Tarea 2.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Tarea 2: análisis de sentimientos con aprendizaje supervisado\n",
    "\n",
    "##### Por: Daniela Flores Villanueva \n",
    "\n",
    "## Sobre las librerías"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "En primer lugar, se cargan todas las librerías empleadas en esta tarea:\n",
    "- `pandas`: librería utilizada para cargar la matriz GloVe y posteriormente, para confeccionar un `DataFrame` con palabras y su sentimiento asociado.\n",
    "- `csv`: utilizada para cargar los vectores, según lo dispuesto en [este](https://stackoverflow.com/questions/37793118/load-pretrained-glove-vectors-in-python) sitio.\n",
    "- `nltk`: se usó para cargar el *opinion lexicon* descrito en el enunciado de manera conveniente. Se probó además su método para realizar lematización, proceso que consiste en, dada una palabra flexionada (en forma plural, verbo conjugado, etc), encontrar su lema, es decir, algo similar a la forma en que sería encontrada en un diccionario.\n",
    "- `string`: para obtener signos de puntuación, con el objeto de pre-procesar oraciones.\n",
    "- `sklearn`: se utilizaron sus implementaciones de *Support Vector Machines*, *K-Nearest Neighbors* y *Random Forest* para construir clasificadores para el sentimiento de palabras y posteriormente oraciones. Se aprovechó también `train_test_split` para separar los datos en conjuntos de entrenamiento y de prueba, `GridSearchCV` para la calibración de hiperparámetros y los métodos `accuracy_score`, `precision_score`, `recall_score` y `f1-score` para cuantificar el desempeño de los modelos. Finalmente `joblib` se utiliza para serializar y guardar los modelos entrenados.\n",
    "- `pickle`: para serializar la matriz de *embeddings*.\n",
    "- `matplotlib`: para mostrar resultados de manera gráfica.\n",
    "- `countries`: diccionario con información de países para la sección de prejuicios."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[nltk_data] Downloading package stopwords to\n",
      "[nltk_data]     /Users/DanielaFlores/nltk_data...\n",
      "[nltk_data]   Package stopwords is already up-to-date!\n",
      "[nltk_data] Downloading package opinion_lexicon to\n",
      "[nltk_data]     /Users/DanielaFlores/nltk_data...\n",
      "[nltk_data]   Package opinion_lexicon is already up-to-date!\n",
      "[nltk_data] Downloading package punkt to\n",
      "[nltk_data]     /Users/DanielaFlores/nltk_data...\n",
      "[nltk_data]   Package punkt is already up-to-date!\n"
     ]
    }
   ],
   "source": [
    "import pandas as pd\n",
    "import csv\n",
    "from nltk.stem import WordNetLemmatizer\n",
    "import numpy as np\n",
    "import nltk\n",
    "nltk.download(\"stopwords\")\n",
    "nltk.download(\"opinion_lexicon\")\n",
    "nltk.download('punkt')\n",
    "from nltk.corpus import opinion_lexicon\n",
    "from nltk.corpus import stopwords\n",
    "from nltk.tokenize import word_tokenize\n",
    "from string import punctuation\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.model_selection import GridSearchCV\n",
    "from sklearn.svm import SVC\n",
    "from sklearn.neighbors import KNeighborsClassifier\n",
    "from sklearn.ensemble import RandomForestClassifier\n",
    "from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score\n",
    "from sklearn.externals import joblib\n",
    "import pickle\n",
    "import matplotlib.pyplot as plt\n",
    "from countryinfo import countries"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "np.set_printoptions(formatter={'float_kind':'{:f}'.format})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "GLOVE_PATH = \"./glove.42B.300d.txt\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "NON_WORDS = list(punctuation)\n",
    "NON_WORDS.extend(map(str, range(10)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "ENGLISH_STOPWORDS = stopwords.words(\"english\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "La forma de cargar el opinion lexicon es la descrita en la [documentación](http://www.nltk.org/api/nltk.corpus.reader.html?highlight=wordnet) de NLTK."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "positive_data = opinion_lexicon.positive()\n",
    "negative_data = opinion_lexicon.negative()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Con el fin de tener una representación más general de las palabras, se convierte cada vocablo en las listas anteriores a minúsculas."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "positive_data = [word.lower() for word in positive_data]\n",
    "negative_data = [word.lower() for word in negative_data]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "A continuación, se carga la matriz de *embeddings*:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "glove_matrix = pd.read_csv(GLOVE_PATH, sep=\" \", index_col=0, header=None, quoting=csv.QUOTE_NONE, na_values=None, keep_default_na=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "En el mismo recurso para cargar la matriz en memoria (adjunto en la descripción de las librerías utilizadas), se menciona una función que, dada una palabra, retorna su *embedding*. Esta función es `vec(w)`, definida a continuación:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "def vec(w): \n",
    "    return glove_matrix.loc[w].as_matrix()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Breve análisis exploratorio\n",
    "\n",
    "No se puede empezar a trabajar sin tener al menos una noción básica de la composición de la matriz y de las palabras con etiqueta. Por esta razón, en primer lugar cabe preguntarse cuántas palabras hay en la matriz de *embeddings*."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1917494\n"
     ]
    }
   ],
   "source": [
    "glove_words = set(glove_matrix.index.tolist())\n",
    "print(len(glove_words))\n",
    "pickle.dump(glove_words, open(\"glove_words.pkl\", \"wb\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Podría ocurrir que no todas las palabras que tenemos etiquetadas según su polaridad tengan su representación vectorial en GloVe, por lo que conviene realizar el siguiente análisis:\n",
    "Dada la unión entre las palabras de polaridad positiva y las de polaridad negativa, se revisa si están o no en la matriz. De no estar, se agregan a la lista `missing_words`, para decidir qué hacer con ellas futuramente."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "161\n"
     ]
    }
   ],
   "source": [
    "missing_words = set()\n",
    "for word in set(positive_data).union(set(negative_data)):\n",
    "    if word not in glove_words:\n",
    "        missing_words.add(word)\n",
    "print(len(missing_words))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Es posible notar que hay 161 palabras del *opinion lexicon* para las que no se tiene una representación vectorial. Antes de decidir qué hacer con estas palabras faltantes, conviene preguntarse si existen palabras que estén etiquetadas tanto de polaridad positiva como de polaridad negativa."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'envious', 'enviously', 'enviousness'}\n"
     ]
    }
   ],
   "source": [
    "wrong_tag = set(positive_data).intersection(negative_data)\n",
    "print(wrong_tag)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Así, se evidencia que tres palabras relacionadas a la envidia parecen estar mal etiquetadas, pues no deberían aparecer entre las palabras positivas."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "A continuación, se procede a realizar lematización de los vocablos positivos y negativos, para ver si existe algún cambio en la cantidad palabras sin *embedding* tras este cambio."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "positive_data_lemmatized = {nltk.stem.WordNetLemmatizer().lemmatize(word) for word in positive_data}\n",
    "negative_data_lemmatized = {nltk.stem.WordNetLemmatizer().lemmatize(word) for word in negative_data}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "160\n"
     ]
    }
   ],
   "source": [
    "missing_words_lemmatized = set()\n",
    "for word in positive_data_lemmatized.union(set(negative_data_lemmatized)):\n",
    "    if word not in glove_words:\n",
    "        missing_words_lemmatized.add(word)\n",
    "print(len(missing_words_lemmatized))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'envious', 'enviously', 'plea', 'enviousness'}\n"
     ]
    }
   ],
   "source": [
    "wrong_tag_lemmatized = set(positive_data_lemmatized).intersection(negative_data_lemmatized)\n",
    "print(wrong_tag_lemmatized)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "positive_data_lemmatized = positive_data_lemmatized - wrong_tag_lemmatized"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Al lematizar, es posible notar que hay una palabra menos que no tiene representación vectorial. En la siguiente celda, podrá evidenciarse qué palabra es la que hace la diferencia:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'spoilages'}\n"
     ]
    }
   ],
   "source": [
    "print(missing_words - missing_words_lemmatized)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Preprocesamiento de los datos\n",
    "\n",
    "En las siguientes líneas, se iterará por sobre todas las palabras en los conjuntos de palabras reducidas a su lema (`positive_data_lemmatized` y `negative_data_lemmatized`), con el fin de encontrar su vector en la matriz de *embeddings* GloVe. Estos vectores se guardarán en las matrices `positive_matrix` y `negative_matrix`, que contendrán, respectivamente, las palabras de sentimiento positivo y negativo.\n",
    "\n",
    "Es importante destacar el trato que reciben las palabras para las que no se dispone una representación vectorial. Tras [investigar](https://github.com/stanfordnlp/GloVe/search?utf8=%E2%9C%93&q=unk&type=), se descubrió que GloVe incluye un vector estándar para las palabras sin *word embedding*, el que en esta tarea puede obtenerse con `vec(\"unk\")`. Así, a todas las palabras del *opinion lexicon* que no tienen *embedding* en GloVe, se les asigna el vector antes mencionado."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "def build_matrices(words):\n",
    "    matrix = np.empty((0, 300))\n",
    "    for word in words:\n",
    "        if word in glove_words:\n",
    "            matrix = np.append(matrix, [vec(word)], axis=0)\n",
    "        else:\n",
    "            matrix = np.append(matrix, [vec(\"unk\")], axis=0) \n",
    "    return matrix"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "positive_data_lemmatized = list(positive_data_lemmatized)\n",
    "negative_data_lemmatized = list(negative_data_lemmatized)\n",
    "positive_matrix = build_matrices(positive_data_lemmatized)\n",
    "negative_matrix = build_matrices(negative_data_lemmatized)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Se observa que las matrices recién formadas efectivamente contienen todas las representaciones vectoriales que corresponden:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1951 (1951, 300)\n",
      "4492 (4492, 300)\n"
     ]
    }
   ],
   "source": [
    "print(len(positive_data_lemmatized), positive_matrix.shape)\n",
    "print(len(negative_data_lemmatized), negative_matrix.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Se definen las etiquetas para cada ejemplo en el *data set*."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "positive_labels = [1] * positive_matrix.shape[0]\n",
    "negative_labels = [0] * negative_matrix.shape[0]\n",
    "all_labels = positive_labels + negative_labels"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Finalmente, se concatenan las matrices `positive_matrix` y `negative_matrix`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(6443, 300)\n"
     ]
    }
   ],
   "source": [
    "concatenated_matrix = np.concatenate([positive_matrix, negative_matrix], axis=0)\n",
    "print(concatenated_matrix.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Clasificación de sentimiento de palabras\n",
    "\n",
    "### Separación entre entrenamiento y *testing*\n",
    "\n",
    "Con el objeto de hacer una correcta clasificación, se separan los datos en entrenamiento y *testing*. En particular, se utiliza el 80% de los datos para entrenar los modelos y 20% para probarlos."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "all_words = positive_data_lemmatized + negative_data_lemmatized\n",
    "X_train, X_test, y_train, y_test, words_train, words_test = train_test_split(concatenated_matrix, all_labels, all_words, test_size=0.3, random_state=7)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Calibración de híper-parámetros\n",
    "\n",
    "Para mejorar el desempeño de los modelos, se define la siguiente función para calibrar los híper-parámetros correspondientes. Esto se hace posible gracias a `GridSearchCV`, que recibe un modelo, los datos de entrenamiento, los híper-parámetros a testear y la cantidad de *folds* en que se separarán los datos (de acuerdo a la documentación, los datos se separán de forma que se conserven las proporciones originales de las clases). Con esta información, `GridSearchCV` probará todas las combinaciones de híper-parámetros y se realizará validación cruzada con ellas. Finalmente se retornará la combinación de híper-parámetros que haya obtenido mejor `cross_val_score`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "def estimator_grid_search(estimator, parameters):\n",
    "    clf = GridSearchCV(estimator=estimator, param_grid=parameters, n_jobs=-1, cv=5)\n",
    "    clf.fit(X_train, y_train)   \n",
    "    print(\"Mejor accuracy: {}\".format(clf.best_score_))\n",
    "    return clf"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Medición del desempeño de los modelos\n",
    "\n",
    "Para evaluar los modelos, es necesario cuantificar su desempeño. Por esta razón, se define la función `get_metrics` que dado un conjunto de entrenamiento con sus respectivas etiquetas, permite obtener un diccionario con las principales métricas para tareas de clasificación:\n",
    "\n",
    "- *Accuracy*: porcentaje de elementos bien clasificados.\n",
    "- *Precision*: de la cantidad de ejemplos que el modelo clasificó como positivos, ¿qué porcentaje de ellos de verdad lo eran?\n",
    "- *Recall*: ¿qué porcentaje de los verdaderos ejemplos positivos fueron clasificados como tal por el modelo?\n",
    "- *F1-score*: media armónica de *precision* y *recall*."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_metrics(clf, X_test, y_test):\n",
    "    metrics = {}\n",
    "    predictions = clf.predict(X_test)\n",
    "    metrics[\"accuracy\"] = accuracy_score(y_test, predictions)\n",
    "    metrics[\"precision\"] = precision_score(y_test, predictions)\n",
    "    metrics[\"recall\"] = recall_score(y_test, predictions)\n",
    "    metrics[\"f1-score\"] = f1_score(y_test, predictions)\n",
    "    return metrics"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Guardar modelos\n",
    "\n",
    "La función `save_model` permite guardar los clasificadores entrenados en esta tarea."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "def save_model(clf, clf_name):\n",
    "    joblib.dump(clf, \"best_{}.pkl\".format(clf_name))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### SVM\n",
    "\n",
    "\n",
    "El primer modelo entrenado en esta tarea corresponde a *Support Vector Machines*. Para ajustarlo, primero se realiza la calibración de híper-parámetros. Una vez encontrada la mejor combinación de ellos, se entrena el clasificador con ellos y los datos de entrenamiento. Finalmente, se obtienen las métricas antes descritas sobre el set de *testing*. Como los datos de este conjunto no han sido vistos por el modelo, puede asumirse que las métricas son una buena muestra de la calidad del clasificador."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Mejor accuracy: 0.9374722838137473\n",
      "Mejor C: 10\n",
      "Mejor kernel: rbf\n",
      "Mejor Gamma: 0.001\n",
      "{'accuracy': 0.9436109674081739, 'precision': 0.9330855018587361, 'recall': 0.8730434782608696, 'f1-score': 0.9020664869721473}\n"
     ]
    }
   ],
   "source": [
    "svm_params = {\n",
    "    \"C\": [1, 10],\n",
    "    \"gamma\": [0.001, 1/300],\n",
    "    \"kernel\": [\"linear\", \"rbf\"]\n",
    "}\n",
    "tuning_svm = estimator_grid_search(SVC(), svm_params)\n",
    "print(\"Mejor C: {}\".format(tuning_svm.best_estimator_.C))\n",
    "print(\"Mejor kernel: {}\".format(tuning_svm.best_estimator_.kernel))\n",
    "print(\"Mejor Gamma: {}\".format(tuning_svm.best_estimator_.gamma))\n",
    "final_svm = SVC(C=tuning_svm.best_estimator_.C, \n",
    "                kernel=tuning_svm.best_estimator_.kernel, \n",
    "                gamma=tuning_svm.best_estimator_.gamma, probability=True)\n",
    "final_svm.fit(X_train, y_train)\n",
    "svm_final_metrics = get_metrics(final_svm, X_test, y_test)\n",
    "print(svm_final_metrics)\n",
    "save_model(final_svm, \"SVM\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### KNN\n",
    "\n",
    "Se realizan los mismos pasos que en el caso del SVM, pero ahora para entrenar un clasificador KNN."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Mejor accuracy: 0.90509977827051\n",
      "Mejor K: 3\n",
      "Mejor algoritmo de búsqueda de vecinos: kd_tree\n",
      "{'accuracy': 0.9146404552509053, 'precision': 0.9035433070866141, 'recall': 0.7982608695652174, 'f1-score': 0.8476454293628809}\n"
     ]
    }
   ],
   "source": [
    "knn_params = {\"n_neighbors\": np.arange(3) + 1, \"algorithm\": [\"kd_tree\", \"ball_tree\"]}\n",
    "tuning_knn = estimator_grid_search(KNeighborsClassifier(), knn_params)\n",
    "print(\"Mejor K: {}\".format(tuning_knn.best_estimator_.n_neighbors)) \n",
    "print(\"Mejor algoritmo de búsqueda de vecinos: {}\".format(tuning_knn.best_estimator_.algorithm))\n",
    "final_knn = KNeighborsClassifier(n_neighbors=tuning_knn.best_estimator_.n_neighbors, \n",
    "                                 algorithm=tuning_knn.best_estimator_.algorithm)\n",
    "final_knn.fit(X_train, y_train)\n",
    "knn_final_metrics = get_metrics(final_knn, X_test, y_test)\n",
    "print(knn_final_metrics)\n",
    "save_model(final_knn, \"KNN\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Random Forest\n",
    "\n",
    "Finalmente, se evaluó el desempeño de un ensamble de árboles de decisión sobre esta tarea de clasificación."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Mejor accuracy: 0.8583148558758315\n",
      "Mejor cantidad de árboles: 9\n",
      "Mejor criterio de división: entropy\n",
      "Mejor cantidad máxima de features: None\n",
      "{'accuracy': 0.8613554061045008, 'precision': 0.8259023354564756, 'recall': 0.6765217391304348, 'f1-score': 0.7437858508604207}\n"
     ]
    }
   ],
   "source": [
    "rf_params = {\n",
    "    \"n_estimators\": np.arange(5, 10) + 1, \n",
    "    \"criterion\": [\"gini\", \"entropy\"], \n",
    "    \"max_features\": [\"sqrt\", \"log2\", None]}\n",
    "tuning_rf = estimator_grid_search(RandomForestClassifier(), rf_params)\n",
    "print(\"Mejor cantidad de árboles: {}\".format(tuning_rf.best_estimator_.n_estimators))\n",
    "print(\"Mejor criterio de división: {}\".format(tuning_rf.best_estimator_.criterion))\n",
    "print(\"Mejor cantidad máxima de features: {}\".format(tuning_rf.best_estimator_.max_features))\n",
    "final_rf = RandomForestClassifier(n_estimators=tuning_rf.best_estimator_.n_estimators,\n",
    "                                 criterion=tuning_rf.best_estimator_.criterion,\n",
    "                                 max_features=tuning_rf.best_estimator_.max_features)\n",
    "final_rf.fit(X_train, y_train)\n",
    "rf_final_metrics = get_metrics(final_rf, X_test, y_test)\n",
    "print(rf_final_metrics)\n",
    "save_model(final_rf, \"RF\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "En el enunciado de la tarea, se pide que los modelos tengan salida continua. Es por eso que se define la función `model_predict`, que retorna la probabilidad de pertenencia a cada una de las clases (positivo/negativo)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "def model_predict(clf, test_set):\n",
    "    return clf.predict_proba(test_set)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Comparación de clasificadores\n",
    "\n",
    "A continuación, se comparan los clasificadores según las métricas anteriormente descritas:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/matplotlib/cbook/deprecation.py:107: MatplotlibDeprecationWarning: Adding an axes using the same arguments as a previous axes currently reuses the earlier instance.  In a future version, a new instance will always be created and returned.  Meanwhile, this warning can be suppressed, and the future behavior ensured, by passing a unique label to each axes instance.\n",
      "  warnings.warn(message, mplDeprecation, stacklevel=1)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "Text(0.5,1,'Comparación de métricas de los distintos clasificadores')"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "r1 = np.arange(4)\n",
    "r2 = [x + 0.25 for x in r1]\n",
    "r3 = [x + 0.25 for x in r2]\n",
    "\n",
    "labels = [k for k, v in svm_final_metrics.items()]\n",
    "svm_values = [v for k, v in svm_final_metrics.items()]\n",
    "knn_values = [v for k, v in knn_final_metrics.items()]\n",
    "rf_values = [v for k, v in rf_final_metrics.items()]\n",
    "\n",
    "plt.xticks([r + 0.25 for r in range(4)], labels)\n",
    "ax = plt.subplot(111)\n",
    "ax.bar(r1, svm_values,width=0.25,color='b', align='center')\n",
    "ax.bar(r2, knn_values, width=0.25,color='g', align='center')\n",
    "ax.bar(r3, rf_values, width=0.25, color='r', align='center')\n",
    "plt.legend([\"SVM\", \"KNN\", \"RF\"])\n",
    "plt.title(\"Comparación de métricas de los distintos clasificadores\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Para comparar las salidas continuas con las verdaderas etiquetas de los datos, se crean los siguientes `DataFrame`:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "svm_df = pd.DataFrame(model_predict(final_svm,  X_test), columns=[\"negative_probability\", \"positive_probability\"])\n",
    "knn_df = pd.DataFrame(model_predict(final_knn, X_test), columns=[\"negative_probability\", \"positive_probability\"])\n",
    "rf_df = pd.DataFrame(model_predict(final_rf, X_test), columns=[\"negative_probability\", \"positive_probability\"])\n",
    "svm_df[\"word\"] = words_test\n",
    "knn_df[\"word\"] = words_test\n",
    "rf_df[\"word\"] = words_test\n",
    "svm_df[\"true_label\"] = y_test\n",
    "knn_df[\"true_label\"] = y_test\n",
    "rf_df[\"true_label\"] = y_test\n",
    "\n",
    "svm_df = svm_df[[\"word\", \"negative_probability\", \"positive_probability\", \"true_label\"]]\n",
    "knn_df = knn_df[[\"word\", \"negative_probability\", \"positive_probability\", \"true_label\"]]\n",
    "rf_df = rf_df[[\"word\", \"negative_probability\", \"positive_probability\", \"true_label\"]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "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>word</th>\n",
       "      <th>negative_probability</th>\n",
       "      <th>positive_probability</th>\n",
       "      <th>true_label</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>290</th>\n",
       "      <td>stressful</td>\n",
       "      <td>0.971489</td>\n",
       "      <td>0.028511</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>436</th>\n",
       "      <td>impending</td>\n",
       "      <td>0.994916</td>\n",
       "      <td>0.005084</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>942</th>\n",
       "      <td>incorrigibly</td>\n",
       "      <td>0.994051</td>\n",
       "      <td>0.005949</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1541</th>\n",
       "      <td>lurid</td>\n",
       "      <td>0.998807</td>\n",
       "      <td>0.001193</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1833</th>\n",
       "      <td>crashed</td>\n",
       "      <td>0.999742</td>\n",
       "      <td>0.000258</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "              word  negative_probability  positive_probability  true_label\n",
       "290      stressful              0.971489              0.028511           0\n",
       "436      impending              0.994916              0.005084           0\n",
       "942   incorrigibly              0.994051              0.005949           0\n",
       "1541         lurid              0.998807              0.001193           0\n",
       "1833       crashed              0.999742              0.000258           0"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "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>word</th>\n",
       "      <th>negative_probability</th>\n",
       "      <th>positive_probability</th>\n",
       "      <th>true_label</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>290</th>\n",
       "      <td>stressful</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>436</th>\n",
       "      <td>impending</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>942</th>\n",
       "      <td>incorrigibly</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1541</th>\n",
       "      <td>lurid</td>\n",
       "      <td>0.666667</td>\n",
       "      <td>0.333333</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1833</th>\n",
       "      <td>crashed</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "              word  negative_probability  positive_probability  true_label\n",
       "290      stressful              1.000000              0.000000           0\n",
       "436      impending              1.000000              0.000000           0\n",
       "942   incorrigibly              1.000000              0.000000           0\n",
       "1541         lurid              0.666667              0.333333           0\n",
       "1833       crashed              1.000000              0.000000           0"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "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>word</th>\n",
       "      <th>negative_probability</th>\n",
       "      <th>positive_probability</th>\n",
       "      <th>true_label</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>290</th>\n",
       "      <td>stressful</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>436</th>\n",
       "      <td>impending</td>\n",
       "      <td>0.888889</td>\n",
       "      <td>0.111111</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>942</th>\n",
       "      <td>incorrigibly</td>\n",
       "      <td>0.888889</td>\n",
       "      <td>0.111111</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1541</th>\n",
       "      <td>lurid</td>\n",
       "      <td>0.777778</td>\n",
       "      <td>0.222222</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1833</th>\n",
       "      <td>crashed</td>\n",
       "      <td>0.888889</td>\n",
       "      <td>0.111111</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "              word  negative_probability  positive_probability  true_label\n",
       "290      stressful              1.000000              0.000000           0\n",
       "436      impending              0.888889              0.111111           0\n",
       "942   incorrigibly              0.888889              0.111111           0\n",
       "1541         lurid              0.777778              0.222222           0\n",
       "1833       crashed              0.888889              0.111111           0"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "random_words = svm_df.sample(5)[\"word\"].tolist()\n",
    "\n",
    "display(svm_df[svm_df[\"word\"].isin(random_words)])\n",
    "display(knn_df[knn_df[\"word\"].isin(random_words)])\n",
    "display(rf_df[rf_df[\"word\"].isin(random_words)])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Predicción de sentimiento deoraciones\n",
    "\n",
    "En esta sección se detalla lo necesario para extender el poder de los modelos a la clasificación de oraciones de largo arbitrario.\n",
    "### Preprocesamiento de las oraciones\n",
    "\n",
    "Con el fin de eliminar las partes de las oraciones que no aportan demasiado a su significado, se realiza lo siguiente dentro de la función `tokenize_lemmatize`:\n",
    "- Se convierte la oración recibida a minúsculas.\n",
    "- Se eliminan los caracteres que no corresponden a letras (signos de puntuación y números).\n",
    "- Se separa la oración en *tokens*.\n",
    "- Se remueven las *stopwords*, palabras comunes que no contribuyen al significado de una oración.\n",
    "- Lematización, tal como se explicó en una sección anterior."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "def tokenize_lemmatize(sentence):\n",
    "    sentence = sentence.lower()\n",
    "    words_sentence = ''.join([c for c in sentence if c not in NON_WORDS])\n",
    "    tokenized_sentence = word_tokenize(words_sentence)\n",
    "    removed_stopwords = [word for word in tokenized_sentence if word not in ENGLISH_STOPWORDS]\n",
    "    lemmatized_sentence = [nltk.stem.WordNetLemmatizer().lemmatize(word) for word in removed_stopwords]\n",
    "    return lemmatized_sentence"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "La obtención de la representación vectorial es similar a la utilizada para las palabras. En el caso de las oraciones, se obtiene el vector de cada palabra en la oración y se guarda en `sentence_matrix`. De no contar con el *embedding*, se utiliza el vector de \"unk\". Posteriormente, se promedian todas las coordenadas de los vectores en `sentence_matrix` y se guarda en `mean_sentence_vector`. Este será el vector que represente a una oración."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "def vectorize_sentence(sentence):\n",
    "    sentence_matrix = np.empty((0, 300))\n",
    "    for word in sentence:\n",
    "        if word in glove_words:\n",
    "            sentence_matrix = np.append(sentence_matrix, [vec(word)], axis=0)\n",
    "        else:\n",
    "            sentence_matrix = np.append(sentence_matrix, [vec(\"unk\")], axis=0)\n",
    "    mean_sentence_vector = sentence_matrix.mean(0)\n",
    "    if np.isnan(np.sum(mean_sentence_vector)):\n",
    "        mean_sentence_vector = vec(\"unk\")\n",
    "    return mean_sentence_vector"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "La siguiente función ejecuta las dos funciones anteriores y retorna la representación vectorial de la oración entregada."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "def prepare_sentence(sentence):\n",
    "    tokenized = tokenize_lemmatize(sentence)\n",
    "    vectorized = vectorize_sentence(tokenized)\n",
    "    return vectorized"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Evaluación de los modelos sobre frases reales\n",
    "\n",
    "Para evaluar el desempeño del modelo, se utiliza el *data set* [Sentiment Labelled Sentences](https://archive.ics.uci.edu/ml/datasets/Sentiment+Labelled+Sentences). Las siguientes líneas cargan y pre-procesan estos nuevos datos."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "amazon = pd.read_csv(\"amazon_cells_labelled.txt\", sep=\"\\t\", names=[\"sentence\", \"polarity\"])\n",
    "imdb = pd.read_csv(\"imdb_labelled.txt\", sep=\"\\t\", names=[\"sentence\", \"polarity\"])\n",
    "yelp = pd.read_csv(\"yelp_labelled.txt\", sep=\"\\t\", names=[\"sentence\", \"polarity\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "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>sentence</th>\n",
       "      <th>polarity</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>So there is no way for me to plug it in here i...</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Good case, Excellent value.</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>Great for the jawbone.</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Tied to charger for conversations lasting more...</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>The mic is great.</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                            sentence  polarity\n",
       "0  So there is no way for me to plug it in here i...         0\n",
       "1                        Good case, Excellent value.         1\n",
       "2                             Great for the jawbone.         1\n",
       "3  Tied to charger for conversations lasting more...         0\n",
       "4                                  The mic is great.         1"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_sentences = pd.concat([amazon, imdb, yelp])\n",
    "all_sentences.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_sentence_test = all_sentences[\"sentence\"].tolist()\n",
    "y_sentence_test = all_sentences[\"polarity\"]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/ipykernel_launcher.py:8: RuntimeWarning: Mean of empty slice.\n",
      "  \n",
      "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/numpy/core/_methods.py:73: RuntimeWarning: invalid value encountered in true_divide\n",
      "  ret, rcount, out=ret, casting='unsafe', subok=False)\n"
     ]
    }
   ],
   "source": [
    "vectorized_sentences = list(map(prepare_sentence, X_sentence_test))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Lo anterior permite obtener las siguientes métricas: "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'accuracy': 0.7183406113537117, 'precision': 0.6550151975683891, 'recall': 0.9329004329004329, 'f1-score': 0.7696428571428572}\n",
      "{'accuracy': 0.6371906841339156, 'precision': 0.5900046274872744, 'recall': 0.9199134199134199, 'f1-score': 0.7189173949816747}\n",
      "{'accuracy': 0.665938864628821, 'precision': 0.6346375143843498, 'recall': 0.7958152958152959, 'f1-score': 0.706145966709347}\n"
     ]
    }
   ],
   "source": [
    "print(get_metrics(final_svm, vectorized_sentences, y_sentence_test))\n",
    "print(get_metrics(final_knn, vectorized_sentences, y_sentence_test))\n",
    "print(get_metrics(final_rf, vectorized_sentences, y_sentence_test))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Reentrenamiento de los modelos\n",
    "\n",
    "Se observa que el desempeño no es tan bueno como el que se tiene para la predicción de sentimiento de palabras. Por esta razón, se reentrenan los modelos anteriores con los datos para palabras y el 80% de los nuevos datos."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "new_concatenated_matrix = np.concatenate([positive_matrix, negative_matrix, vectorized_sentences], axis=0)\n",
    "new_all_labels = positive_labels + negative_labels + y_sentence_test.tolist()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [],
   "source": [
    "words_and_sentences = all_words + X_sentence_test\n",
    "new_X_train, new_X_test, new_y_train, new_y_test, new_words_train, new_words_test  = train_test_split(new_concatenated_matrix, new_all_labels, words_and_sentences, test_size=0.2, random_state=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "new_svm = SVC(C=tuning_svm.best_estimator_.C, \n",
    "                kernel=tuning_svm.best_estimator_.kernel, \n",
    "                gamma=tuning_svm.best_estimator_.gamma, probability=True)\n",
    "new_knn = KNeighborsClassifier(n_neighbors=tuning_knn.best_estimator_.n_neighbors, \n",
    "                                 algorithm=tuning_knn.best_estimator_.algorithm)\n",
    "new_rf = RandomForestClassifier(n_estimators=tuning_rf.best_estimator_.n_estimators,\n",
    "                                 criterion=tuning_rf.best_estimator_.criterion,\n",
    "                                 max_features=tuning_rf.best_estimator_.max_features)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "RandomForestClassifier(bootstrap=True, class_weight=None, criterion='entropy',\n",
       "            max_depth=None, max_features=None, max_leaf_nodes=None,\n",
       "            min_impurity_decrease=0.0, min_impurity_split=None,\n",
       "            min_samples_leaf=1, min_samples_split=2,\n",
       "            min_weight_fraction_leaf=0.0, n_estimators=9, n_jobs=1,\n",
       "            oob_score=False, random_state=None, verbose=0,\n",
       "            warm_start=False)"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "new_svm.fit(new_X_train, new_y_train)\n",
    "new_knn.fit(new_X_train, new_y_train)\n",
    "new_rf.fit(new_X_train, new_y_train)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Así, se obtienen las siguientes métricas."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'accuracy': 0.9053833605220228, 'precision': 0.8865814696485623, 'recall': 0.8434650455927052, 'f1-score': 0.8644859813084111}\n",
      "{'accuracy': 0.8504622077215879, 'precision': 0.7978227060653188, 'recall': 0.7796352583586627, 'f1-score': 0.7886241352805534}\n",
      "{'accuracy': 0.8297988036976618, 'precision': 0.7938671209540034, 'recall': 0.7082066869300911, 'f1-score': 0.7485943775100401}\n"
     ]
    }
   ],
   "source": [
    "new_svm_metrics = get_metrics(new_svm, new_X_test, new_y_test)\n",
    "new_knn_metrics = get_metrics(new_knn, new_X_test, new_y_test)\n",
    "new_rf_metrics = get_metrics(new_rf, new_X_test, new_y_test)\n",
    "print(new_svm_metrics)\n",
    "print(new_knn_metrics)\n",
    "print(new_rf_metrics)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/matplotlib/cbook/deprecation.py:107: MatplotlibDeprecationWarning: Adding an axes using the same arguments as a previous axes currently reuses the earlier instance.  In a future version, a new instance will always be created and returned.  Meanwhile, this warning can be suppressed, and the future behavior ensured, by passing a unique label to each axes instance.\n",
      "  warnings.warn(message, mplDeprecation, stacklevel=1)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "Text(0.5,1,'Comparación de modelos según distintas métricas')"
      ]
     },
     "execution_count": 47,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "r1 = np.arange(4)\n",
    "r2 = [x + 0.25 for x in r1]\n",
    "r3 = [x + 0.25 for x in r2]\n",
    "\n",
    "labels = [k for k, v in new_svm_metrics.items()]\n",
    "svm_values = [v for k, v in new_svm_metrics.items()]\n",
    "knn_values = [v for k, v in new_knn_metrics.items()]\n",
    "rf_values = [v for k, v in new_rf_metrics.items()]\n",
    "plt.xticks([r + 0.25 for r in range(4)], labels)\n",
    "ax = plt.subplot(111)\n",
    "ax.bar(r1, svm_values,width=0.25,color='b', align='center')\n",
    "ax.bar(r2, knn_values, width=0.25,color='g', align='center')\n",
    "ax.bar(r3, rf_values, width=0.25, color='r', align='center')\n",
    "plt.legend([\"SVM\", \"KNN\", \"RF\"])\n",
    "plt.title(\"Comparación de modelos según distintas métricas\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Los siguientes `DataFrames` permiten visualizar las salidas continuas y la verdadera etiqueta de los datos de *testing*."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [],
   "source": [
    "new_svm_df = pd.DataFrame(model_predict(new_svm,  new_X_test), columns=[\"negative_probability\", \"positive_probability\"])\n",
    "new_knn_df = pd.DataFrame(model_predict(new_knn, new_X_test), columns=[\"negative_probability\", \"positive_probability\"])\n",
    "new_rf_df = pd.DataFrame(model_predict(new_rf, new_X_test), columns=[\"negative_probability\", \"positive_probability\"])\n",
    "new_svm_df[\"word_sentence\"] = new_words_test\n",
    "new_knn_df[\"word_sentence\"] = new_words_test\n",
    "new_rf_df[\"word_sentence\"] = new_words_test\n",
    "new_svm_df[\"true_label\"] = new_y_test\n",
    "new_knn_df[\"true_label\"] = new_y_test\n",
    "new_rf_df[\"true_label\"] = new_y_test\n",
    "new_svm_df = new_svm_df[[\"word_sentence\", \"negative_probability\", \"positive_probability\", \"true_label\"]]\n",
    "new_knn_df = new_knn_df[[\"word_sentence\", \"negative_probability\", \"positive_probability\", \"true_label\"]]\n",
    "new_rf_df = new_rf_df[[\"word_sentence\", \"negative_probability\", \"positive_probability\", \"true_label\"]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "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>word_sentence</th>\n",
       "      <th>negative_probability</th>\n",
       "      <th>positive_probability</th>\n",
       "      <th>true_label</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>498</th>\n",
       "      <td>flexible</td>\n",
       "      <td>0.000017</td>\n",
       "      <td>0.999983</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>713</th>\n",
       "      <td>This one wants to surf on the small wave of sp...</td>\n",
       "      <td>0.701126</td>\n",
       "      <td>0.298874</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1074</th>\n",
       "      <td>oversize</td>\n",
       "      <td>0.982785</td>\n",
       "      <td>0.017215</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1287</th>\n",
       "      <td>self-satisfaction</td>\n",
       "      <td>0.507796</td>\n",
       "      <td>0.492204</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1345</th>\n",
       "      <td>dirty</td>\n",
       "      <td>0.999860</td>\n",
       "      <td>0.000140</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                          word_sentence  negative_probability  \\\n",
       "498                                            flexible              0.000017   \n",
       "713   This one wants to surf on the small wave of sp...              0.701126   \n",
       "1074                                           oversize              0.982785   \n",
       "1287                                  self-satisfaction              0.507796   \n",
       "1345                                              dirty              0.999860   \n",
       "\n",
       "      positive_probability  true_label  \n",
       "498               0.999983           1  \n",
       "713               0.298874           0  \n",
       "1074              0.017215           0  \n",
       "1287              0.492204           1  \n",
       "1345              0.000140           0  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "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>word_sentence</th>\n",
       "      <th>negative_probability</th>\n",
       "      <th>positive_probability</th>\n",
       "      <th>true_label</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>498</th>\n",
       "      <td>flexible</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>713</th>\n",
       "      <td>This one wants to surf on the small wave of sp...</td>\n",
       "      <td>0.666667</td>\n",
       "      <td>0.333333</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1074</th>\n",
       "      <td>oversize</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1287</th>\n",
       "      <td>self-satisfaction</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1345</th>\n",
       "      <td>dirty</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                          word_sentence  negative_probability  \\\n",
       "498                                            flexible              0.000000   \n",
       "713   This one wants to surf on the small wave of sp...              0.666667   \n",
       "1074                                           oversize              1.000000   \n",
       "1287                                  self-satisfaction              1.000000   \n",
       "1345                                              dirty              1.000000   \n",
       "\n",
       "      positive_probability  true_label  \n",
       "498               1.000000           1  \n",
       "713               0.333333           0  \n",
       "1074              0.000000           0  \n",
       "1287              0.000000           1  \n",
       "1345              0.000000           0  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "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>word_sentence</th>\n",
       "      <th>negative_probability</th>\n",
       "      <th>positive_probability</th>\n",
       "      <th>true_label</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>498</th>\n",
       "      <td>flexible</td>\n",
       "      <td>0.111111</td>\n",
       "      <td>0.888889</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>713</th>\n",
       "      <td>This one wants to surf on the small wave of sp...</td>\n",
       "      <td>0.222222</td>\n",
       "      <td>0.777778</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1074</th>\n",
       "      <td>oversize</td>\n",
       "      <td>0.333333</td>\n",
       "      <td>0.666667</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1287</th>\n",
       "      <td>self-satisfaction</td>\n",
       "      <td>0.777778</td>\n",
       "      <td>0.222222</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1345</th>\n",
       "      <td>dirty</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                          word_sentence  negative_probability  \\\n",
       "498                                            flexible              0.111111   \n",
       "713   This one wants to surf on the small wave of sp...              0.222222   \n",
       "1074                                           oversize              0.333333   \n",
       "1287                                  self-satisfaction              0.777778   \n",
       "1345                                              dirty              1.000000   \n",
       "\n",
       "      positive_probability  true_label  \n",
       "498               0.888889           1  \n",
       "713               0.777778           0  \n",
       "1074              0.666667           0  \n",
       "1287              0.222222           1  \n",
       "1345              0.000000           0  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "random_words_sentences = new_svm_df.sample(5)[\"word_sentence\"].tolist()\n",
    "\n",
    "display(new_svm_df[new_svm_df[\"word_sentence\"].isin(random_words_sentences)])\n",
    "display(new_knn_df[new_knn_df[\"word_sentence\"].isin(random_words_sentences)])\n",
    "display(new_rf_df[new_rf_df[\"word_sentence\"].isin(random_words_sentences)])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Finalmente, se serializan los modelos para referencias futuras."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [],
   "source": [
    "save_model(new_svm, \"sentences_SVM\")\n",
    "save_model(new_knn, \"sentences_KNN\")\n",
    "save_model(new_rf, \"sentences_RF\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Evaluación y caracterización de prejuicios\n",
    "\n",
    "En esta sección, se detallan los pasos seguidos para detectar racismo, cuantificarlo y luego intentar remediarlo. Primeramente, se obtuvo datos sobre países y gentilicios, para evaluar la existencia de prejuicios según continente. Las fuentes de estos datos son [countryinfo.py](https://gist.github.com/pamelafox/986163) y [demonyms.csv](https://github.com/knowitall/chunkedextractor/blob/master/src/main/resources/edu/knowitall/chunkedextractor/demonyms.csv), respectivamente.\n",
    "\n",
    "### Primeros acercamientos a la evaluación de prejuicios\n",
    "\n",
    "A continuación, se definen los siguientes métodos que permiten, respectivamente, obtener el nombre de un país y sus gentilicios asociados."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_continent_countries(continent):\n",
    "    return [country[\"name\"] for country in countries if continent in country[\"continent\"]]\n",
    "\n",
    "def get_continent_demonyms(demonyms, continent_countries):\n",
    "    return demonyms[demonyms[\"place\"].isin(continent_countries)][\"demonym\"].tolist()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [],
   "source": [
    "american_countries = get_continent_countries(\"America\")\n",
    "european_countries = get_continent_countries(\"Europe\")\n",
    "african_countries = get_continent_countries(\"Africa\")\n",
    "asian_countries = get_continent_countries(\"Asia\")\n",
    "oceanian_countries = get_continent_countries(\"Oceania\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [],
   "source": [
    "all_countries = american_countries + european_countries + african_countries + asian_countries + oceanian_countries"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Se carga en memoria el modelo que dio mejores resultados: SVM."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_svm = joblib.load(\"best_sentences_SVM.pkl\") "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Con la información de `demonyms.csv` se obtienen los gentilicios para cada país."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [],
   "source": [
    "demonyms = pd.read_csv(\"demonyms.csv\", header=None)\n",
    "demonyms.columns = [\"demonym\", \"place\"]\n",
    "american_demonyms = get_continent_demonyms(demonyms, american_countries)\n",
    "european_demonyms = get_continent_demonyms(demonyms, european_countries)\n",
    "african_demonyms = get_continent_demonyms(demonyms, african_countries)\n",
    "asian_demonyms = get_continent_demonyms(demonyms, asian_countries)\n",
    "oceanian_demonyms = get_continent_demonyms(demonyms, oceanian_countries)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "La siguiente función permite, dada una lista de países o nacionalidades, generar desde una oración de template una lista de oraciones para cada país o nacionalidad. Por ejemplo, al entregar los parámetros: `\"demonym\"`, `\"Let's go get {} food\"`, `american_demonyms`, `american_countries`, se retornan oraciones del tipo: \"Let's go get mexican food\"."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [],
   "source": [
    "def generate_sentence(country_attribute, sentence, continent_demonyms, continent_countries):\n",
    "    if country_attribute not in countries[0].keys():\n",
    "        return list({sentence.format(demonym) for demonym in continent_demonyms})\n",
    "    else:\n",
    "        return list({sentence.format(country) for country in continent_countries})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [],
   "source": [
    "american_food_sentence = generate_sentence(\"demonym\", \"Let's go get {} food\", american_demonyms, american_countries)\n",
    "european_food_sentence = generate_sentence(\"demonym\", \"Let's go get {} food\", european_demonyms, european_countries)\n",
    "african_food_sentence = generate_sentence(\"demonym\", \"Let's go get {} food\", african_demonyms, african_countries)\n",
    "asian_food_sentence = generate_sentence(\"demonym\", \"Let's go get {} food\", asian_demonyms, asian_countries)\n",
    "oceanian_food_sentence = generate_sentence(\"demonym\", \"Let's go get {} food\", oceanian_demonyms, oceanian_countries)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Para que el modelo pueda evaluar las oraciones anteriores, se debe obtener su representación vectorial."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [],
   "source": [
    "vectorized_food_american = list(map(prepare_sentence, american_food_sentence))\n",
    "vectorized_food_european = list(map(prepare_sentence, european_food_sentence))\n",
    "vectorized_food_african = list(map(prepare_sentence, african_food_sentence))\n",
    "vectorized_food_asian = list(map(prepare_sentence, asian_food_sentence))\n",
    "vectorized_food_oceanian = list(map(prepare_sentence, oceanian_food_sentence))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Con los *embeddings*, se procede a guardar en `DataFrames` las predicciones del modelo para las frases según continente."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [],
   "source": [
    "american_food_predictions = pd.DataFrame(model_predict(test_svm, vectorized_food_american), columns=[\"negative_probability\", \"positive_probability\"])\n",
    "european_food_predictions = pd.DataFrame(model_predict(test_svm, vectorized_food_european), columns=[\"negative_probability\", \"positive_probability\"])\n",
    "african_food_predictions = pd.DataFrame(model_predict(test_svm, vectorized_food_african), columns=[\"negative_probability\", \"positive_probability\"])\n",
    "asian_food_predictions = pd.DataFrame(model_predict(test_svm, vectorized_food_asian), columns=[\"negative_probability\", \"positive_probability\"])\n",
    "oceanian_food_predictions = pd.DataFrame(model_predict(test_svm, vectorized_food_oceanian), columns=[\"negative_probability\", \"positive_probability\"])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Posteriormente, se calcula el promedio para las predicciones."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [],
   "source": [
    "american_food_predictions_mean = american_food_predictions.mean().tolist()\n",
    "european_food_predictions_mean = european_food_predictions.mean().tolist()\n",
    "african_food_predictions_mean = african_food_predictions.mean().tolist()\n",
    "asian_food_predictions_mean = asian_food_predictions.mean().tolist()\n",
    "oceanian_food_predictions_mean = oceanian_food_predictions.mean().tolist()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "La información anterior permite construir un diccionario con los promedios antes calculados."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [],
   "source": [
    "continents_food_mean = {\n",
    "    \"América\": american_food_predictions_mean,\n",
    "    \"Europa\": european_food_predictions_mean,\n",
    "    \"África\": african_food_predictions_mean,\n",
    "    \"Asia\": asian_food_predictions_mean,\n",
    "    \"Oceanía\": oceanian_food_predictions_mean,\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'América': [0.6111782797899482, 0.38882172021005174], 'Europa': [0.5735651383883401, 0.42643486161165983], 'África': [0.6324345195026476, 0.36756548049735266], 'Asia': [0.6196386003447514, 0.3803613996552484], 'Oceanía': [0.5964573393570225, 0.40354266064297756]}\n"
     ]
    }
   ],
   "source": [
    "print(continents_food_mean)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Se preparan los datos para graficar el promedio de sentimiento negativo según continente."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [],
   "source": [
    "negative_food_means = {k: v[0] for k, v in continents_food_mean.items()}\n",
    "positive_food_means = {k: v[1] for k, v in continents_food_mean.items()}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0,0.5,'Probabilidad')"
      ]
     },
     "execution_count": 64,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.bar(negative_food_means.keys(), negative_food_means.values())\n",
    "plt.title(\"Probabilidad promedio de sentimiento negativo para la frase 'Let's go get {gentilicio} food' según continente\")\n",
    "plt.xlabel(\"Continente\")\n",
    "plt.ylabel(\"Probabilidad\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "El gráfico permite visualizar que, en promedio, la frase \"Let's go get {gentilicio} food\", arroja una probabilidad de ser negativa más alta en el caso de los países africanos. Esto permite conjeturar que podría existir un sesgo racista. De forma similar a lo expuesto anteriormente, se grafica el sentimiento positivo promedio."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0,0.5,'Probabilidad')"
      ]
     },
     "execution_count": 65,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.bar(positive_food_means.keys(), positive_food_means.values())\n",
    "plt.title(\"Probabilidad promedio de sentimiento positivo para la frase 'Let's go get {gentilicio} food' según continente\")\n",
    "plt.xlabel(\"Continente\")\n",
    "plt.ylabel(\"Probabilidad\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "En las celdas siguientes se quiso evaluar el sentimiento promedio por continente para \"Let's go to {país}\"."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [],
   "source": [
    "american_places_sentence = generate_sentence(\"name\", \"Let's go to {}\", american_demonyms, american_countries)\n",
    "european_places_sentence = generate_sentence(\"name\", \"Let's go to {}\", european_demonyms, european_countries)\n",
    "african_places_sentence = generate_sentence(\"name\", \"Let's go to {}\", african_demonyms, african_countries)\n",
    "asian_places_sentence = generate_sentence(\"name\", \"Let's go to {}\", asian_demonyms, asian_countries)\n",
    "oceanian_places_sentence = generate_sentence(\"name\", \"Let's go to {}\", oceanian_demonyms, oceanian_countries)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [],
   "source": [
    "vectorized_place_american = list(map(prepare_sentence, american_places_sentence))\n",
    "vectorized_place_european = list(map(prepare_sentence, european_places_sentence))\n",
    "vectorized_place_african = list(map(prepare_sentence, african_places_sentence))\n",
    "vectorized_place_asian = list(map(prepare_sentence, asian_places_sentence))\n",
    "vectorized_place_oceanian = list(map(prepare_sentence, oceanian_places_sentence))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [],
   "source": [
    "american_places_predictions = pd.DataFrame(model_predict(test_svm, vectorized_place_american), columns=[\"negative_probability\", \"positive_probability\"])\n",
    "european_places_predictions = pd.DataFrame(model_predict(test_svm, vectorized_place_european), columns=[\"negative_probability\", \"positive_probability\"])\n",
    "african_places_predictions = pd.DataFrame(model_predict(test_svm, vectorized_place_african), columns=[\"negative_probability\", \"positive_probability\"])\n",
    "asian_places_predictions = pd.DataFrame(model_predict(test_svm, vectorized_place_asian), columns=[\"negative_probability\", \"positive_probability\"])\n",
    "oceanian_places_predictions = pd.DataFrame(model_predict(test_svm, vectorized_place_oceanian), columns=[\"negative_probability\", \"positive_probability\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [],
   "source": [
    "american_places_predictions_mean = american_places_predictions.mean().tolist()\n",
    "european_places_predictions_mean = european_places_predictions.mean().tolist()\n",
    "african_places_predictions_mean = african_places_predictions.mean().tolist()\n",
    "asian_places_predictions_mean = asian_places_predictions.mean().tolist()\n",
    "oceanian_places_predictions_mean = oceanian_places_predictions.mean().tolist()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [],
   "source": [
    "continents_places_mean = {\n",
    "    \"América\": american_places_predictions_mean,\n",
    "    \"Europa\": european_places_predictions_mean,\n",
    "    \"África\": african_places_predictions_mean,\n",
    "    \"Asia\": asian_places_predictions_mean,\n",
    "    \"Oceanía\": oceanian_places_predictions_mean,\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {},
   "outputs": [],
   "source": [
    "negative_places_means = {k: v[0] for k, v in continents_places_mean.items()}\n",
    "positive_places_means = {k: v[1] for k, v in continents_places_mean.items()}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5,1,\"Probabilidad promedio de sentimiento negativo para la frase 'Let's go to {país}' según continente\")"
      ]
     },
     "execution_count": 72,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.bar(negative_places_means.keys(), negative_places_means.values())\n",
    "plt.xlabel(\"Continente\")\n",
    "plt.ylabel(\"Probabilidad\")\n",
    "plt.title(\"Probabilidad promedio de sentimiento negativo para la frase 'Let's go to {país}' según continente\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "El gráfico permite distinguir que el sentimiento negativo hacia las visitas a países de África es bastante más marcado que el del resto de los continentes. Ahora sí es posible afirmar que existe el sesgo mencionado anteriormente."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5,1,\"Probabilidad promedio de sentimiento positivo para la frase 'Let's go to {país}' según continente\")"
      ]
     },
     "execution_count": 73,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.bar(positive_places_means.keys(), positive_places_means.values())\n",
    "plt.xlabel(\"Continente\")\n",
    "plt.ylabel(\"Probabilidad\")\n",
    "plt.title(\"Probabilidad promedio de sentimiento positivo para la frase 'Let's go to {país}' según continente\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Definición de una métrica para el prejuicio racista\n",
    "\n",
    "En esta sección se detallan los pasos seguidos para definir una métrica para el prejuicio racista. En primer lugar, se realiza una normalización de los gentilicios, a través del proceso de lematización."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "metadata": {},
   "outputs": [],
   "source": [
    "american_demonyms_lemmatized = {nltk.stem.WordNetLemmatizer().lemmatize(word.lower()) for word in american_demonyms}\n",
    "european_demonyms_lemmatized = {nltk.stem.WordNetLemmatizer().lemmatize(word.lower()) for word in european_demonyms}\n",
    "african_demonyms_lemmatized = {nltk.stem.WordNetLemmatizer().lemmatize(word.lower()) for word in african_demonyms}\n",
    "asian_demonyms_lemmatized = {nltk.stem.WordNetLemmatizer().lemmatize(word.lower()) for word in asian_demonyms}\n",
    "oceanian_demonyms_lemmatized = {nltk.stem.WordNetLemmatizer().lemmatize(word.lower()) for word in oceanian_demonyms}"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Los gráficos anteriores permitieron dar cuenta de que en general los países europeos tienen asociado un sentimiento positivo mayor al del resto de los continentes. Por esta razón, se confeccionaron vectores a partir de una matriz cuyas filas correspondían a países y nacionalidades europeas. Posteriormente, se calculó el mínimo, el máximo y el promedio por dimensión de todos los vectores que conformaban la matriz. Así, se escogió como estándar de comparación el vector construido a partir del promedio por dimensión (`european_mean_vector`), pues es el que ofrece mayor neutralidad (la diferencia entre el sentimiento positivo y negativo asociado es mennor que en el caso del mínimo y del máximo). "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[0.007280 0.992720]]\n",
      "[[0.816033 0.183967]]\n",
      "[[0.241519 0.758481]]\n"
     ]
    }
   ],
   "source": [
    "european_demonym_matrix = np.empty((0, 300))\n",
    "for word in european_demonyms_lemmatized:\n",
    "    if word in glove_words:\n",
    "        european_demonym_matrix = np.append(european_demonym_matrix, [vec(word)], axis=0)\n",
    "    else:\n",
    "        continue\n",
    "for word in european_countries:\n",
    "    if word.lower() in glove_words:\n",
    "        european_demonym_matrix = np.append(european_demonym_matrix, [vec(word.lower())], axis=0)\n",
    "    else:\n",
    "        continue\n",
    "european_min_vector = european_demonym_matrix.min(0)\n",
    "european_max_vector = european_demonym_matrix.max(0)\n",
    "european_mean_vector = european_demonym_matrix.mean(0)\n",
    "european_min_prediction = model_predict(test_svm, [european_min_vector])\n",
    "european_max_prediction = model_predict(test_svm, [european_max_vector])\n",
    "european_mean_prediction = model_predict(test_svm, [european_mean_vector])\n",
    "\n",
    "print(european_min_prediction)\n",
    "print(european_max_prediction)\n",
    "print(european_mean_prediction)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Lo anterior permite definir la métrica de prejuicio de la siguiente forma: la diferencia absoluta entre la probabilidad de sentimiento negativo del vector promedio europeo y el vector al cual se le quiere medir el prejuicio."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_prejudge_metric(vector, model):\n",
    "    return abs(european_mean_prediction[0][1] - model_predict(model, [vector])[0][1])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Las siguientes tres celdas permiten obtener el promedio de la métrica de prejuicio por continente y visualizar estos valores en un gráfico."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 77,
   "metadata": {},
   "outputs": [],
   "source": [
    "demonym_american_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence, american_demonyms_lemmatized)]\n",
    "demonym_european_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence, european_demonyms_lemmatized)]\n",
    "demonym_african_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence, african_demonyms_lemmatized)]\n",
    "demonym_asian_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence, asian_demonyms_lemmatized)]\n",
    "demonym_oceanian_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence, oceanian_demonyms_lemmatized)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "metadata": {},
   "outputs": [],
   "source": [
    "demonym_prejudge = {\n",
    "    \"América\": np.array(demonym_american_prejudge).mean(),\n",
    "    \"Europa\": np.array(demonym_european_prejudge).mean(),\n",
    "    \"África\": np.array(demonym_african_prejudge).mean(),\n",
    "    \"Asia\": np.array(demonym_asian_prejudge).mean(),\n",
    "    \"Oceanía\": np.array(demonym_oceanian_prejudge).mean(),\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5,1,'Métrica de prejuicio para gentilicios según continente')"
      ]
     },
     "execution_count": 79,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.bar(demonym_prejudge.keys(), demonym_prejudge.values())\n",
    "plt.xlabel(\"Continente\")\n",
    "plt.ylabel(\"Métrica de prejuicio\")\n",
    "plt.title(\"Métrica de prejuicio para gentilicios según continente\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Se evidencia que la métrica de prejuicio definida presenta un valor más alto para los gentilicios del continente africano, lo que coincide con los análisis expuestos anteriormente. Las siguientes tres celdas muestran el mismo análisis, pero para nombres de países por continente."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 80,
   "metadata": {},
   "outputs": [],
   "source": [
    "american_countries_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence, american_countries)]\n",
    "european_countries_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence, european_countries)]\n",
    "african_countries_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence, african_countries)]\n",
    "asian_countries_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence, asian_countries)]\n",
    "oceanian_countries_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence, oceanian_countries)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 81,
   "metadata": {},
   "outputs": [],
   "source": [
    "countries_prejudge = {\n",
    "    \"América\": np.array(american_countries_prejudge).mean(),\n",
    "    \"Europa\": np.array(european_countries_prejudge).mean(),\n",
    "    \"África\": np.array(african_countries_prejudge).mean(),\n",
    "    \"Asia\": np.array(asian_countries_prejudge).mean(),\n",
    "    \"Oceanía\": np.array(oceanian_countries_prejudge).mean(),\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 82,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5,1,'Métrica de prejuicio para países según continente')"
      ]
     },
     "execution_count": 82,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.bar(countries_prejudge.keys(), countries_prejudge.values())\n",
    "plt.xlabel(\"Continente\")\n",
    "plt.ylabel(\"Métrica de prejuicio\")\n",
    "plt.title(\"Métrica de prejuicio para países según continente\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Los gráficos y análisis anteriores motivan a buscar una forma de reducir el prejuicio racista de los modelos entregados. Por esta razón, se define la función `prejudge_free_vectorize`. Su funcionamiento es similar a la función `vectorize_sentence` anterior, la diferencia radica en que si alguna palabra en la frase corresponde a un gentilicio o a un nombre de país, el *embedding* asociado a ella ahora será el vector definido como `european_mean_vector`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {},
   "outputs": [],
   "source": [
    "all_demonyms = demonyms[\"demonym\"].tolist()\n",
    "all_demonyms_lemmatized = {nltk.stem.WordNetLemmatizer().lemmatize(word.lower()) for word in all_demonyms}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 84,
   "metadata": {},
   "outputs": [],
   "source": [
    "def prejudge_free_vectorize(sentence):\n",
    "    sentence_matrix = np.empty((0, 300))\n",
    "    for word in sentence:\n",
    "        if word in all_demonyms_lemmatized.union(set(map(lambda x: x.lower(), all_countries))):\n",
    "            sentence_matrix = np.append(sentence_matrix, [european_mean_vector], axis=0)\n",
    "            continue\n",
    "        if word in glove_words:\n",
    "            sentence_matrix = np.append(sentence_matrix, [vec(word)], axis=0)\n",
    "        else:\n",
    "            sentence_matrix = np.append(sentence_matrix, [vec(\"unk\")], axis=0)\n",
    "    mean_sentence_vector = sentence_matrix.mean(0)\n",
    "    if np.isnan(np.sum(mean_sentence_vector)):\n",
    "        mean_sentence_vector = vec(\"unk\")\n",
    "    return mean_sentence_vector"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 85,
   "metadata": {},
   "outputs": [],
   "source": [
    "def prepare_sentence_prejudge_free(sentence):\n",
    "    tokenized = tokenize_lemmatize(sentence)\n",
    "    vectorized = prejudge_free_vectorize(tokenized)\n",
    "    return vectorized"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Las tres celdas anteriores permiten calcular el nuevo valor de la métrica de prejuicio para los gentilicios según continente."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 86,
   "metadata": {},
   "outputs": [],
   "source": [
    "new_demonym_american_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence_prejudge_free, american_demonyms_lemmatized)]\n",
    "new_demonym_european_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence_prejudge_free, european_demonyms_lemmatized)]\n",
    "new_demonym_african_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence_prejudge_free, african_demonyms_lemmatized)]\n",
    "new_demonym_asian_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence_prejudge_free, asian_demonyms_lemmatized)]\n",
    "new_demonym_oceanian_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence_prejudge_free, oceanian_demonyms_lemmatized)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 87,
   "metadata": {},
   "outputs": [],
   "source": [
    "new_prejudge_demonym = {\n",
    "    \"América\": np.array(new_demonym_american_prejudge).mean(),\n",
    "    \"Europa\": np.array(new_demonym_european_prejudge).mean(),\n",
    "    \"África\": np.array(new_demonym_african_prejudge).mean(),\n",
    "    \"Asia\": np.array(new_demonym_asian_prejudge).mean(),\n",
    "    \"Oceanía\": np.array(new_demonym_oceanian_prejudge).mean(),\n",
    "}"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "El siguiente gráfico permite comparar los valores antiguos y los obtenidos tras realizar el cambio en la *vectorización*."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5,1,'Comparación de métrica de prejuicio para gentilicio según continente')"
      ]
     },
     "execution_count": 88,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.bar(new_prejudge_demonym.keys(), new_prejudge_demonym.values(), color=\"green\")\n",
    "plt.bar(demonym_prejudge.keys(), demonym_prejudge.values(), color=\"red\", alpha=0.2)\n",
    "plt.legend([\"Nuevo valor de métrica\", \"Valor anterior de métrica\"])\n",
    "plt.xlabel(\"Continente\")\n",
    "plt.ylabel(\"Métrica de prejuicio\")\n",
    "plt.title(\"Comparación de métrica de prejuicio para gentilicio según continente\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Así, efectivamente hay una reducción del prejuicio (al menos en la forma en que fue cuantificado en este informe) tras el cambio en los *embeddings* asociados a gentilicios y países. Ahora, quiere evaluarse los valores de prejuicio para la frase \"Let's go get {gentilicio} food\"."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 89,
   "metadata": {},
   "outputs": [],
   "source": [
    "food_american_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence, american_food_sentence)]\n",
    "food_european_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence, european_food_sentence)]\n",
    "food_african_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence, african_food_sentence)]\n",
    "food_asian_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence, asian_food_sentence)]\n",
    "food_oceanian_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence, oceanian_food_sentence)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 90,
   "metadata": {},
   "outputs": [],
   "source": [
    "food_prejudge = {\n",
    "    \"América\": np.array(food_american_prejudge).mean(),\n",
    "    \"Europa\": np.array(food_european_prejudge).mean(),\n",
    "    \"África\": np.array(food_african_prejudge).mean(),\n",
    "    \"Asia\": np.array(food_asian_prejudge).mean(),\n",
    "    \"Oceanía\": np.array(food_oceanian_prejudge).mean(),\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 91,
   "metadata": {},
   "outputs": [],
   "source": [
    "new_food_american_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                         map(prepare_sentence_prejudge_free, american_food_sentence)]\n",
    "new_food_european_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence_prejudge_free, european_food_sentence)]\n",
    "new_food_african_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence_prejudge_free, african_food_sentence)]\n",
    "new_food_asian_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence_prejudge_free, asian_food_sentence)]\n",
    "new_food_oceanian_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence_prejudge_free, oceanian_food_sentence)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 92,
   "metadata": {},
   "outputs": [],
   "source": [
    "new_food_prejudge = {\n",
    "    \"América\": np.array(new_food_american_prejudge).mean(),\n",
    "    \"Europa\": np.array(new_food_european_prejudge).mean(),\n",
    "    \"África\": np.array(new_food_african_prejudge).mean(),\n",
    "    \"Asia\": np.array(new_food_asian_prejudge).mean(),\n",
    "    \"Oceanía\": np.array(new_food_oceanian_prejudge).mean(),\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 93,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5,1,\"Comparación de métrica de prejuicio para 'Let's go get {} food' según continente\")"
      ]
     },
     "execution_count": 93,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.bar(new_food_prejudge.keys(), new_food_prejudge.values(), color=\"green\")\n",
    "plt.bar(food_prejudge.keys(), food_prejudge.values(), color=\"red\", alpha=0.2)\n",
    "plt.legend([\"Nuevo valor de métrica\", \"Valor anterior de métrica\"])\n",
    "plt.xlabel(\"Continente\")\n",
    "plt.ylabel(\"Métrica de prejuicio\")\n",
    "plt.title(\"Comparación de métrica de prejuicio para 'Let's go get {} food' según continente\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "El gráfico indica que también hubo una reducción del prejuicio medido, pero la diferencia no es tan notoria como en el caso de los prejuicios por sí solos, como en el gráfico anterior. Finalmente, se quiso cuantificar el prejuicio para \"Let's go to {país}\"."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "metadata": {},
   "outputs": [],
   "source": [
    "american_places_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence, american_places_sentence)]\n",
    "european_places_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence, european_places_sentence)]\n",
    "african_places_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence, african_places_sentence)]\n",
    "asian_places_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence, asian_places_sentence)]\n",
    "oceanian_places_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence, oceanian_places_sentence)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {},
   "outputs": [],
   "source": [
    "places_prejudge = {\n",
    "    \"América\": np.array(american_places_prejudge).mean(),\n",
    "    \"Europa\": np.array(european_places_prejudge).mean(),\n",
    "    \"África\": np.array(african_places_prejudge).mean(),\n",
    "    \"Asia\": np.array(asian_places_prejudge).mean(),\n",
    "    \"Oceanía\": np.array(oceanian_places_prejudge).mean(),\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 96,
   "metadata": {},
   "outputs": [],
   "source": [
    "new_american_places_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence_prejudge_free, american_places_sentence)]\n",
    "new_european_places_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence_prejudge_free, european_places_sentence)]\n",
    "new_african_places_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence_prejudge_free, african_places_sentence)]\n",
    "new_asian_places_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence_prejudge_free, asian_places_sentence)]\n",
    "new_oceanian_places_prejudge = [get_prejudge_metric(vector, test_svm) for vector in \n",
    "                     map(prepare_sentence_prejudge_free, oceanian_places_sentence)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 97,
   "metadata": {},
   "outputs": [],
   "source": [
    "new_places_prejudge = {\n",
    "    \"América\": np.array(new_american_places_prejudge).mean(),\n",
    "    \"Europa\": np.array(new_european_places_prejudge).mean(),\n",
    "    \"África\": np.array(new_african_places_prejudge).mean(),\n",
    "    \"Asia\": np.array(new_asian_places_prejudge).mean(),\n",
    "    \"Oceanía\": np.array(new_oceanian_places_prejudge).mean(),\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 98,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5,1,\"Comparación de métrica de prejuicio para 'Let's go to {}' según continente\")"
      ]
     },
     "execution_count": 98,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.bar(new_places_prejudge.keys(), new_places_prejudge.values(), color=\"green\")\n",
    "plt.bar(places_prejudge.keys(), places_prejudge.values(), color=\"red\", alpha=0.2)\n",
    "plt.legend([\"Nuevo valor de métrica\", \"Valor anterior de métrica\"])\n",
    "plt.xlabel(\"Continente\")\n",
    "plt.ylabel(\"Métrica de prejuicio\")\n",
    "plt.title(\"Comparación de métrica de prejuicio para 'Let's go to {}' según continente\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "El gráfico permite observar que el prejuicio medido se redujo de forma considerable en el caso de África, que era el continente más perjudicado. También se observan disminuciones para América, Asia y Oceanía, aunque menos marcadas. Se observa además un pequeño aumento en el caso de Europa. Esto es positivo, puesto que el nuevo valor es más similar al del resto de los continentes que el anterior. Finalmente, este era el objetivo a perseguir: que el sentimiento asociado a países y sus gentilicios fuera lo más neutro posible. Los valores más uniformes y disminuidos de la métrica de prejuicio en comparación a los primeros *outputs* del modelo, permiten afirmar que lo expuesto en este informe es un primer paso para no propagar (o, al menos, mitigar) terribles sesgos humanos en el área de aprendizaje de máquina."
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}