Federated Learning Example.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from Node import Node\n",
    "from util import fetch_minibatch\n",
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "from time import time\n",
    "import matplotlib.pyplot as plt\n",
    "plt.style.use(\"ggplot\")\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn import datasets\n",
    "iris = datasets.load_iris()\n",
    "X_full = iris.data\n",
    "y_full = np.eye(3)[iris.target]\n",
    "n_features = X_full[0].shape[0]\n",
    "n_classes = y_full[0].shape[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "untrained accuracy:  0.333333\n",
      "accuracy, using updated weights 0.667 \n"
     ]
    }
   ],
   "source": [
    "server_batch_size = 30\n",
    "\n",
    "X_test, y_test = fetch_minibatch(X_full, y_full, server_batch_size)\n",
    "\n",
    "with tf.Session() as sess:    \n",
    "    Master = Node(sess, n_features, n_classes)\n",
    "    sess.run(tf.global_variables_initializer())\n",
    "        \n",
    "    print(\"untrained accuracy: \", Master.score(X_full, y_full))\n",
    "    \n",
    "    n_epochs = 20\n",
    "    X_server, y_server = fetch_minibatch(X_full, y_full, server_batch_size)\n",
    "    Master.fit(X_server, y_server, n_epochs)\n",
    "    \n",
    "    print(\"accuracy, using updated weights {0:.3f} \".format(Master.score(X_full, y_full)))\n",
    "    \n",
    "    server_w1 = Master.fetch_weights()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "original_server_w1 = server_w1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "round 0 execution time: 3.1388421058654785 accuracy: 0.8999999761581421\n",
      "round 1 execution time: 2.8116440773010254 accuracy: 0.95333331823349\n",
      "round 2 execution time: 2.782147169113159 accuracy: 0.9599999785423279\n",
      "round 3 execution time: 2.7059502601623535 accuracy: 0.9599999785423279\n",
      "round 4 execution time: 2.756247043609619 accuracy: 0.9666666388511658\n",
      "round 5 execution time: 2.7820329666137695 accuracy: 0.9399999976158142\n",
      "round 6 execution time: 2.8027403354644775 accuracy: 0.95333331823349\n",
      "round 7 execution time: 2.7404510974884033 accuracy: 0.6666666865348816\n",
      "round 8 execution time: 2.8029439449310303 accuracy: 0.95333331823349\n",
      "round 9 execution time: 2.7011499404907227 accuracy: 0.6666666865348816\n",
      "completed\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x11cc4d978>]"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAX0AAAD8CAYAAACb4nSYAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x11c9ad358>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learning_rate = .1\n",
    "n_epochs = 500\n",
    "n_clients = 5\n",
    "client_batch_size = 5\n",
    "n_rounds = 10\n",
    "ls_accuracy = []\n",
    "\n",
    "with tf.Session() as sess:\n",
    "    Worker = Node(sess, n_features, n_classes)\n",
    "    sess.run(tf.global_variables_initializer())\n",
    "    \n",
    "    for r in range(n_rounds):\n",
    "        s = time()\n",
    "        \n",
    "        # each round needs a new average client weights\n",
    "        ls_client_w1 = []\n",
    "        \n",
    "        for client in range(n_clients):\n",
    "\n",
    "            # each client receives the same server weight\n",
    "            client_w1 = server_w1\n",
    "\n",
    "            # each client has their own minibatch\n",
    "            X_client, y_client = fetch_minibatch(X_full, y_full, client_batch_size)\n",
    "            \n",
    "            # calculate term b gradient using (1) server weights and (2) client data\n",
    "            term_b = Worker.fetch_grad(server_w1, X_client, y_client)\n",
    "            \n",
    "            # compute term c gradient using (1) server weights and (2) server data\n",
    "            term_c = Worker.fetch_grad(server_w1, X_server, y_server)\n",
    "\n",
    "            # client parameter updates\n",
    "            for n in range(n_epochs):\n",
    "                # compute term a gradient using (1) client weights and (2) client data\n",
    "                term_a = Worker.fetch_grad(client_w1, X_client, y_client)\n",
    "\n",
    "                # update client weights\n",
    "                client_w1 = client_w1 - learning_rate * (term_a - term_b + term_c)\n",
    "\n",
    "            # track weights for each client\n",
    "            ls_client_w1.append(client_w1)\n",
    "        \n",
    "        # after all clients are accounted for, update server weights each round\n",
    "        server_w1 = np.mean(ls_client_w1, axis=0)\n",
    "        Worker.load(server_w1)\n",
    "        acc = Worker.score(X_full, y_full)\n",
    "        print(\"round {} execution time: {} accuracy: {}\".format(r, time() - s, acc))\n",
    "        ls_accuracy.append(acc)\n",
    "    \n",
    "print(\"completed\")\n",
    "plt.plot(ls_accuracy)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "with tf.Session() as sess:    \n",
    "    Master = Client(sess, n_features, n_classes)\n",
    "    sess.run(tf.global_variables_initializer())\n",
    "    \n",
    "    Master.load(original_server_w1)\n",
    "    print(\"untrained accuracy: \", Master.score(X_test, y_test))\n",
    "    Master.load(server_w1)  # use average weights from clients\n",
    "    print(\"accuracy, using updated weights: \", Master.score(X_test, y_test))"
   ]
  }
 ],
 "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": "3.6.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}