contextual_bandit_sim.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "import numpy as np\n",
    "from matplotlib import pylab as plt\n",
    "#from mpltools import style # uncomment for prettier plots\n",
    "#style.use(['ggplot'])\n",
    "\n",
    "import OnlineVariance\n",
    "import DataGenerator\n",
    "import BinaryStrategy\n",
    "import PositiveStrategy\n",
    "import Simulator"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 96,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "arms=3\n",
    "features=5\n",
    "rewardType='positive'\n",
    "#rewardType='binary'\n",
    "featureType='integer'\n",
    "#featureType='binary'\n",
    "\n",
    "# define number of samples and number of choices\n",
    "num_samples = 10000\n",
    "num_batches = 100\n",
    "num_experiments = 3\n",
    "\n",
    "dg = DataGenerator.DataGenerator(arms,features,feature_type=featureType,reward_type=rewardType)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 97,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "experiment: 0\n",
      "experiment: 1\n",
      "experiment: 2\n"
     ]
    }
   ],
   "source": [
    "total_regret = []\n",
    "total_rmse = []\n",
    "\n",
    "for e in range(0,num_experiments):\n",
    "    print \"experiment: %d\" % e\n",
    "    \n",
    "    #appropriate model for reward_type='positive' or reward_type='mixed'\n",
    "    positiveStrategy = PositiveStrategy.PositiveStrategy(arms,features)\n",
    "    simulator = Simulator.Simulator(positiveStrategy)\n",
    "\n",
    "    previous_rmse = 0.\n",
    "    if previous_rmse == 0:\n",
    "        initial_rmse = rmse[0][-1]\n",
    "    \n",
    "    for b in range(0,num_batches):\n",
    "        (sample_features, sample_rewards) = dg.generate_samples(num_samples)    \n",
    "        regret, rmse = simulator.simulate(sample_features, sample_rewards, dg.W)\n",
    "        \n",
    "        if previous_rmse == 0:\n",
    "            initial_rmse = rmse[0][-1]\n",
    "            previous_rmse = rmse[0][-1]\n",
    "        #print \"\\tbatch: %d, started at: %f, now: %f\" % (b, abs(initial_rmse), abs(rmse[0][-1]))\n",
    "        if (len(total_rmse) == 0):\n",
    "            total_rmse = rmse\n",
    "            total_regret = regret\n",
    "        else:\n",
    "            total_rmse += rmse\n",
    "            total_regret += regret\n",
    "#appropriate model for reward_type='binary'\n",
    "#binaryStrategy = BinaryStrategy.BinaryStrategy(arms,features)\n",
    "#regret,delta = binaryStrategy.simulate(sample_features,sample_rewards,dg.W)\n",
    "\n",
    "mean_regret = total_regret / num_experiments\n",
    "mean_rmse   = total_rmse   / num_experiments\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 93,
   "metadata": {
    "collapsed": false,
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "#print dg.W\n",
    "#print positiveStrategy.mu()\n",
    "#print sample_rewards\n",
    "#print np.cumsum(total_regret / num_experiments)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAY4AAAEZCAYAAACAZ8KHAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x10ce6a290>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#regret_accumulator = np.cumsum(sample_regret)\n",
    "#np.cumsum(sample_regret)\n",
    "plt.semilogy(np.cumsum(mean_regret)/num_experiments)\n",
    "plt.title('Simulated Bandit Performance for K = ' + str(arms))\n",
    "plt.ylabel('Cumulative Expected Regret')\n",
    "plt.xlabel('Round Index')\n",
    "#plt.legend(('Random','Naive','Epsilon-Greedy','(1 - 1/t) UCB','95% UCB',\"thompson\", \"thomson normal\"),loc='lower right')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAY4AAAEeCAYAAACdYvI/AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nnr8X4AAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x10ce6ebd0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#regret_accumulator = np.cumsum(sample_regret)\n",
    "#np.cumsum(sample_regret)\n",
    "plt.semilogx(mean_rmse/num_experiments)\n",
    "plt.title('Simulated Bandit Performance for K = ' + str(arms))\n",
    "plt.ylabel('RMSE')\n",
    "plt.xlabel('Round Index')\n",
    "#plt.legend(('Random','Naive','Epsilon-Greedy','(1 - 1/t) UCB','95% UCB',\"thompson\", \"thomson normal\"),loc='lower right')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "plt.semilogy(regret_accumulator/number_experiments)\n",
    "plt.title('Simulated Bandit Performance for K = 5')\n",
    "plt.ylabel('Cumulative Expected Regret')\n",
    "plt.xlabel('Round Index')\n",
    "#plt.legend(('Random','Naive','Epsilon-Greedy','(1 - 1/t) UCB','95% UCB',\"thompson\", \"thomson normal\"),loc='lower right')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Running experiment: 1\n",
      "Running experiment: 2\n"
     ]
    }
   ],
   "source": [
    "regret_accumulator = np.zeros((num_samples,1))\n",
    "for i in range(num_experiments):\n",
    "    print \"Running experiment:\", i+1\n",
    "    true_rewards,CTRs_that_generated_data = dg.generate_bernoulli_bandit_data(num_samples)\n",
    "    regret_accumulator[:,0] += dg.run_bandit_alg(true_rewards,CTRs_that_generated_data,dg.thompson_sampling)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'CTRs_that_generated_data' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-8-d641f0b81392>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mCTRs_that_generated_data\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m: name 'CTRs_that_generated_data' is not defined"
     ]
    }
   ],
   "source": [
    "CTRs_that_generated_data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "stats = np.empty((arms,features),dtype=object)\n",
    "for k in range(0,arms):\n",
    "    for d in range(0,features):\n",
    "        stats[k,d] = OnlineVariance.OnlineVariance(ddof=0)\n",
    "F = np.array([0,0,1,1,0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([-0.        , -0.        ,  1.20563   , -1.25494569,  0.        ])"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "F * map(lambda x: np.random.normal(x.mean, x.std if x.std > 0 else 1), stats[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}