\n",
                     "  text-decoration: none;\n",
                     "}\n",
                     "\n",
                     "/* Span, style for the box shown on hovering the info icon */\n",
                     ".sk-estimator-doc-link span {\n",
                     "  display: none;\n",
                     "  z-index: 9999;\n",
                     "  position: relative;\n",
                     "  font-weight: normal;\n",
                     "  right: .2ex;\n",
                     "  padding: .5ex;\n",
                     "  margin: .5ex;\n",
                     "  width: min-content;\n",
                     "  min-width: 20ex;\n",
                     "  max-width: 50ex;\n",
                     "  color: var(--sklearn-color-text);\n",
                     "  box-shadow: 2pt 2pt 4pt #999;\n",
                     "  /* unfitted */\n",
                     "  background: var(--sklearn-color-unfitted-level-0);\n",
                     "  border: .5pt solid var(--sklearn-color-unfitted-level-3);\n",
                     "}\n",
                     "\n",
                     ".sk-estimator-doc-link.fitted span {\n",
                     "  /* fitted */\n",
                     "  background: var(--sklearn-color-fitted-level-0);\n",
                     "  border: var(--sklearn-color-fitted-level-3);\n",
                     "}\n",
                     "\n",
                     ".sk-estimator-doc-link:hover span {\n",
                     "  display: block;\n",
                     "}\n",
                     "\n",
                     "/* \"?\"-specific style due to the `<a>` HTML tag */\n",
                     "\n",
                     "#sk-container-id-1 a.estimator_doc_link {\n",
                     "  float: right;\n",
                     "  font-size: 1rem;\n",
                     "  line-height: 1em;\n",
                     "  font-family: monospace;\n",
                     "  background-color: var(--sklearn-color-background);\n",
                     "  border-radius: 1rem;\n",
                     "  height: 1rem;\n",
                     "  width: 1rem;\n",
                     "  text-decoration: none;\n",
                     "  /* unfitted */\n",
                     "  color: var(--sklearn-color-unfitted-level-1);\n",
                     "  border: var(--sklearn-color-unfitted-level-1) 1pt solid;\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-1 a.estimator_doc_link.fitted {\n",
                     "  /* fitted */\n",
                     "  border: var(--sklearn-color-fitted-level-1) 1pt solid;\n",
                     "  color: var(--sklearn-color-fitted-level-1);\n",
                     "}\n",
                     "\n",
                     "/* On hover */\n",
                     "#sk-container-id-1 a.estimator_doc_link:hover {\n",
                     "  /* unfitted */\n",
                     "  background-color: var(--sklearn-color-unfitted-level-3);\n",
                     "  color: var(--sklearn-color-background);\n",
                     "  text-decoration: none;\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-1 a.estimator_doc_link.fitted:hover {\n",
                     "  /* fitted */\n",
                     "  background-color: var(--sklearn-color-fitted-level-3);\n",
                     "}\n",
                     "</style><div id=\"sk-container-id-1\" class=\"sk-top-container\"><div class=\"sk-text-repr-fallback\"><pre>DecisionTreeClassifier(criterion=&#x27;entropy&#x27;, random_state=0)</pre><b>In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook. <br />On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.</b></div><div class=\"sk-container\" hidden><div class=\"sk-item\"><div class=\"sk-estimator fitted sk-toggleable\"><input class=\"sk-toggleable__control sk-hidden--visually\" id=\"sk-estimator-id-1\" type=\"checkbox\" checked><label for=\"sk-estimator-id-1\" class=\"sk-toggleable__label fitted sk-toggleable__label-arrow fitted\">&nbsp;&nbsp;DecisionTreeClassifier<a class=\"sk-estimator-doc-link fitted\" rel=\"noreferrer\" target=\"_blank\" href=\"https://scikit-learn.org/1.5/modules/generated/sklearn.tree.DecisionTreeClassifier.html\">?<span>Documentation for DecisionTreeClassifier</span></a><span class=\"sk-estimator-doc-link fitted\">i<span>Fitted</span></span></label><div class=\"sk-toggleable__content fitted\"><pre>DecisionTreeClassifier(criterion=&#x27;entropy&#x27;, random_state=0)</pre></div> </div></div></div></div>"
                  ],
                  "text/plain": [
                     "DecisionTreeClassifier(criterion='entropy', random_state=0)"
                  ]
               },
               "execution_count": 14,
               "metadata": {},
               "output_type": "execute_result"
            }
         ],
         "source": [
            "from sklearn.tree import DecisionTreeClassifier\n",
            "from sklearn.metrics import roc_auc_score\n",
            "\n",
            "tree_classifier = DecisionTreeClassifier(random_state=0, criterion=\"entropy\")\n",
            "\n",
            "tree_classifier.fit(X_train, y_train)\n",
            "\n",
            "y_prob = tree_classifier.predict_proba(X_test)[:, 1]\n",
            "\n",
            "auroc = roc_auc_score(y_test, y_prob)\n",
            "\n",
            "print(f\"AUROC: {auroc:.4f}\")\n",
            "\n",
            "tree_classifier"
         ]
      },
      {
         "cell_type": "code",
         "execution_count": 15,
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "outputs": [
            {
               "name": "stdout",
               "output_type": "stream",
               "text": [
                  "Solution is correct!\n"
               ]
            }
         ],
         "source": [
            "assert auroc > 0.7\n",
            "\n",
            "print(\"Solution is correct!\")"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "Wynik AUROC = 0.72 wydaje się nie być taki zły, choć zdecydowanie widać miejsce do poprawy w drodze do naprawdę dobrego modelu."
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "source": [
            "## Uczenie zespołowe, bagging, lasy losowe"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "source": [
            "Bardzo często wiele klasyfikatorów działających razem daje lepsze wyniki niż pojedynczy klasyfikator. Takie podejście nazywa się **uczeniem zespołowym (ensemble learning)**. Istnieje wiele różnych podejść do tworzenia takich klasyfikatorów złożonych (ensemble classifiers).\n",
            "\n",
            "Podstawową metodą jest **bagging**:\n",
            "1. Wylosuj N (np. 100, 500, ...) próbek boostrapowych (boostrap sample) ze zbioru treningowego. Próbka boostrapowa to po prostu losowanie ze zwracaniem, gdzie dla wejściowego zbioru z M wierszami losujemy M próbek. Będą tam powtórzenia, średnio nawet 1/3, ale się tym nie przejmujemy.\n",
            "2. Wytrenuj klasyfikator bazowy (base classifier) na każdej z próbek boostrapowych.\n",
            "3. Stwórz klasyfikator złożony poprzez uśrednienie predykcji każdego z klasyfikatorów bazowych.\n",
            "\n",
            "<img src = https://upload.wikimedia.org/wikipedia/commons/thumb/c/c8/Ensemble_Bagging.svg/440px-Ensemble_Bagging.svg.png width = \"440\" height = \"248\" >\n",
            "\n",
            "Typowo klasyfikatory bazowe są bardzo proste, żeby można było szybko wytrenować ich dużą liczbę. Prawie zawsze używa się do tego drzew decyzyjnych. Dla klasyfikacji uśrednienie wyników polega na głosowaniu - dla nowej próbki każdy klasyfikator bazowy ją klasyfikuje, sumuje się głosy na każdą klasę i zwraca najbardziej popularną decyzję.\n",
            "\n",
            "Taki sposób ensemblingu zmniejsza wariancję klasyfikatora. Intuicyjnie, skoro coś uśredniamy, to siłą rzeczy będzie mniej rozrzucone, bo dużo ciężej będzie osiągnąć jakąś skrajność. Redukuje to też overfitting.\n",
            "\n",
            "**Lasy losowe (Random Forests)** to ulepszenie baggingu. Zaobserwowano, że pomimo losowania próbek boostrapowych, w baggingu poszczególne drzewa są do siebie bardzo podobne (są skorelowane), używają podobnych cech ze zbioru. My natomiast chcemy zróżnicowania, żeby mieć niski bias - redukcją wariancji zajmuje się uśrednianie. Dlatego używa się metody losowej podprzestrzeni (random subspace method) - przy każdym podziale drzewa losuje się tylko pewien podzbiór cech, których możemy użyć do tego podziału. Typowo jest to pierwiastek kwadratowy z ogólnej liczby cech.\n",
            "\n",
            "Zarówno bagging, jak i lasy losowe mają dodatkowo bardzo przyjemną własność - są mało czułe na hiperparametry, szczególnie na liczbę drzew. W praktyce wystarczy ustawić 500 czy 1000 drzew i będzie dobrze działać. Dalsze dostrajanie hiperparametrów może jeszcze trochę poprawić wyniki, ale nie tak bardzo, jak przy innych klasyfikatorach. Jest to zatem doskonały wybór domyślny, kiedy nie wiemy, jakiego klasyfikatora użyć.\n",
            "\n",
            "Dodatkowo jest to problem **embarassingly parallel** - drzewa można trenować w 100% równolegle, dzięki czemu jest to dodatkowo wydajna obliczeniowo metoda.\n",
            "\n",
            "Głębsze wytłumaczenie, z interaktywnymi wizualizacjami, można znaleźć [tutaj](https://mlu-explain.github.io/random-forest/). Dobrze tłumaczy je też [ta seria filmów](https://www.youtube.com/watch?v=J4Wdy0Wc_xQ&t=480s)."
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "### Zadanie 4 (0.5 punktu)"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "1. Wytrenuj klasyfikator Random Forest (klasa `RandomForestClassifier`). Użyj 500 drzew i entropii jako kryterium podziału.\n",
            "2. Sprawdź AUROC na zbiorze testowym.\n",
            "3. Skomentuj wynik w odniesieniu do drzewa decyzyjnego.\n",
            "\n",
            "**Uwaga:** pamiętaj o ustawieniu `random_state=0`. Dla przyspieszenia ustaw `n_jobs=-1` (użyje tylu procesów, ile masz dostępnych rdzeni procesora). Pamiętaj też o przekazaniu prawdopodobieństw do metryki AUROC."
         ]
      },
      {
         "cell_type": "code",
         "execution_count": 16,
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%%\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "outputs": [
            {
               "name": "stdout",
               "output_type": "stream",
               "text": [
                  "AUROC: 0.9011\n"
               ]
            },
            {
               "data": {
                  "text/html": [
                     "<style>#sk-container-id-2 {\n",
                     "  /* Definition of color scheme common for light and dark mode */\n",
                     "  --sklearn-color-text: black;\n",
                     "  --sklearn-color-line: gray;\n",
                     "  /* Definition of color scheme for unfitted estimators */\n",
                     "  --sklearn-color-unfitted-level-0: #fff5e6;\n",
                     "  --sklearn-color-unfitted-level-1: #f6e4d2;\n",
                     "  --sklearn-color-unfitted-level-2: #ffe0b3;\n",
                     "  --sklearn-color-unfitted-level-3: chocolate;\n",
                     "  /* Definition of color scheme for fitted estimators */\n",
                     "  --sklearn-color-fitted-level-0: #f0f8ff;\n",
                     "  --sklearn-color-fitted-level-1: #d4ebff;\n",
                     "  --sklearn-color-fitted-level-2: #b3dbfd;\n",
                     "  --sklearn-color-fitted-level-3: cornflowerblue;\n",
                     "\n",
                     "  /* Specific color for light theme */\n",
                     "  --sklearn-color-text-on-default-background: var(--sg-text-color, var(--theme-code-foreground, var(--jp-content-font-color1, black)));\n",
                     "  --sklearn-color-background: var(--sg-background-color, var(--theme-background, var(--jp-layout-color0, white)));\n",
                     "  --sklearn-color-border-box: var(--sg-text-color, var(--theme-code-foreground, var(--jp-content-font-color1, black)));\n",
                     "  --sklearn-color-icon: #696969;\n",
                     "\n",
                     "  @media (prefers-color-scheme: dark) {\n",
                     "    /* Redefinition of color scheme for dark theme */\n",
                     "    --sklearn-color-text-on-default-background: var(--sg-text-color, var(--theme-code-foreground, var(--jp-content-font-color1, white)));\n",
                     "    --sklearn-color-background: var(--sg-background-color, var(--theme-background, var(--jp-layout-color0, #111)));\n",
                     "    --sklearn-color-border-box: var(--sg-text-color, var(--theme-code-foreground, var(--jp-content-font-color1, white)));\n",
                     "    --sklearn-color-icon: #878787;\n",
                     "  }\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-2 {\n",
                     "  color: var(--sklearn-color-text);\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-2 pre {\n",
                     "  padding: 0;\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-2 input.sk-hidden--visually {\n",
                     "  border: 0;\n",
                     "  clip: rect(1px 1px 1px 1px);\n",
                     "  clip: rect(1px, 1px, 1px, 1px);\n",
                     "  height: 1px;\n",
                     "  margin: -1px;\n",
                     "  overflow: hidden;\n",
                     "  padding: 0;\n",
                     "  position: absolute;\n",
                     "  width: 1px;\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-2 div.sk-dashed-wrapped {\n",
                     "  border: 1px dashed var(--sklearn-color-line);\n",
                     "  margin: 0 0.4em 0.5em 0.4em;\n",
                     "  box-sizing: border-box;\n",
                     "  padding-bottom: 0.4em;\n",
                     "  background-color: var(--sklearn-color-background);\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-2 div.sk-container {\n",
                     "  /* jupyter's `normalize.less` sets `[hidden] { display: none; }`\n",
                     "     but bootstrap.min.css set `[hidden] { display: none !important; }`\n",
                     "     so we also need the `!important` here to be able to override the\n",
                     "     default hidden behavior on the sphinx rendered scikit-learn.org.\n",
                     "     See: https://github.com/scikit-learn/scikit-learn/issues/21755 */\n",
                     "  display: inline-block !important;\n",
                     "  position: relative;\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-2 div.sk-text-repr-fallback {\n",
                     "  display: none;\n",
                     "}\n",
                     "\n",
                     "div.sk-parallel-item,\n",
                     "div.sk-serial,\n",
                     "div.sk-item {\n",
                     "  /* draw centered vertical line to link estimators */\n",
                     "  background-image: linear-gradient(var(--sklearn-color-text-on-default-background), var(--sklearn-color-text-on-default-background));\n",
                     "  background-size: 2px 100%;\n",
                     "  background-repeat: no-repeat;\n",
                     "  background-position: center center;\n",
                     "}\n",
                     "\n",
                     "/* Parallel-specific style estimator block */\n",
                     "\n",
                     "#sk-container-id-2 div.sk-parallel-item::after {\n",
                     "  content: \"\";\n",
                     "  width: 100%;\n",
                     "  border-bottom: 2px solid var(--sklearn-color-text-on-default-background);\n",
                     "  flex-grow: 1;\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-2 div.sk-parallel {\n",
                     "  display: flex;\n",
                     "  align-items: stretch;\n",
                     "  justify-content: center;\n",
                     "  background-color: var(--sklearn-color-background);\n",
                     "  position: relative;\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-2 div.sk-parallel-item {\n",
                     "  display: flex;\n",
                     "  flex-direction: column;\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-2 div.sk-parallel-item:first-child::after {\n",
                     "  align-self: flex-end;\n",
                     "  width: 50%;\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-2 div.sk-parallel-item:last-child::after {\n",
                     "  align-self: flex-start;\n",
                     "  width: 50%;\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-2 div.sk-parallel-item:only-child::after {\n",
                     "  width: 0;\n",
                     "}\n",
                     "\n",
                     "/* Serial-specific style estimator block */\n",
                     "\n",
                     "#sk-container-id-2 div.sk-serial {\n",
                     "  display: flex;\n",
                     "  flex-direction: column;\n",
                     "  align-items: center;\n",
                     "  background-color: var(--sklearn-color-background);\n",
                     "  padding-right: 1em;\n",
                     "  padding-left: 1em;\n",
                     "}\n",
                     "\n",
                     "\n",
                     "/* Toggleable style: style used for estimator/Pipeline/ColumnTransformer box that is\n",
                     "clickable and can be expanded/collapsed.\n",
                     "- Pipeline and ColumnTransformer use this feature and define the default style\n",
                     "- Estimators will overwrite some part of the style using the `sk-estimator` class\n",
                     "*/\n",
                     "\n",
                     "/* Pipeline and ColumnTransformer style (default) */\n",
                     "\n",
                     "#sk-container-id-2 div.sk-toggleable {\n",
                     "  /* Default theme specific background. It is overwritten whether we have a\n",
                     "  specific estimator or a Pipeline/ColumnTransformer */\n",
                     "  background-color: var(--sklearn-color-background);\n",
                     "}\n",
                     "\n",
                     "/* Toggleable label */\n",
                     "#sk-container-id-2 label.sk-toggleable__label {\n",
                     "  cursor: pointer;\n",
                     "  display: block;\n",
                     "  width: 100%;\n",
                     "  margin-bottom: 0;\n",
                     "  padding: 0.5em;\n",
                     "  box-sizing: border-box;\n",
                     "  text-align: center;\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-2 label.sk-toggleable__label-arrow:before {\n",
                     "  /* Arrow on the left of the label */\n",
                     "  content: \"▸\";\n",
                     "  float: left;\n",
                     "  margin-right: 0.25em;\n",
                     "  color: var(--sklearn-color-icon);\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-2 label.sk-toggleable__label-arrow:hover:before {\n",
                     "  color: var(--sklearn-color-text);\n",
                     "}\n",
                     "\n",
                     "/* Toggleable content - dropdown */\n",
                     "\n",
                     "#sk-container-id-2 div.sk-toggleable__content {\n",
                     "  max-height: 0;\n",
                     "  max-width: 0;\n",
                     "  overflow: hidden;\n",
                     "  text-align: left;\n",
                     "  /* unfitted */\n",
                     "  background-color: var(--sklearn-color-unfitted-level-0);\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-2 div.sk-toggleable__content.fitted {\n",
                     "  /* fitted */\n",
                     "  background-color: var(--sklearn-color-fitted-level-0);\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-2 div.sk-toggleable__content pre {\n",
                     "  margin: 0.2em;\n",
                     "  border-radius: 0.25em;\n",
                     "  color: var(--sklearn-color-text);\n",
                     "  /* unfitted */\n",
                     "  background-color: var(--sklearn-color-unfitted-level-0);\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-2 div.sk-toggleable__content.fitted pre {\n",
                     "  /* unfitted */\n",
                     "  background-color: var(--sklearn-color-fitted-level-0);\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-2 input.sk-toggleable__control:checked~div.sk-toggleable__content {\n",
                     "  /* Expand drop-down */\n",
                     "  max-height: 200px;\n",
                     "  max-width: 100%;\n",
                     "  overflow: auto;\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-2 input.sk-toggleable__control:checked~label.sk-toggleable__label-arrow:before {\n",
                     "  content: \"▾\";\n",
                     "}\n",
                     "\n",
                     "/* Pipeline/ColumnTransformer-specific style */\n",
                     "\n",
                     "#sk-container-id-2 div.sk-label input.sk-toggleable__control:checked~label.sk-toggleable__label {\n",
                     "  color: var(--sklearn-color-text);\n",
                     "  background-color: var(--sklearn-color-unfitted-level-2);\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-2 div.sk-label.fitted input.sk-toggleable__control:checked~label.sk-toggleable__label {\n",
                     "  background-color: var(--sklearn-color-fitted-level-2);\n",
                     "}\n",
                     "\n",
                     "/* Estimator-specific style */\n",
                     "\n",
                     "/* Colorize estimator box */\n",
                     "#sk-container-id-2 div.sk-estimator input.sk-toggleable__control:checked~label.sk-toggleable__label {\n",
                     "  /* unfitted */\n",
                     "  background-color: var(--sklearn-color-unfitted-level-2);\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-2 div.sk-estimator.fitted input.sk-toggleable__control:checked~label.sk-toggleable__label {\n",
                     "  /* fitted */\n",
                     "  background-color: var(--sklearn-color-fitted-level-2);\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-2 div.sk-label label.sk-toggleable__label,\n",
                     "#sk-container-id-2 div.sk-label label {\n",
                     "  /* The background is the default theme color */\n",
                     "  color: var(--sklearn-color-text-on-default-background);\n",
                     "}\n",
                     "\n",
                     "/* On hover, darken the color of the background */\n",
                     "#sk-container-id-2 div.sk-label:hover label.sk-toggleable__label {\n",
                     "  color: var(--sklearn-color-text);\n",
                     "  background-color: var(--sklearn-color-unfitted-level-2);\n",
                     "}\n",
                     "\n",
                     "/* Label box, darken color on hover, fitted */\n",
                     "#sk-container-id-2 div.sk-label.fitted:hover label.sk-toggleable__label.fitted {\n",
                     "  color: var(--sklearn-color-text);\n",
                     "  background-color: var(--sklearn-color-fitted-level-2);\n",
                     "}\n",
                     "\n",
                     "/* Estimator label */\n",
                     "\n",
                     "#sk-container-id-2 div.sk-label label {\n",
                     "  font-family: monospace;\n",
                     "  font-weight: bold;\n",
                     "  display: inline-block;\n",
                     "  line-height: 1.2em;\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-2 div.sk-label-container {\n",
                     "  text-align: center;\n",
                     "}\n",
                     "\n",
                     "/* Estimator-specific */\n",
                     "#sk-container-id-2 div.sk-estimator {\n",
                     "  font-family: monospace;\n",
                     "  border: 1px dotted var(--sklearn-color-border-box);\n",
                     "  border-radius: 0.25em;\n",
                     "  box-sizing: border-box;\n",
                     "  margin-bottom: 0.5em;\n",
                     "  /* unfitted */\n",
                     "  background-color: var(--sklearn-color-unfitted-level-0);\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-2 div.sk-estimator.fitted {\n",
                     "  /* fitted */\n",
                     "  background-color: var(--sklearn-color-fitted-level-0);\n",
                     "}\n",
                     "\n",
                     "/* on hover */\n",
                     "#sk-container-id-2 div.sk-estimator:hover {\n",
                     "  /* unfitted */\n",
                     "  background-color: var(--sklearn-color-unfitted-level-2);\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-2 div.sk-estimator.fitted:hover {\n",
                     "  /* fitted */\n",
                     "  background-color: var(--sklearn-color-fitted-level-2);\n",
                     "}\n",
                     "\n",
                     "/* Specification for estimator info (e.g. \"i\" and \"?\") */\n",
                     "\n",
                     "/* Common style for \"i\" and \"?\" */\n",
                     "\n",
                     ".sk-estimator-doc-link,\n",
                     "a:link.sk-estimator-doc-link,\n",
                     "a:visited.sk-estimator-doc-link {\n",
                     "  float: right;\n",
                     "  font-size: smaller;\n",
                     "  line-height: 1em;\n",
                     "  font-family: monospace;\n",
                     "  background-color: var(--sklearn-color-background);\n",
                     "  border-radius: 1em;\n",
                     "  height: 1em;\n",
                     "  width: 1em;\n",
                     "  text-decoration: none !important;\n",
                     "  margin-left: 1ex;\n",
                     "  /* unfitted */\n",
                     "  border: var(--sklearn-color-unfitted-level-1) 1pt solid;\n",
                     "  color: var(--sklearn-color-unfitted-level-1);\n",
                     "}\n",
                     "\n",
                     ".sk-estimator-doc-link.fitted,\n",
                     "a:link.sk-estimator-doc-link.fitted,\n",
                     "a:visited.sk-estimator-doc-link.fitted {\n",
                     "  /* fitted */\n",
                     "  border: var(--sklearn-color-fitted-level-1) 1pt solid;\n",
                     "  color: var(--sklearn-color-fitted-level-1);\n",
                     "}\n",
                     "\n",
                     "/* On hover */\n",
                     "div.sk-estimator:hover .sk-estimator-doc-link:hover,\n",
                     ".sk-estimator-doc-link:hover,\n",
                     "div.sk-label-container:hover .sk-estimator-doc-link:hover,\n",
                     ".sk-estimator-doc-link:hover {\n",
                     "  /* unfitted */\n",
                     "  background-color: var(--sklearn-color-unfitted-level-3);\n",
                     "  color: var(--sklearn-color-background);\n",
                     "  text-decoration: none;\n",
                     "}\n",
                     "\n",
                     "div.sk-estimator.fitted:hover .sk-estimator-doc-link.fitted:hover,\n",
                     ".sk-estimator-doc-link.fitted:hover,\n",
                     "div.sk-label-container:hover .sk-estimator-doc-link.fitted:hover,\n",
                     ".sk-estimator-doc-link.fitted:hover {\n",
                     "  /* fitted */\n",
                     "  background-color: var(--sklearn-color-fitted-level-3);\n",
                     "  color: var(--sklearn-color-background);\n",
                     "  text-decoration: none;\n",
                     "}\n",
                     "\n",
                     "/* Span, style for the box shown on hovering the info icon */\n",
                     ".sk-estimator-doc-link span {\n",
                     "  display: none;\n",
                     "  z-index: 9999;\n",
                     "  position: relative;\n",
                     "  font-weight: normal;\n",
                     "  right: .2ex;\n",
                     "  padding: .5ex;\n",
                     "  margin: .5ex;\n",
                     "  width: min-content;\n",
                     "  min-width: 20ex;\n",
                     "  max-width: 50ex;\n",
                     "  color: var(--sklearn-color-text);\n",
                     "  box-shadow: 2pt 2pt 4pt #999;\n",
                     "  /* unfitted */\n",
                     "  background: var(--sklearn-color-unfitted-level-0);\n",
                     "  border: .5pt solid var(--sklearn-color-unfitted-level-3);\n",
                     "}\n",
                     "\n",
                     ".sk-estimator-doc-link.fitted span {\n",
                     "  /* fitted */\n",
                     "  background: var(--sklearn-color-fitted-level-0);\n",
                     "  border: var(--sklearn-color-fitted-level-3);\n",
                     "}\n",
                     "\n",
                     ".sk-estimator-doc-link:hover span {\n",
                     "  display: block;\n",
                     "}\n",
                     "\n",
                     "/* \"?\"-specific style due to the `<a>` HTML tag */\n",
                     "\n",
                     "#sk-container-id-2 a.estimator_doc_link {\n",
                     "  float: right;\n",
                     "  font-size: 1rem;\n",
                     "  line-height: 1em;\n",
                     "  font-family: monospace;\n",
                     "  background-color: var(--sklearn-color-background);\n",
                     "  border-radius: 1rem;\n",
                     "  height: 1rem;\n",
                     "  width: 1rem;\n",
                     "  text-decoration: none;\n",
                     "  /* unfitted */\n",
                     "  color: var(--sklearn-color-unfitted-level-1);\n",
                     "  border: var(--sklearn-color-unfitted-level-1) 1pt solid;\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-2 a.estimator_doc_link.fitted {\n",
                     "  /* fitted */\n",
                     "  border: var(--sklearn-color-fitted-level-1) 1pt solid;\n",
                     "  color: var(--sklearn-color-fitted-level-1);\n",
                     "}\n",
                     "\n",
                     "/* On hover */\n",
                     "#sk-container-id-2 a.estimator_doc_link:hover {\n",
                     "  /* unfitted */\n",
                     "  background-color: var(--sklearn-color-unfitted-level-3);\n",
                     "  color: var(--sklearn-color-background);\n",
                     "  text-decoration: none;\n",
                     "}\n",
                     "\n",
                     "#sk-container-id-2 a.estimator_doc_link.fitted:hover {\n",
                     "  /* fitted */\n",
                     "  background-color: var(--sklearn-color-fitted-level-3);\n",
                     "}\n",
                     "</style><div id=\"sk-container-id-2\" class=\"sk-top-container\"><div class=\"sk-text-repr-fallback\"><pre>RandomForestClassifier(criterion=&#x27;entropy&#x27;, n_estimators=500, n_jobs=-1,\n",
                     "                       random_state=0)</pre><b>In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook. <br />On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.</b></div><div class=\"sk-container\" hidden><div class=\"sk-item\"><div class=\"sk-estimator fitted sk-toggleable\"><input class=\"sk-toggleable__control sk-hidden--visually\" id=\"sk-estimator-id-2\" type=\"checkbox\" checked><label for=\"sk-estimator-id-2\" class=\"sk-toggleable__label fitted sk-toggleable__label-arrow fitted\">&nbsp;&nbsp;RandomForestClassifier<a class=\"sk-estimator-doc-link fitted\" rel=\"noreferrer\" target=\"_blank\" href=\"https://scikit-learn.org/1.5/modules/generated/sklearn.ensemble.RandomForestClassifier.html\">?<span>Documentation for RandomForestClassifier</span></a><span class=\"sk-estimator-doc-link fitted\">i<span>Fitted</span></span></label><div class=\"sk-toggleable__content fitted\"><pre>RandomForestClassifier(criterion=&#x27;entropy&#x27;, n_estimators=500, n_jobs=-1,\n",
                     "                       random_state=0)</pre></div> </div></div></div></div>"
                  ],
                  "text/plain": [
                     "RandomForestClassifier(criterion='entropy', n_estimators=500, n_jobs=-1,\n",
                     "                       random_state=0)"
                  ]
               },
               "execution_count": 16,
               "metadata": {},
               "output_type": "execute_result"
            }
         ],
         "source": [
            "from sklearn.ensemble import RandomForestClassifier\n",
            "\n",
            "forest_classifier = RandomForestClassifier(random_state=0, n_estimators=500, criterion=\"entropy\", n_jobs=-1)\n",
            "\n",
            "forest_classifier.fit(X_train, y_train)\n",
            "\n",
            "y_prob = forest_classifier.predict_proba(X_test)[:, 1]\n",
            "\n",
            "auroc = roc_auc_score(y_test, y_prob)\n",
            "\n",
            "print(f\"AUROC: {auroc:.4f}\")\n",
            "\n",
            "forest_classifier\n"
         ]
      },
      {
         "cell_type": "code",
         "execution_count": 17,
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "outputs": [
            {
               "name": "stdout",
               "output_type": "stream",
               "text": [
                  "Solution is correct!\n"
               ]
            }
         ],
         "source": [
            "assert auroc > 0.85\n",
            "\n",
            "print(\"Solution is correct!\")"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "Wynik AUROC = 0.9 jest zdecydowanie lepszy niż dla drzewa decyzyjnego, co więcej można by go uznać za obiektywnie bardzo dobry."
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "source": [
            "Jak zobaczymy poniżej, wynik ten możemy jednak jeszcze ulepszyć!"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "source": [
            "## Oversampling, SMOTE"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "source": [
            "W przypadku zbiorów niezbalansowanych można dokonać **balansowania (balancing)** zbioru. Są tutaj 2 metody:\n",
            "- **undersampling**: usunięcie przykładów z klasy dominującej\n",
            "- **oversampling**: wygenerowanie dodatkowych przykładów z klasy mniejszościowej\n",
            "\n",
            "Undersampling działa dobrze, kiedy niezbalansowanie jest niewielkie, a zbiór jest duży (możemy sobie pozwolić na usunięcie jego części). Oversampling typowo daje lepsze wyniki, istnieją dla niego bardzo efektywne algorytmy. W przypadku bardzo dużego niezbalansowania można zrobić oba.\n",
            "\n",
            "Typowym algorytmem oversamplingu jest **SMOTE (Synthetic Minority Oversampling TEchnique)**. Działa on następująco:\n",
            "1. Idź po kolei po przykładach z klasy mniejszościowej\n",
            "2. Znajdź `k` najbliższych przykładów dla próbki, typowo `k=5`\n",
            "3. Wylosuj tylu sąsiadów, ile trzeba do oversamplingu, np. jeżeli chcemy zwiększyć klasę mniejszościową 3 razy (o 200%), to wylosuj 2 z 5 sąsiadów\n",
            "4. Dla każdego z wylosowanych sąsiadów wylosuj punkt na linii prostej między próbką a tym sąsiadem. Dodaj ten punkt jako nową próbkę do zbioru\n",
            "\n",
            "<img src = https://miro.medium.com/max/734/1*yRumRhn89acByodBz0H7oA.png >\n",
            "\n",
            "Taka technika generuje przykłady bardzo podobne do prawdziwych, więc nie zaburza zbioru, a jednocześnie pomaga klasyfikatorom, bo \"zagęszcza\" przestrzeń, w której znajduje się klasa pozytywna.\n",
            "\n",
            "Algorytm SMOTE, jego warianty i inne algorytmy dla problemów niezbalansowanych implementuje biblioteka Imbalanced-learn."
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "execution": {
               "iopub.execute_input": "2024-10-07T13:36:43.138444Z",
               "iopub.status.busy": "2024-10-07T13:36:43.137348Z",
               "iopub.status.idle": "2024-10-07T13:36:43.147546Z",
               "shell.execute_reply": "2024-10-07T13:36:43.144776Z",
               "shell.execute_reply.started": "2024-10-07T13:36:43.138374Z"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "### Zadanie 5 (1 punkt)"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "Użyj SMOTE do zbalansowania zbioru treningowego (nie używa się go na zbiorze testowym!) (klasa `SMOTE`). Wytrenuj drzewo decyzyjne oraz las losowy na zbalansowanym zbiorze, użyj tych samych argumentów co wcześniej. Pamiętaj o użyciu wszędzie stałego `random_state=0` oraz przekazaniu prawdopodobieństw do AUROC. Skomentuj wynik.\n",
            "\n",
            "Wartość ROC drzewa decyzyjnego przypisz do zmiennej `tree_roc`, a lasu do `forest_roc`."
         ]
      },
      {
         "cell_type": "code",
         "execution_count": 18,
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%%\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "outputs": [
            {
               "name": "stdout",
               "output_type": "stream",
               "text": [
                  "Decision Tree AUROC: 0.7100\n",
                  "Random Forest AUROC: 0.9042\n"
               ]
            }
         ],
         "source": [
            "from imblearn.over_sampling import SMOTE\n",
            "\n",
            "smote = SMOTE(random_state=0)\n",
            "X_train_resampled, y_train_resampled = smote.fit_resample(X_train, y_train)\n",
            "\n",
            "tree_classifier = DecisionTreeClassifier(random_state=0, criterion=\"entropy\")\n",
            "tree_classifier.fit(X_train_resampled, y_train_resampled)\n",
            "\n",
            "y_prob = tree_classifier.predict_proba(X_test)[:, 1]\n",
            "tree_roc = roc_auc_score(y_test, y_prob)\n",
            "\n",
            "print(f\"Decision Tree AUROC: {tree_roc:.4f}\")\n",
            "\n",
            "forest_classifier = RandomForestClassifier(random_state=0, n_estimators=500, criterion=\"entropy\", n_jobs=-1)\n",
            "forest_classifier.fit(X_train_resampled, y_train_resampled)\n",
            "\n",
            "y_prob = forest_classifier.predict_proba(X_test)[:, 1]\n",
            "forest_roc = roc_auc_score(y_test, y_prob)\n",
            "\n",
            "print(f\"Random Forest AUROC: {forest_roc:.4f}\")\n",
            "\n",
            "\n"
         ]
      },
      {
         "cell_type": "code",
         "execution_count": 19,
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "outputs": [],
         "source": [
            "assert 0.6 < tree_roc < 0.8\n",
            "assert 0.8 < forest_roc < 0.95"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "Jak widać wyniki obu klasyfikacji są lekko lepsze po zastosowaniu balansowania."
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "source": [
            "W dalszej części laboratorium używaj zbioru po zastosowaniu SMOTE do treningu klasyfikatorów."
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "source": [
            "## Dostrajanie (tuning) hiperparametrów"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "source": [
            "Lasy losowe są stosunkowo mało czułe na dobór hiperparametrów - i dobrze, bo mają ich dość dużo. Można zawsze jednak spróbować to zrobić, a w szczególności najważniejszy jest parametr `max_features`, oznaczający, ile cech losować przy każdym podziale drzewa. Typowo sprawdza się wartości z zakresu `[0.1, 0.5]`.\n",
            "\n",
            "W kwestii szybkości, kiedy dostrajamy hiperparametry, to mniej oczywiste jest, jakiego `n_jobs` użyć. Z jednej strony klasyfikator może być trenowany na wielu procesach, a z drugiej można trenować wiele klasyfikatorów na różnych zestawach hiperparametrów równolegle. Jeżeli nasz klasyfikator bardzo dobrze się uwspółbieżnia (jak Random Forest), to można dać mu nawet wszystkie rdzenie, a za to wypróbowywać kolejne zestawy hiperparametrów sekwencyjnie. Warto ustawić parametr `verbose` na 2 lub więcej, żeby dostać logi podczas długiego treningu i mierzyć czas wykonania. W praktyce ustawia się to metodą prób i błędów."
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "### Zadanie 6 (1 punkt)"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "1. Dobierz wartość hiperparametru `max_features`:\n",
            "   - użyj grid search z 5 foldami\n",
            "   - wypróbuj wartości `[0.1, 0.2, 0.3, 0.4, 0.5]`\n",
            "   - wybierz model o najwyższym AUROC (argument `scoring`)\n",
            "2. Sprawdź, jaka była optymalna wartość `max_features`. Jest to atrybut wytrenowanego `GridSearchCV`.\n",
            "3. Skomentuj wynik. Czy warto było poświęcić czas i zasoby na tę procedurę?\n",
            "4. Wynik przypisz do zmiennej `auroc`.\n",
            "\n",
            "**Uwaga:**\n",
            "- pamiętaj, żeby jako estymatora przekazanego do grid search'a użyć instancji Random Forest, która ma już ustawione `random_state=0` i `n_jobs`"
         ]
      },
      {
         "cell_type": "code",
         "execution_count": 20,
         "metadata": {
            "editable": true,
            "pycharm": {
               "is_executing": true,
               "name": "#%%\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "outputs": [
            {
               "name": "stdout",
               "output_type": "stream",
               "text": [
                  "Optimal 'max_features': 0.1\n",
                  "Test AUROC: 0.9024\n"
               ]
            }
         ],
         "source": [
            "from sklearn.model_selection import GridSearchCV\n",
            "from sklearn.ensemble import RandomForestClassifier\n",
            "from sklearn.metrics import roc_auc_score, make_scorer\n",
            "\n",
            "rfc = RandomForestClassifier(random_state=0, n_estimators=500, criterion=\"entropy\")\n",
            "\n",
            "param_grid = { \n",
            "    'max_features': [0.1, 0.2, 0.3, 0.4, 0.5],\n",
            "}\n",
            "\n",
            "gsc = GridSearchCV(\n",
            "    estimator=rfc,\n",
            "    param_grid=param_grid,\n",
            "    scoring=make_scorer(roc_auc_score),\n",
            "    cv=5,\n",
            "    n_jobs=-1\n",
            ")\n",
            "\n",
            "gsc.fit(X_train_resampled, y_train_resampled)\n",
            "\n",
            "best_model = gsc.best_estimator_\n",
            "best_max_features = gsc.best_params_['max_features']\n",
            "\n",
            "y_pred_probs = best_model.predict_proba(X_test)[:, 1]\n",
            "auroc = roc_auc_score(y_test, y_pred_probs)\n",
            "\n",
            "print(f\"Optimal 'max_features': {best_max_features}\")\n",
            "print(f\"Test AUROC: {auroc:.4f}\")"
         ]
      },
      {
         "cell_type": "code",
         "execution_count": 21,
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "outputs": [
            {
               "name": "stdout",
               "output_type": "stream",
               "text": [
                  "Solution is correct!\n"
               ]
            }
         ],
         "source": [
            "assert 0.9 <= auroc <= 0.95\n",
            "\n",
            "print(\"Solution is correct!\")"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "Nie warto było poświęcać czasu i zasobów, ponieważ wynik nie jest znacząco wyższy od otrzymanego wcześniej."
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "source": [
            "W praktycznych zastosowaniach data scientist wedle własnego uznana, doświadczenia, dostępnego czasu i zasobów wybiera, czy dostrajać hiperparametry i w jak szerokim zakresie. Dla Random Forest na szczęście często może nie być znaczącej potrzeby, i za to go lubimy :)"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "source": [
            "**Random Forest - podsumowanie**\n",
            "\n",
            "1. Model oparty o uczenie zespołowe\n",
            "2. Kluczowe elementy:\n",
            "   - bagging: uczenie wielu klasyfikatorów na próbkach boostrapowych\n",
            "   - metoda losowej podprzestrzeni: losujemy podzbiór cech do każdego podziału drzewa\n",
            "   - uśredniamy głosy klasyfikatorów\n",
            "3. Dość odporny na overfitting, zmniejsza wariancję błędu dzięki uśrednianiu\n",
            "4. Mało czuły na hiperparametry\n",
            "5. Przeciętnie bardzo dobre wyniki, doskonały wybór domyślny przy wybieraniu algorytmu klasyfikacji"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "source": [
            "## Boosting"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "source": [
            "Drugą bardzo ważną grupą algorytmów ensemblingu jest **boosting**, też oparty o drzewa decyzyjne. O ile Random Forest trenował wszystkie klasyfikatory bazowe równolegle i je uśredniał, o tyle boosting robi to sekwencyjnie. Drzewa te uczą się na całym zbiorze, nie na próbkach boostrapowych. Idea jest następująca: trenujemy drzewo decyzyjne, radzi sobie przeciętnie i popełnia błędy na częsci przykładów treningowych. Dokładamy kolejne, ale znające błędy swojego poprzednika, dzięki czemu może to uwzględnić i je poprawić. W związku z tym \"boostuje\" się dzięki wiedzy od poprzednika. Dokładamy kolejne drzewa zgodnie z tą samą zasadą.\n",
            "\n",
            "Jak uczyć się na błędach poprzednika? Jest to pewna **funkcja kosztu** (błędu), którą chcemy zminimalizować. Zakłada się jakąś jej konkretną postać, np. squared error dla regresji, albo logistic loss dla klasyfikacji. Później wykorzystuje się spadek wzdłuż gradientu (gradient descent), aby nauczyć się, w jakim kierunku powinny optymalizować kolejne drzewa, żeby zminimalizować błędy poprzednika. Jest to konkretnie **gradient boosting**, absolutnie najpopularniejsza forma boostingu, i jeden z najpopularniejszych i osiągających najlepsze wyniki algorytmów ML.\n",
            "\n",
            "Tyle co do intuicji. Ogólny algorytm gradient boostingu jest trochę bardziej skomplikowany. Bardzo dobrze i krok po kroku tłumaczy go [ta seria filmów na YT](https://www.youtube.com/watch?v=3CC4N4z3GJc). Szczególnie ważne implementacje gradient boostingu to **XGBoost (Extreme Gradient Boosting)** oraz **LightGBM (Light Gradient Boosting Machine)**. XGBoost był prawdziwym przełomem w ML, uzyskując doskonałe wyniki i bardzo dobrze się skalując - był wykorzystany w CERNie do wykrywania cząstki Higgsa w zbiorze z pomiarów LHC mającym 10 milionów próbek. Jego implementacja jest dość złożona, ale dobrze tłumaczy ją [inna seria filmików na YT](https://www.youtube.com/watch?v=OtD8wVaFm6E).\n",
            "\n",
            "![](xgboost.png)\n",
            "\n",
            "Obecnie najczęściej wykorzystuje się LightGBM. Został stworzony przez Microsoft na podstawie doświadczeń z XGBoostem. Został jeszcze bardziej ulepszony i przyspieszony, ale różnice są głównie implementacyjne. Różnice dobrze tłumaczy [ta prezentacja z konferencji PyData](https://www.youtube.com/watch?v=5CWwwtEM2TA) oraz [prezentacja Microsoftu](https://www.youtube.com/watch?v=5nKSMXBFhes). Dla zainteresowanych - [praktyczne aspekty LightGBM](https://www.kaggle.com/code/prashant111/lightgbm-classifier-in-python/notebook)."
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "### Zadanie 7 (0.5 punktu)"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "1. Wytrenuj klasyfikator LightGBM (klasa `LGBMClassifier`). Przekaż `importance_type=\"gain\"` - przyda nam się to za chwilę.\n",
            "2. Sprawdź AUROC na zbiorze testowym.\n",
            "3. Skomentuj wynik w odniesieniu do wcześniejszych algorytmów.\n",
            "\n",
            "Pamiętaj o `random_state`, `n_jobs` i prawdopodobieństwach dla AUROC."
         ]
      },
      {
         "cell_type": "code",
         "execution_count": 38,
         "metadata": {
            "editable": true,
            "pycharm": {
               "is_executing": true,
               "name": "#%%\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "outputs": [
            {
               "name": "stdout",
               "output_type": "stream",
               "text": [
                  "AUROC: 0.9527\n",
                  "\n",
                  "Classification Report (Default Hyperparameters):\n",
                  "              precision    recall  f1-score   support\n",
                  "\n",
                  "           0       0.98      0.99      0.98      2002\n",
                  "           1       0.76      0.55      0.64        99\n",
                  "\n",
                  "    accuracy                           0.97      2101\n",
                  "   macro avg       0.87      0.77      0.81      2101\n",
                  "weighted avg       0.97      0.97      0.97      2101\n",
                  "\n"
               ]
            }
         ],
         "source": [
            "from lightgbm import LGBMClassifier\n",
            "from sklearn.metrics import classification_report\n",
            "\n",
            "lgbm_classifier = LGBMClassifier(importance_type = \"gain\", random_state=0, n_estimators=500, n_jobs=-1)\n",
            "\n",
            "lgbm_classifier.fit(X_train_resampled, y_train_resampled)\n",
            "\n",
            "y_pred_prob = lgbm_classifier.predict_proba(X_test)[:, 1]\n",
            "\n",
            "auroc = roc_auc_score(y_test, y_pred_prob)\n",
            "\n",
            "print(f\"AUROC: {auroc:.4f}\")\n",
            "\n",
            "y_pred_tuned = lgbm_classifier.predict(X_test)\n",
            "print(\"\\nClassification Report (Default Hyperparameters):\")\n",
            "print(classification_report(y_test, y_pred_tuned))\n"
         ]
      },
      {
         "cell_type": "code",
         "execution_count": 23,
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "outputs": [
            {
               "name": "stdout",
               "output_type": "stream",
               "text": [
                  "Solution is correct!\n"
               ]
            }
         ],
         "source": [
            "assert 0.9 <= auroc <= 0.97\n",
            "\n",
            "print(\"Solution is correct!\")"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "Wynik jest zdecydowanie lepszy co pokazuje, że dynamiczny charakter tej metody daje efekty(uczenie na podstawie błędów poprzednika)."
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "source": [
            "Boosting dzięki uczeniu na poprzednich drzewach redukuje nie tylko wariancję, ale też bias w błędzie, dzięki czemu może w wielu przypadkach osiągnąć lepsze rezultaty od lasu losowego. Do tego dzięki znakomitej implementacji LightGBM jest szybszy.\n",
            "\n",
            "Boosting jest jednak o wiele bardziej czuły na hiperparametry niż Random Forest. W szczególności bardzo łatwo go przeuczyć, a większość hiperparametrów, których jest dużo, wiąże się z regularyzacją modelu. To, że teraz poszło nam lepiej z domyślnymi, jest rzadkim przypadkiem.\n",
            "\n",
            "W związku z tym, że przestrzeń hiperparametrów jest duża, przeszukanie wszystkich kombinacji nie wchodzi w grę. Zamiast tego można wylosować zadaną liczbę zestawów hiperparametrów i tylko je sprawdzić - chociaż im więcej, tym lepsze wyniki powinniśmy dostać. Służy do tego `RandomizedSearchCV`. Co więcej, klasa ta potrafi próbkować rozkłady prawdopodobieństwa, a nie tylko sztywne listy wartości, co jest bardzo przydatne przy parametrach ciągłych.\n",
            "\n",
            "Hiperparametry LightGBMa są dobrze opisane w oficjalnej dokumentacji: [wersja krótsza](https://lightgbm.readthedocs.io/en/latest/pythonapi/lightgbm.LGBMClassifier.html#lightgbm.LGBMClassifier) i [wersja dłuższa](https://lightgbm.readthedocs.io/en/latest/Parameters.html). Jest ich dużo, więc nie będziemy ich tutaj omawiać. Jeżeli chodzi o ich dostrajanie w praktyce, to przydatny jest [oficjalny guide](https://lightgbm.readthedocs.io/en/latest/Parameters-Tuning.html) oraz dyskusje na Kaggle."
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "### Zadanie 8 (1.5 punktu)"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "1. Zaimplementuj random search dla LightGBMa (klasa `RandomizedSearchCV`):\n",
            "   - użyj tylu prób, na ile pozwalają twoje zasoby obliczeniowe, ale przynajmniej 30\n",
            "   - przeszukaj przestrzeń hiperparametrów:\n",
            "    ```\n",
            "    param_grid = {\n",
            "        \"n_estimators\": [100, 250, 500],\n",
            "        \"learning_rate\": [0.05, 0.1, 0.2],\n",
            "        \"num_leaves\": [31, 48, 64],\n",
            "        \"colsample_bytree\": [0.8, 0.9, 1.0],\n",
            "        \"subsample\": [0.8, 0.9, 1.0],\n",
            "    }\n",
            "    ```\n",
            "2. Wypisz znalezione optymalne hiperparametry.\n",
            "3. Wypisz raporty z klasyfikacji (funkcja `classification_report`), dla modelu LightGBM bez i z dostrajaniem hiperparametrów.\n",
            "4. Skomentuj różnicę precyzji (precision) i czułości (recall) między modelami bez i z dostrajaniem hiperparametrów. Czy jest to pożądane zjawisko w tym przypadku?\n",
            "5. Wartość ROC przypisz do zmiennej `auroc`.\n",
            "\n",
            "**Uwaga:** \n",
            "- koniecznie ustaw `verbose=-1` przy tworzeniu `LGBMClassifier`, żeby uniknąć kolosalnej ilości logów, która potrafi też wyłączyć Jupytera\n",
            "- pamiętaj o ustawieniu `importance_type`, `random_state=0` i `n_jobs`, oraz ewentualnie `verbose` w `RandomizedSearchCV` dla śledzenia przebiegu\n",
            "- istnieje możliwość, że ustawienie `n_jobs` dla grid searcha będzie szybsze niż dla samego LightGBM; odpowiada to tuningowi wielu klasyfikatorów równolegle, przy wolniejszym treningu pojedynczych klasyfikatorów\n",
            "- nie ustawiaj wszędzie `n_jobs=-1`, bo wtedy stworzysz więcej procesów niż rdzeni i spowodujesz thread contention"
         ]
      },
      {
         "cell_type": "code",
         "execution_count": 41,
         "metadata": {
            "editable": true,
            "pycharm": {
               "is_executing": true,
               "name": "#%%\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "outputs": [
            {
               "name": "stdout",
               "output_type": "stream",
               "text": [
                  "Optimal parameters: {'subsample': 1.0, 'num_leaves': 48, 'n_estimators': 250, 'learning_rate': 0.2, 'colsample_bytree': 0.9}\n",
                  "Test AUROC: 0.9397\n",
                  "\n",
                  "Classification Report (Tuned Hyperparameters):\n",
                  "              precision    recall  f1-score   support\n",
                  "\n",
                  "           0       0.98      0.99      0.99      2002\n",
                  "           1       0.80      0.53      0.63        99\n",
                  "\n",
                  "    accuracy                           0.97      2101\n",
                  "   macro avg       0.89      0.76      0.81      2101\n",
                  "weighted avg       0.97      0.97      0.97      2101\n",
                  "\n"
               ]
            }
         ],
         "source": [
            "from sklearn.model_selection import RandomizedSearchCV\n",
            "\n",
            "param_grid = {\n",
            "    \"n_estimators\": [100, 250, 500],\n",
            "    \"learning_rate\": [0.05, 0.1, 0.2],\n",
            "    \"num_leaves\": [31, 48, 64],\n",
            "    \"colsample_bytree\": [0.8, 0.9, 1.0],\n",
            "    \"subsample\": [0.8, 0.9, 1.0],\n",
            "}\n",
            "\n",
            "rsc = RandomizedSearchCV(\n",
            "    estimator=LGBMClassifier(importance_type = \"gain\", random_state=0, n_estimators=500, verbose=-1), \n",
            "    param_distributions=param_grid, \n",
            "    n_iter=150, \n",
            "    scoring=make_scorer(roc_auc_score), \n",
            "    cv=5, \n",
            "    n_jobs=-1\n",
            ")\n",
            "\n",
            "rsc.fit(X_train_resampled, y_train_resampled)\n",
            "\n",
            "best_model = rsc.best_estimator_\n",
            "best_params = rsc.best_params_\n",
            "\n",
            "y_pred_probs = best_model.predict_proba(X_test)[:, 1]\n",
            "auroc = roc_auc_score(y_test, y_pred_probs)\n",
            "\n",
            "print(f\"Optimal parameters: {best_params}\")\n",
            "print(f\"Test AUROC: {auroc:.4f}\")\n",
            "\n",
            "y_pred_tuned = best_model.predict(X_test)\n",
            "print(\"\\nClassification Report (Tuned Hyperparameters):\")\n",
            "print(classification_report(y_test, y_pred_tuned))\n"
         ]
      },
      {
         "cell_type": "code",
         "execution_count": 25,
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "outputs": [],
         "source": [
            "assert 0.9 <= auroc <= 0.99"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "Dla klasy mniejszościowej precision wzrosło z 0.76 do 0.8, a recall lekko spadł z 0.55 do 0.53, co oznacza, że mimo, iż model rzadziej klasyfikuje \"false positive\" to częściej \"false negative\". W naszym przypadku jest to nie porządane zjawisko, gdyż chcielibyśmy wykrywać z jak największą precyzją firmy, które upadną, natomiast pomyłka w drugą stronę nie jest aż tak bolesna."
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "source": [
            "**Boosting - podsumowanie**\n",
            "\n",
            "1. Model oparty o uczenie zespołowe\n",
            "2. Kolejne modele są dodawane sekwencyjnie i uczą się na błędach poprzedników\n",
            "3. Nauka typowo jest oparta o minimalizację funkcji kosztu (błędu), z użyciem spadku wzdłuż gradientu\n",
            "4. Wiodący model klasyfikacji dla danych tabelarycznych, z 2 głównymi implementacjami: XGBoost i LightGBM\n",
            "5. Liczne hiperparametry, wymagające odpowiednich metod dostrajania"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "source": [
            "## Wyjaśnialna AI"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": []
         },
         "source": [
            "W ostatnich latach zaczęto zwracać coraz większą uwagę na wpływ sztucznej inteligencji na społeczeństwo, a na niektórych czołowych konferencjach ML nawet obowiązkowa jest sekcja \"Social impact\" w artykułach naukowych. Typowo im lepszy model, tym bardziej złożony, a najpopularniejsze modele boostingu są z natury skomplikowane. Kiedy mają podejmować krytyczne decyzje, to musimy wiedzieć, czemu predykcja jest taka, a nie inna. Jest to poddziedzina uczenia maszynowego - **wyjaśnialna AI (explainable AI, XAI)**.\n",
            "\n",
            "Taka informacja jest cenna, bo dzięki temu lepiej wiemy, co robi model. Jest to ważne z kilku powodów:\n",
            "1. Wymogi prawne - wdrażanie algorytmów w ekonomii, prawie etc. ma coraz częściej konkretne wymagania prawne co do wyjaśnialności predykcji\n",
            "2. Dodatkowa wiedza dla użytkowników - często dodatkowe obserwacje co do próbek są ciekawe same w sobie i dają wiedzę użytkownikowi (często posiadającemu specjalistyczną wiedzę z dziedziny), czasem nawet bardziej niż sam model predykcyjny\n",
            "3. Analiza modelu - dodatkowa wiedza o wewnętrznym działaniu algorytmu pozwala go lepiej zrozumieć i ulepszyć wyniki, np. przez lepszy preprocessing danych\n",
            "\n",
            "W szczególności można ją podzielić na **globalną** oraz **lokalną interpretowalność (global / local interpretability)**. Ta pierwsza próbuje wyjaśnić, czemu ogólnie model działa tak, jak działa. Analizuje strukturę modelu oraz trendy w jego predykcjach, aby podsumować w prostszy sposób jego tok myślenia. Interpretowalność lokalna z kolei dotyczy predykcji dla konkretnych próbek - czemu dla danego przykładu model podejmuje dla niego taką, a nie inną decyzję o klasyfikacji.\n",
            "\n",
            "W szczególności podstawowym sposobem interpretowalności jest **ważność cech (feature importance)**. Wyznacza ona, jak ważne są poszczególne cechy:\n",
            "- w wariancie globalnym, jak mocno model opiera się na poszczególnych cechach\n",
            "- w wariancie lokalnym, jak mocno konkretne wartości cech wpłynęły na predykcję, i w jaki sposób\n",
            "\n",
            "Teraz będzie nas interesować globalna ważność cech. Dla modeli drzewiastych definiuje się ją bardzo prosto. Każdy podział w drzewie decyzyjnym wykorzystuje jakąś cechę, i redukuje z pomocą podziału funkcję kosztu (np. entropię) o określoną ilość. Dla drzewa decyzyjnego ważność to sumaryczna redukcja entropii, jaką udało się uzyskać za pomocą danej cechy. Dla lasów losowych i boostingu sumujemy te wartości dla wszystkich drzew. Alternatywnie można też użyć liczby splitów, w jakiej została użyta dana cecha, ale jest to mniej standardowe.\n",
            "\n",
            "Warto zauważyć, że taka ważność cech jest **względna**:\n",
            "- nie mówimy, jak bardzo ogólnie ważna jest jakaś cecha, tylko jak bardzo przydatna była dla naszego modelu w celu jego wytrenowania\n",
            "- ważność cech można tylko porównywać ze sobą, np. jedna jest 2 razy ważniejsza od drugiej; nie ma ogólnych progów ważności\n",
            "\n",
            "Ze względu na powyższe, ważności cech normalizuje się często do zakresu [0, 1] dla łatwiejszego porównywania."
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "### Zadanie 9 (0.5 punktu)"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "1. Wybierz 5 najważniejszych cech dla drzewa decyzyjnego. Przedstaw wyniki na poziomym wykresie słupkowym. Użyj czytelnych nazw cech ze zmiennej `feature_names`.\n",
            "2. Powtórz powyższe dla lasu losowego, oraz dla boostingu (tutaj znormalizuj wyniki - patrz uwaga niżej). Wybierz te hiperparametry, które dały wcześniej najlepsze wyniki.\n",
            "3. Skomentuj, czy wybrane cechy twoim zdaniem mają sens jako najważniejsze cechy.\n",
            "\n",
            "**Uwaga:** Scikit-learn normalizuje ważności do zakresu [0, 1], natomiast LightGBM nie. Musisz to znormalizować samodzielnie, dzieląc przez sumę."
         ]
      },
      {
         "cell_type": "code",
         "execution_count": 40,
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "outputs": [
            {
               "data": {
                  "image/png": "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",
                  "text/plain": [
                     "<Figure size 800x400 with 1 Axes>"
                  ]
               },
               "metadata": {},
               "output_type": "display_data"
            },
            {
               "data": {
                  "image/png": "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",
                  "text/plain": [
                     "<Figure size 800x400 with 1 Axes>"
                  ]
               },
               "metadata": {},
               "output_type": "display_data"
            },
            {
               "data": {
                  "image/png": "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",
                  "text/plain": [
                     "<Figure size 800x400 with 1 Axes>"
                  ]
               },
               "metadata": {},
               "output_type": "display_data"
            }
         ],
         "source": [
            "def plot_feature_importances(top_features, model_name):\n",
            "    names, importances = zip(*top_features)\n",
            "    plt.figure(figsize=(8, 4))\n",
            "    plt.barh(names, importances, align='center')\n",
            "    plt.xlabel('Importance')\n",
            "    plt.title(f'Top 5 Features - {model_name}')\n",
            "    plt.show()\n",
            "\n",
            "dt_features = tree_classifier.feature_importances_\n",
            "dt_indices = np.argsort(dt_features)[-5:]\n",
            "dt_top_features = [(feature_names[i], dt_features[i]) for i in dt_indices]\n",
            "\n",
            "rf_features = forest_classifier.feature_importances_\n",
            "rf_indices = np.argsort(rf_features)[-5:]\n",
            "rf_top_features = [(feature_names[i], rf_features[i]) for i in rf_indices]\n",
            "\n",
            "lgbm_features = lgbm_classifier.feature_importances_\n",
            "lgbm_features_normalized = lgbm_features / np.sum(lgbm_features)\n",
            "lgbm_indices = np.argsort(lgbm_features)[-5:]\n",
            "lgbm_top_features = [(feature_names[i], lgbm_features_normalized[i]) for i in lgbm_indices]\n",
            "\n",
            "\n",
            "plot_feature_importances(dt_top_features, \"Decision Tree\")\n",
            "plot_feature_importances(rf_top_features, \"Random Forest\")\n",
            "plot_feature_importances(lgbm_top_features, \"LightGBM\")\n"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "Zdecydowanie otrzymane cechy mają największe znaczenie. Mówią o stosunku dochodów do kosztów, bądź obecnych wynikach firmy, co faktycznie powinno przekładać się na prawdopodobieństwo bankrupctwa."
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {},
         "source": [
            "### Dla zainteresowanych\n",
            "\n",
            "Najpopularniejszym podejściem do interpretowalności lokalnych jest **SHAP (SHapley Additive exPlanations)**, metoda oparta o kooperatywną teorię gier. Traktuje się cechy modelu jak zbiór graczy, podzielonych na dwie drużyny (koalicje): jedna chce zaklasyfikować próbkę jako negatywną, a druga jako pozytywną. O ostatecznej decyzji decyduje model, który wykorzystuje te wartości cech. Powstaje pytanie - w jakim stopniu wartości cech przyczyniły się do wyniku swojej drużyny? Można to obliczyć jako wartości Shapleya (Shapley values), które dla modeli ML oblicza algorytm SHAP. Ma on bardzo znaczące, udowodnione matematycznie zalety, a dodatkowo posiada wyjątkowo efektywną implementację dla modeli drzewiastych oraz dobre wizualizacje.\n",
            "\n",
            "Bardzo intuicyjnie, na prostym przykładzie, SHAPa wyjaśnia [pierwsza część tego artykułu](https://iancovert.com/blog/understanding-shap-sage/). Dobrze i dość szczegółówo SHAPa wyjaśnia jego autor [w tym filmie](https://www.youtube.com/watch?v=-taOhqkiuIo)."
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "pycharm": {
               "name": "#%% md\n"
            }
         },
         "source": [
            "**Wyjaśnialna AI - podsumowanie**\n",
            "\n",
            "1. Problem zrozumienia, jak wnioskuje model i czemu podejmuje dane decyzje\n",
            "2. Ważne zarówno z perspektywy data scientist'a, jak i użytkowników systemu\n",
            "3. Można wyjaśniać model lokalnie (konkretne predykcje) lub globalnie (wpływ poszczególnych cech)"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "## Zadanie 10 dla chętnych (3 punkty)"
         ]
      },
      {
         "cell_type": "markdown",
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%% md\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "source": [
            "Dokonaj selekcji cech, usuwając 20% najsłabszych cech. Może się tu przydać klasa `SelectPercentile`. Czy Random Forest i LightGBM (bez dostrajania hiperparametrów, dla uproszczenia) wytrenowane bez najsłabszych cech dają lepszy wynik (AUROC lub innej metryki)?\n",
            "\n",
            "Wykorzystaj po 1 algorytmie z 3 grup algorytmów selekcji cech:\n",
            "1. Filter methods - mierzymy ważność każdej cechy niezależnie, za pomocą pewnej miary (typowo ze statystyki lub teorii informacji), a potem odrzucamy (filtrujemy) te o najniższej ważności. Są to np. `chi2` i `mutual_info_classif` z pakietu `sklearn.feature_selection`.\n",
            "2. Embedded methods - klasyfikator sam zwraca ważność cech, jest jego wbudowaną cechą (stąd nazwa). Jest to w szczególności właściwość wszystkich zespołowych klasyfikatorów drzewiastych. Mają po wytrenowaniu atrybut `feature_importances_`.\n",
            "2. Wrapper methods - algorytmy wykorzystujące w środku używany model (stąd nazwa), mierzące ważność cech za pomocą ich wpływu na jakość klasyfikatora. Jest to np. recursive feature elimination (klasa `RFE`). W tym algorytmie trenujemy klasyfikator na wszystkich cechach, wyrzucamy najsłabszą, trenujemy znowu i tak dalej.\n",
            "\n",
            "Typowo metody filter są najszybsze, ale dają najsłabszy wynik, natomiast metody wrapper są najwolniejsze i dają najlepszy wynik. Metody embedded są gdzieś pośrodku.\n",
            "\n",
            "Dla zainteresowanych, inne znane i bardzo dobre algorytmy:\n",
            "- Relief (filter method) oraz warianty, szczególnie ReliefF, SURF i MultiSURF (biblioteka `ReBATE`): [Wikipedia](https://en.wikipedia.org/wiki/Relief_(feature_selection)), [artykuł \"Benchmarking Relief-Based Feature Selection Methods\"](https://www.researchgate.net/publication/321307194_Benchmarking_Relief-Based_Feature_Selection_Methods)\n",
            "- Boruta (wrapper method), stworzony na Uniwersytecie Warszawskim, łączący Random Forest oraz testy statystyczne (biblioteka `boruta_py`): [link 1](https://towardsdatascience.com/boruta-explained-the-way-i-wish-someone-explained-it-to-me-4489d70e154a), [link 2](https://danielhomola.com/feature%20selection/phd/borutapy-an-all-relevant-feature-selection-method/)"
         ]
      },
      {
         "cell_type": "code",
         "execution_count": null,
         "metadata": {
            "editable": true,
            "pycharm": {
               "name": "#%%\n"
            },
            "slideshow": {
               "slide_type": ""
            },
            "tags": [
               "ex"
            ]
         },
         "outputs": [],
         "source": []
      }
   ],
   "metadata": {
      "jupytext": {
         "formats": "ipynb,py:percent"
      },
      "kernelspec": {
         "display_name": ".venv",
         "language": "python",
         "name": "python3"
      },
      "language_info": {
         "codemirror_mode": {
            "name": "ipython",
            "version": 3
         },
         "file_extension": ".py",
         "mimetype": "text/x-python",
         "name": "python",
         "nbconvert_exporter": "python",
         "pygments_lexer": "ipython3",
         "version": "3.12.5"
      }
   },
   "nbformat": 4,
   "nbformat_minor": 4
}