examples/quickstart/online-bandit-vs-opl-simulation.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "source": [
    "# On-policy bandit vs Greedy off-policy optimised policies\n",
    "This notebook shows an empirical experiment comparing selection of bandit algorithms with off-policy optimised policies.\n",
    "\n",
    "We present two experiments that try to answer the following research questions\n",
    "1. What is the effect of the number of rounds (training data) on the difference in reward of on-policy vs off-policy?\n",
    "2. What is the difference between reward between different types of bandit algorithms when used as logged data for an Inverse Propensity Weighting (IPW) learner.\n",
    "\n",
    "Both experiments in this notebook have the following setup.\n",
    "- The environment is stationary\n",
    "- Feedback is instant (no delay)\n",
    "- Bandit algorithms keep exploring / learning during the evaluation cycle\n",
    "- Propensities for the IPW are learned by fitting a classifier predicting $P(A=a|c)$ on the training or evaluation dataset. Note that we fit a separate propensity model for the training and evaluation data and predict on the same dataset as it was trained on."
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%% md\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "outputs": [],
   "source": [
    "import copy\n",
    "\n",
    "import numpy as np\n",
    "from sklearn.linear_model import LogisticRegression\n",
    "\n",
    "from obp.policy.offline import IPWLearner\n",
    "from obp.dataset import logistic_sparse_reward_function\n",
    "from obp.policy import Random, EpsilonGreedy, BernoulliTS, LinTS\n",
    "\n",
    "from sklearn.ensemble import RandomForestClassifier as RandomForest\n",
    "\n",
    "from obp.simulator.simulator import BanditEnvironmentSimulator, BanditPolicySimulator\n"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "### Off-policy vs on-policy at different number of rounds\n",
    "In this section, we show how various bandit algorithms can be compared to an IPW learning given the amount of training and evaluation rounds."
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%% md\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 1000/1000 [00:00<00:00, 71107.98it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 62524.10it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 41594.07it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 41387.62it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 3733.33it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 2261.46it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 72191.12it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 74326.24it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 18544.59it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 27509.59it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 29930.24it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 21766.09it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4331.40it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4737.45it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 74643.70it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 75466.98it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 26941.14it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 28371.14it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 23014.52it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 35251.29it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4133.55it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4631.43it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 75249.00it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 71325.64it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 50579.49it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 62065.20it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 16328.38it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 35920.90it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 3539.61it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4786.43it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 79674.49it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 84617.17it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 25811.10it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 90927.51it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 42778.94it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 50236.00it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4131.76it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4882.14it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 76728.82it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 79668.43it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 35534.11it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 17749.61it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 45804.35it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 33578.34it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4020.59it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4633.56it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 61207.48it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 56076.58it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 54733.78it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 51546.71it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 36036.02it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 28116.67it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 3781.21it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4924.77it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 85845.06it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 84175.64it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 30837.52it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 34989.56it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 19203.81it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 25137.12it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4065.59it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4776.98it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 91107.24it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 80443.11it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 46294.24it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 47454.39it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 36234.32it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 27203.41it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4133.00it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4806.86it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 83536.90it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 80001.22it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 69798.04it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 56420.55it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 24037.78it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 34633.04it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 2292.01it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4601.36it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 74041.52it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 85829.25it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 35902.76it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 50591.69it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 31273.70it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 23204.62it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 3895.19it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 5207.01it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 96365.40it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 100181.62it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 34358.70it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 23197.04it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 26888.29it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 41850.13it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4318.20it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 5368.52it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 105152.03it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 102425.01it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 51172.51it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 24376.70it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 22060.54it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 26079.76it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4278.90it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 5329.52it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 92652.90it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 97703.28it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 40528.59it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 18429.13it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 35920.90it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 30259.10it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4201.83it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 5292.04it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 95758.18it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 94752.27it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 41173.91it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 36722.56it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 13265.22it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 26905.54it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4689.13it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 5330.69it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 94160.92it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 106657.44it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 24212.48it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 48870.99it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 58569.03it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 43915.73it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 3704.40it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 5232.85it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 92346.90it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 92998.03it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 49045.86it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 29521.14it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 28941.00it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 43935.06it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4531.74it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 5275.77it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 78080.05it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 79428.55it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 24876.95it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 66942.85it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 9167.00it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 26286.69it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4863.95it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 5183.44it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 82711.58it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 90564.29it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 29106.29it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 20519.17it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 17148.24it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 61720.88it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4514.00it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 5297.54it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 94984.01it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 99334.60it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 52319.58it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 42889.18it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 32028.59it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 33229.05it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4136.73it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 5302.98it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 105673.93it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 99947.67it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 41588.37it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 83669.21it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 63242.76it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 62533.98it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5235.78it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5218.14it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 114962.52it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 110670.86it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 42194.99it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 77898.64it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 63686.89it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 62580.44it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5338.83it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5382.39it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 114187.89it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 114518.68it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 46767.43it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 85419.01it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 64737.90it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 64525.67it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5431.24it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5351.08it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 113584.90it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 120437.03it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 37788.22it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 85580.40it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 66150.27it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 65314.75it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5340.48it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5281.06it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 110338.67it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 104751.54it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 40686.12it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 87049.82it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 63905.12it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 64238.48it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5273.75it/s]\n",
      "100%|██████████| 10000/10000 [00:02<00:00, 4117.27it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 103564.85it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 114510.87it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 37788.91it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 81710.14it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 58567.56it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 61200.25it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5288.68it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5230.10it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 107634.57it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 101431.25it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 43174.11it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 72039.85it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 59916.49it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 61308.31it/s]\n",
      "100%|██████████| 10000/10000 [00:02<00:00, 4996.90it/s]\n",
      "100%|██████████| 10000/10000 [00:02<00:00, 4991.65it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 108472.44it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 108634.55it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 52500.86it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 64569.77it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 63328.32it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 60316.11it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5228.31it/s]\n",
      "100%|██████████| 10000/10000 [00:02<00:00, 4869.45it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 108129.04it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 88605.38it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 46349.54it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 77567.44it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 61436.37it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 62741.55it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5360.50it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5248.47it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 114992.46it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 117937.12it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 39768.05it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 83591.67it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 62480.88it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 61806.19it/s]\n",
      "100%|██████████| 10000/10000 [00:02<00:00, 4800.24it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5176.17it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 110405.77it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 116860.65it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 42126.33it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 75767.99it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 62039.77it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 63640.89it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5268.36it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5321.78it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 119850.95it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 110988.02it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 46005.96it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 77145.14it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 59916.57it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 60829.15it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5242.65it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5246.00it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 115238.90it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 108199.34it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 44603.27it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 77523.14it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 64042.61it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 62469.53it/s]\n",
      "100%|██████████| 10000/10000 [00:02<00:00, 4843.53it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5258.26it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 109979.63it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 116189.89it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 35496.88it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 84844.83it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 61452.66it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 61858.97it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5266.60it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5326.52it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 107583.50it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 115835.95it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 38256.28it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 82661.69it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 58550.39it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 61353.15it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5223.12it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5241.39it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 103692.61it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 108274.19it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 46193.03it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 73881.88it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 63770.75it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 64335.14it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5212.76it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5250.88it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 121867.92it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 113397.88it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 60568.70it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 69276.61it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 64596.52it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 64564.80it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5331.46it/s]\n",
      "100%|██████████| 10000/10000 [00:02<00:00, 4959.83it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 111483.63it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 121331.37it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 41648.09it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 83804.12it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 59838.27it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 61995.11it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5384.63it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5331.86it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 112103.70it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 120111.11it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 40536.35it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 85497.89it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 63655.67it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 64796.11it/s]\n",
      "100%|██████████| 10000/10000 [00:02<00:00, 4635.08it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5254.46it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 113579.36it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 117464.47it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 41121.80it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 84685.00it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 64443.28it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 63973.74it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5292.28it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5396.45it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 109452.41it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 102721.00it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 80844.72it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 86888.15it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 65295.28it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 66037.73it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5224.77it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5230.21it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 120718.71it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 123657.28it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 67091.04it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 86110.77it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 66535.06it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 64994.19it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5217.92it/s]\n",
      "100%|██████████| 100000/100000 [00:18<00:00, 5300.88it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 105755.30it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 117501.79it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 82205.89it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 86265.65it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 57269.44it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 67035.49it/s]\n",
      "100%|██████████| 100000/100000 [00:18<00:00, 5278.93it/s]\n",
      "100%|██████████| 100000/100000 [00:18<00:00, 5312.67it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 123252.73it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 124270.67it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 78416.85it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 85422.42it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 66924.90it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 64341.29it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5233.95it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5220.02it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 125678.81it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 120258.95it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 81906.59it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 87087.07it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 65777.61it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 65798.42it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5064.07it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5217.12it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 120520.68it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 122089.25it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 79017.02it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 80694.93it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 64330.64it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 63133.15it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5025.01it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5132.86it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 121774.80it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 123144.68it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 83586.81it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 85931.68it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 64137.43it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 64637.27it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5186.42it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5222.09it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 116659.00it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 121473.33it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 71388.18it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 84924.19it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 65890.71it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 64362.28it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5213.84it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5245.70it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 102729.91it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 118182.33it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 80849.42it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 73515.20it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 64309.83it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 66069.27it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5199.00it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5258.08it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 124933.47it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 97743.40it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 81551.11it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 86576.17it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 63869.49it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 66821.03it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5240.42it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5227.93it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 125935.56it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 124014.71it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 82247.19it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 86833.07it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 56871.05it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 66530.95it/s]\n",
      "100%|██████████| 100000/100000 [00:18<00:00, 5286.47it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5241.78it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 122634.26it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 123095.85it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 80225.01it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 86266.54it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 66644.85it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 56181.04it/s]\n",
      "100%|██████████| 100000/100000 [00:18<00:00, 5272.49it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5189.57it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 123020.61it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 121131.25it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 81712.48it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 87205.78it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 65694.95it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 66348.28it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5222.50it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5213.11it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 126041.45it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 125530.83it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 81303.54it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 86919.50it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 64838.28it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 65648.11it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5213.12it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5200.08it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 119641.93it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 124819.29it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 80129.44it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 86704.98it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 65270.92it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 64907.71it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5193.13it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5167.46it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 120544.51it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 122220.99it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 68141.15it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 83382.43it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 66138.40it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 65400.94it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5109.01it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5063.46it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 119566.73it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 121223.15it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 81883.97it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 86030.83it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 64462.23it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 65545.02it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5210.54it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5189.36it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 120772.35it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 123340.62it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 80505.45it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 86540.54it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 60354.03it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 64152.74it/s]\n",
      "100%|██████████| 100000/100000 [00:18<00:00, 5276.95it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5223.78it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 122448.20it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 120764.17it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 82463.28it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 85601.80it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 65378.11it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 66430.62it/s]\n",
      "100%|██████████| 100000/100000 [00:18<00:00, 5272.22it/s]\n",
      "100%|██████████| 100000/100000 [00:18<00:00, 5295.28it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 124919.56it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 123839.90it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 73319.75it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 84553.09it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 66251.40it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 65569.27it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5213.16it/s]\n",
      "100%|██████████| 100000/100000 [00:19<00:00, 5261.67it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 96357.96it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 119114.67it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 86347.70it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 83673.58it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 63763.41it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 66017.87it/s]\n",
      "100%|██████████| 500000/500000 [01:36<00:00, 5193.01it/s]\n",
      "100%|██████████| 500000/500000 [01:36<00:00, 5187.82it/s]\n",
      "100%|██████████| 500000/500000 [00:03<00:00, 125294.85it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 120853.69it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 78214.02it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 85901.81it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 63749.70it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 63055.69it/s]\n",
      "100%|██████████| 500000/500000 [01:36<00:00, 5205.26it/s]\n",
      "100%|██████████| 500000/500000 [01:35<00:00, 5217.87it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 114329.86it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 120642.37it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 84003.06it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 85279.51it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 63566.12it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 63047.32it/s]\n",
      "100%|██████████| 500000/500000 [01:37<00:00, 5153.69it/s]\n",
      "100%|██████████| 500000/500000 [01:37<00:00, 5141.90it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 124015.02it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 119479.80it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 86609.69it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 85828.10it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 63950.54it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 63293.34it/s]\n",
      "100%|██████████| 500000/500000 [01:36<00:00, 5201.99it/s]\n",
      "100%|██████████| 500000/500000 [01:36<00:00, 5197.81it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 115560.26it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 124222.30it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 82774.64it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 76918.40it/s]\n",
      "100%|██████████| 500000/500000 [00:08<00:00, 62322.69it/s]\n",
      "100%|██████████| 500000/500000 [00:08<00:00, 60655.53it/s]\n",
      "100%|██████████| 500000/500000 [01:36<00:00, 5188.68it/s]\n",
      "100%|██████████| 500000/500000 [01:35<00:00, 5214.03it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 124153.30it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 118002.15it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 83227.78it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 86791.78it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 63841.54it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 64071.31it/s]\n",
      "100%|██████████| 500000/500000 [01:35<00:00, 5225.73it/s]\n",
      "100%|██████████| 500000/500000 [01:35<00:00, 5222.09it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 117669.98it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 122589.78it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 84414.43it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 85719.21it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 63491.90it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 63775.89it/s]\n",
      "100%|██████████| 500000/500000 [01:35<00:00, 5214.39it/s]\n",
      "100%|██████████| 500000/500000 [01:36<00:00, 5197.34it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 123150.50it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 119012.06it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 80131.52it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 86045.21it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 63762.87it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 62982.10it/s]\n",
      "100%|██████████| 500000/500000 [01:36<00:00, 5193.69it/s]\n",
      "100%|██████████| 500000/500000 [01:36<00:00, 5188.46it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 123920.31it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 117779.85it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 84557.29it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 85360.75it/s]\n",
      "100%|██████████| 500000/500000 [00:08<00:00, 62488.41it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 65258.49it/s]\n",
      "100%|██████████| 500000/500000 [01:36<00:00, 5190.03it/s]\n",
      "100%|██████████| 500000/500000 [01:36<00:00, 5203.37it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 122933.39it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 123991.34it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 83471.76it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 86043.81it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 63363.62it/s]\n",
      "100%|██████████| 500000/500000 [00:08<00:00, 62089.17it/s]\n",
      "100%|██████████| 500000/500000 [01:37<00:00, 5153.87it/s]\n",
      "100%|██████████| 500000/500000 [01:36<00:00, 5159.88it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 122697.90it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 116719.49it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 86560.68it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 84334.33it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 63030.73it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 64963.13it/s]\n",
      "100%|██████████| 500000/500000 [01:36<00:00, 5174.19it/s]\n",
      "100%|██████████| 500000/500000 [01:37<00:00, 5136.84it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 122605.64it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 117194.11it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 81181.06it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 86635.91it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 63666.51it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 63358.27it/s]\n",
      "100%|██████████| 500000/500000 [01:35<00:00, 5223.64it/s]\n",
      "100%|██████████| 500000/500000 [01:36<00:00, 5207.36it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 123051.72it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 115655.01it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 85207.55it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 86229.04it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 63358.46it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 64591.02it/s]\n",
      "100%|██████████| 500000/500000 [01:36<00:00, 5181.24it/s]\n",
      "100%|██████████| 500000/500000 [01:36<00:00, 5192.40it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 123472.46it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 124574.33it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 82328.19it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 86207.70it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 63752.76it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 63432.97it/s]\n",
      "100%|██████████| 500000/500000 [01:36<00:00, 5207.02it/s]\n",
      "100%|██████████| 500000/500000 [01:36<00:00, 5197.79it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 123436.43it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 116687.25it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 86110.51it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 83839.67it/s]\n",
      "100%|██████████| 500000/500000 [00:08<00:00, 61873.42it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 65480.83it/s]\n",
      "100%|██████████| 500000/500000 [01:36<00:00, 5208.03it/s]\n",
      "100%|██████████| 500000/500000 [01:35<00:00, 5210.08it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 123395.74it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 122616.07it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 86638.44it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 82815.12it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 64711.65it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 63626.23it/s]\n",
      "100%|██████████| 500000/500000 [01:37<00:00, 5127.15it/s]\n",
      "100%|██████████| 500000/500000 [01:36<00:00, 5166.52it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 117759.63it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 123414.70it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 86155.08it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 86556.98it/s]\n",
      "100%|██████████| 500000/500000 [00:08<00:00, 60705.89it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 63587.92it/s]\n",
      "100%|██████████| 500000/500000 [01:35<00:00, 5254.50it/s]\n",
      "100%|██████████| 500000/500000 [01:35<00:00, 5228.20it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 119981.65it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 119961.84it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 83975.27it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 85731.13it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 63449.90it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 65156.38it/s]\n",
      "100%|██████████| 500000/500000 [01:36<00:00, 5171.00it/s]\n",
      "100%|██████████| 500000/500000 [01:37<00:00, 5147.19it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 123398.47it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 121213.83it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 82114.49it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 86144.43it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 63861.96it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 62968.21it/s]\n",
      "100%|██████████| 500000/500000 [01:36<00:00, 5157.77it/s]\n",
      "100%|██████████| 500000/500000 [01:37<00:00, 5150.31it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 121825.44it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 115650.98it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 85680.30it/s]\n",
      "100%|██████████| 500000/500000 [00:05<00:00, 84211.76it/s]\n",
      "100%|██████████| 500000/500000 [00:08<00:00, 62214.68it/s]\n",
      "100%|██████████| 500000/500000 [00:07<00:00, 64961.56it/s]\n",
      "100%|██████████| 500000/500000 [01:36<00:00, 5173.64it/s]\n",
      "100%|██████████| 500000/500000 [01:36<00:00, 5158.55it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 122484.18it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 122330.71it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 85010.72it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 85102.85it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64025.00it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 63448.40it/s]\n",
      "100%|██████████| 1000000/1000000 [03:13<00:00, 5170.31it/s]\n",
      "100%|██████████| 1000000/1000000 [03:12<00:00, 5188.26it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 121399.00it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 119281.24it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 85746.22it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 84953.13it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64576.23it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 63743.67it/s]\n",
      "100%|██████████| 1000000/1000000 [03:10<00:00, 5239.70it/s]\n",
      "100%|██████████| 1000000/1000000 [03:12<00:00, 5201.21it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 123398.80it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 120849.98it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 85776.79it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 85223.47it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64334.39it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 63635.26it/s]\n",
      "100%|██████████| 1000000/1000000 [03:12<00:00, 5203.59it/s]\n",
      "100%|██████████| 1000000/1000000 [03:12<00:00, 5198.03it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 124294.97it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 122749.07it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 83821.57it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 86877.32it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64028.46it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64814.22it/s]\n",
      "100%|██████████| 1000000/1000000 [03:11<00:00, 5219.39it/s]\n",
      "100%|██████████| 1000000/1000000 [03:12<00:00, 5201.59it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 114605.02it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 113468.56it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 85045.24it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 85551.26it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64149.11it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64946.41it/s]\n",
      "100%|██████████| 1000000/1000000 [03:11<00:00, 5220.71it/s]\n",
      "100%|██████████| 1000000/1000000 [03:10<00:00, 5239.00it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 122734.59it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 121020.44it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 84742.46it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 85639.51it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64750.54it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 63984.68it/s]\n",
      "100%|██████████| 1000000/1000000 [03:12<00:00, 5191.73it/s]\n",
      "100%|██████████| 1000000/1000000 [03:12<00:00, 5186.10it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 120266.04it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 118935.58it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 84842.79it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 85774.46it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64560.75it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64030.11it/s]\n",
      "100%|██████████| 1000000/1000000 [03:13<00:00, 5177.71it/s]\n",
      "100%|██████████| 1000000/1000000 [03:12<00:00, 5186.08it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 121147.42it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 120445.66it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 85077.01it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 85527.35it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64372.79it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64068.61it/s]\n",
      "100%|██████████| 1000000/1000000 [03:11<00:00, 5212.12it/s]\n",
      "100%|██████████| 1000000/1000000 [03:11<00:00, 5229.48it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 123945.15it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 121813.86it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 83913.58it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 86620.55it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 63773.92it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64122.13it/s]\n",
      "100%|██████████| 1000000/1000000 [03:13<00:00, 5157.11it/s]\n",
      "100%|██████████| 1000000/1000000 [03:12<00:00, 5188.56it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 124746.22it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 121423.46it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 85295.45it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 84712.78it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64167.18it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 65040.19it/s]\n",
      "100%|██████████| 1000000/1000000 [03:11<00:00, 5212.20it/s]\n",
      "100%|██████████| 1000000/1000000 [03:12<00:00, 5187.24it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 120167.91it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 124485.34it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 87153.96it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 83805.99it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 65210.94it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 65020.84it/s]\n",
      "100%|██████████| 1000000/1000000 [03:11<00:00, 5213.58it/s]\n",
      "100%|██████████| 1000000/1000000 [03:11<00:00, 5217.41it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 122360.26it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 120748.44it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 84827.33it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 85416.54it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 65073.45it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64193.66it/s]\n",
      "100%|██████████| 1000000/1000000 [03:10<00:00, 5236.48it/s]\n",
      "100%|██████████| 1000000/1000000 [03:11<00:00, 5225.93it/s]\n",
      "100%|██████████| 1000000/1000000 [00:07<00:00, 125185.96it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 122298.41it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 85957.72it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 84979.07it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64478.98it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64960.46it/s]\n",
      "100%|██████████| 1000000/1000000 [03:12<00:00, 5195.05it/s]\n",
      "100%|██████████| 1000000/1000000 [03:11<00:00, 5212.72it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 122441.67it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 122384.63it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 86309.05it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 84199.23it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64715.42it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64551.44it/s]\n",
      "100%|██████████| 1000000/1000000 [03:12<00:00, 5190.31it/s]\n",
      "100%|██████████| 1000000/1000000 [03:13<00:00, 5174.18it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 123104.09it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 121177.74it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 85958.16it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 87128.16it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 63790.04it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 63598.23it/s]\n",
      "100%|██████████| 1000000/1000000 [03:12<00:00, 5208.13it/s]\n",
      "100%|██████████| 1000000/1000000 [03:11<00:00, 5215.34it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 120054.32it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 124759.46it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 86829.84it/s]\n",
      "100%|██████████| 1000000/1000000 [00:12<00:00, 83278.55it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64746.13it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64468.20it/s]\n",
      "100%|██████████| 1000000/1000000 [03:12<00:00, 5194.95it/s]\n",
      "100%|██████████| 1000000/1000000 [03:12<00:00, 5206.09it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 124457.94it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 120729.72it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 84320.42it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 86543.14it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 63987.29it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64921.31it/s]\n",
      "100%|██████████| 1000000/1000000 [03:12<00:00, 5199.88it/s]\n",
      "100%|██████████| 1000000/1000000 [03:13<00:00, 5175.28it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 120467.13it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 124672.05it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 86992.54it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 84002.60it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64882.16it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64709.99it/s]\n",
      "100%|██████████| 1000000/1000000 [03:11<00:00, 5210.94it/s]\n",
      "100%|██████████| 1000000/1000000 [03:12<00:00, 5199.03it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 123731.46it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 121437.10it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 85154.88it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 84966.66it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 63147.67it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64871.97it/s]\n",
      "100%|██████████| 1000000/1000000 [03:11<00:00, 5209.33it/s]\n",
      "100%|██████████| 1000000/1000000 [03:11<00:00, 5226.36it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 122616.98it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 121568.31it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 85571.63it/s]\n",
      "100%|██████████| 1000000/1000000 [00:11<00:00, 85540.01it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64839.87it/s]\n",
      "100%|██████████| 1000000/1000000 [00:15<00:00, 64033.11it/s]\n",
      "100%|██████████| 1000000/1000000 [03:11<00:00, 5209.70it/s]\n",
      "100%|██████████| 1000000/1000000 [03:12<00:00, 5191.02it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 121686.26it/s]\n",
      "100%|██████████| 1000000/1000000 [00:08<00:00, 123562.29it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 4h 51min 1s, sys: 16min 43s, total: 5h 7min 44s\n",
      "Wall time: 4h 41min 52s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "experiment_round_size = [1000, 10000, 100000, 500000, 1000000]\n",
    "n_runs_per_round_size = 20\n",
    "n_actions = 5\n",
    "dim_context = 5\n",
    "\n",
    "bandit_policies = [\n",
    "    (Random, {\"n_actions\": n_actions, \"epsilon\": 1.0, \"random_state\": 12345}),\n",
    "    (BernoulliTS, {\"n_actions\": n_actions, \"random_state\": 12345}),\n",
    "    (LinTS, {\"dim\": dim_context, \"n_actions\": n_actions, \"random_state\": 12345}),\n",
    "    (EpsilonGreedy, {\"n_actions\": n_actions, \"epsilon\": 0.1, \"random_state\": 12345}),\n",
    "]\n",
    "\n",
    "dummy_policies = [policy_class(**args) for policy_class, args in bandit_policies]\n",
    "train_rewards = {policy.policy_name: [] for policy in dummy_policies}\n",
    "eval_rewards = {**copy.deepcopy(train_rewards), **{IPWLearner.__name__: []}}\n",
    "train_rewards[\"n_rounds\"] = []\n",
    "eval_rewards[\"n_rounds\"] = []\n",
    "\n",
    "# In this setting the dataset is static, so we can use the same dataset class and redraw batches\n",
    "# for different settings. When drift or delay is added the dataset object needs to be reinitiated with a new seed\n",
    "# for each experiment.\n",
    "env = BanditEnvironmentSimulator(\n",
    "    n_actions=10,\n",
    "    dim_context=5,\n",
    "    reward_type=\"binary\", # \"binary\" or \"continuous\"\n",
    "    reward_function=logistic_sparse_reward_function,\n",
    "    random_state=12345,\n",
    ")\n",
    "\n",
    "for n_rounds in experiment_round_size:\n",
    "    for experiment in range(n_runs_per_round_size):\n",
    "        training_bandit_batch = env.next_bandit_round_batch(n_rounds=n_rounds)\n",
    "        evaluation_bandit_batch = env.next_bandit_round_batch(n_rounds=n_rounds)\n",
    "\n",
    "        # Train the bandit algorithms and get the rewards for the training and evaluation period\n",
    "        for policy_class, args in bandit_policies:\n",
    "            policy = policy_class(**args)\n",
    "\n",
    "            training_simulator = BanditPolicySimulator(policy=policy)\n",
    "\n",
    "            training_simulator.steps(batch_bandit_rounds=training_bandit_batch)\n",
    "            train_rewards[policy.policy_name].append(training_simulator.total_reward)\n",
    "\n",
    "            eval_simulator = BanditPolicySimulator(policy=policy)\n",
    "\n",
    "            eval_simulator.steps(batch_bandit_rounds=evaluation_bandit_batch)\n",
    "            eval_rewards[policy.policy_name].append(eval_simulator.total_reward)\n",
    "\n",
    "        # Train a propensity model on the actions in the training period to get propensities per round\n",
    "        propensity_model = LogisticRegression(random_state=12345)\n",
    "        propensity_model.fit(training_simulator.contexts, training_simulator.selected_actions)\n",
    "        pscores = propensity_model.predict_proba(training_simulator.contexts)\n",
    "\n",
    "        # Train an IPW learning from the logged data and learned propensities\n",
    "        ipw_learner = IPWLearner(n_actions=env.n_actions,\n",
    "                                 base_classifier=RandomForest(n_estimators=30, min_samples_leaf=10, random_state=12345))\n",
    "\n",
    "        ipw_learner.fit(\n",
    "            context=training_simulator.contexts,\n",
    "            action=training_simulator.selected_actions,\n",
    "            reward=training_simulator.obtained_rewards,\n",
    "            pscore=np.choose(training_simulator.selected_actions, pscores.T)\n",
    "        )\n",
    "        eval_action_dists = ipw_learner.predict(\n",
    "            context=eval_simulator.contexts\n",
    "        )\n",
    "\n",
    "        eval_rewards[ipw_learner.policy_name].append(\n",
    "            np.sum(eval_action_dists.squeeze(axis=2) * evaluation_bandit_batch.rewards)\n",
    "        )\n",
    "\n",
    "        train_rewards[\"n_rounds\"].append(n_rounds)\n",
    "        eval_rewards[\"n_rounds\"].append(n_rounds)\n",
    "\n",
    "\n"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "outputs": [],
   "source": [
    "import seaborn as sns\n",
    "import pandas as pd"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "The plot below compares the off-policy model to different on-policy bandit algorithms on the evaluation dataset.\n",
    "\n",
    "IPWLearning has been allowed to learn from n_rounds of samples from the logged training data, the other policies have been exploring on n_rounds and can now exploit (and learn) on the evaluation data.\n",
    "\n",
    "The plot compares the average reward per model. We clearly see that after $10^4$ IPWLearning start outperforming all on-policy models reaching an average reward of around .65 vs the highest on-policy reward of .56. Additionally, the on-policy models do not seem to increase their performance after $10^4$ rounds.\n",
    "\n",
    "This seems to indicate that using an off-policy method to optimise your bandit model can yield higher performance than continious learning with on-policy models."
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%% md\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "outputs": [
    {
     "data": {
      "text/plain": "<Figure size 432x288 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "def plot_average_reward_per_n_rounds(rewards):\n",
    "    rewards_pd = pd.DataFrame(rewards)\n",
    "    rewards_pd = pd.melt(rewards_pd, ['n_rounds'])\n",
    "    rewards_pd[\"average reward\"] = rewards_pd[\"value\"] / rewards_pd[\"n_rounds\"]\n",
    "\n",
    "    plot = sns.lineplot(data=rewards_pd, x=\"n_rounds\", y=\"average reward\", style=\"variable\", hue=\"variable\", markers=True, dashes=False)\n",
    "    plot.set(xscale='log')\n",
    "    plot.legend(bbox_to_anchor=(1.1, 1.05))\n",
    "\n",
    "plot_average_reward_per_n_rounds(eval_rewards)\n",
    "\n"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "### Off-policy on different kinds of bandit policy logs\n",
    "Understanding how counterfactual examples affect the performance of various off-policy optimisers can be useful in deciding which strategy for data collection to take.\n",
    "\n",
    "In this next section, we run a bandit algorithm in order to collect logs that can be used for off-policy learning. These logs are used to train a propensity model for fitting a IPW model to be used for evaluation."
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%% md\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 1000/1000 [00:00<00:00, 81137.15it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 58612.41it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4992.29it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 51322.17it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 26147.07it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 25413.71it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4780.59it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 96032.24it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 21132.66it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 54953.93it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 5120.64it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 110802.13it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 34457.78it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 24740.19it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 5211.13it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 97981.73it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 32533.15it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 26400.52it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4592.51it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 105873.99it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 29863.96it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 21680.70it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4555.45it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 89556.82it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 18315.42it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 24406.92it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 5049.20it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 94760.83it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 38659.32it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 23811.25it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4817.58it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 101500.47it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 12318.65it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 26050.45it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4989.70it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 84322.87it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 18287.87it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 20186.86it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4877.02it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 93756.80it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 20838.57it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 26484.54it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4651.21it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 103532.39it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 18442.66it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 21052.36it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 5285.63it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 107400.30it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 40202.67it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 28427.67it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4411.49it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 97705.55it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 23679.24it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 32725.81it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4830.78it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 94904.49it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 22103.44it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 30856.35it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 5314.96it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 110521.84it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 49627.92it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 21243.97it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4730.21it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 97304.35it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 29210.89it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 38296.80it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 5036.99it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 73947.53it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 19698.13it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 54353.60it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 5143.30it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 101699.82it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 23034.11it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 26207.52it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4030.16it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 100522.56it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 36952.27it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 35522.67it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 4778.77it/s]\n",
      "100%|██████████| 1000/1000 [00:00<00:00, 71017.68it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 38964.95it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 57169.98it/s]\n",
      "100%|██████████| 10000/10000 [00:02<00:00, 4799.38it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 109797.10it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 63253.63it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 63504.93it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5176.11it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 88393.07it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 37115.40it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 65228.61it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5278.39it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 115294.96it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 37645.10it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 61608.64it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5277.01it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 126602.92it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 39218.42it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 67739.22it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5480.38it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 127344.78it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 33739.78it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 67663.49it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5472.06it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 125279.02it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 42106.16it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 67781.80it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5491.17it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 121995.89it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 36170.83it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 67539.28it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5501.65it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 130203.70it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 41154.64it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 67276.68it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5507.95it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 113839.23it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 41471.06it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 65751.85it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5193.88it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 122622.54it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 42233.82it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 67590.43it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5463.87it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 116817.04it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 40171.09it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 65286.08it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5355.93it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 125303.72it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 37740.42it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 64486.98it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5248.05it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 122733.77it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 39806.05it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 64557.85it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5340.55it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 114991.51it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 32428.49it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 56239.52it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5270.60it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 97931.63it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 47859.31it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 64210.95it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5331.61it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 116148.38it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 43719.49it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 65433.96it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5170.83it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 117028.90it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 38670.22it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 64752.29it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5213.54it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 120283.34it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 40270.91it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 63540.14it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5359.26it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 87205.11it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 40721.75it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 60849.45it/s]\n",
      "100%|██████████| 10000/10000 [00:01<00:00, 5289.94it/s]\n",
      "100%|██████████| 10000/10000 [00:00<00:00, 118154.40it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 81671.92it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 63886.00it/s]\n",
      "100%|██████████| 100000/100000 [00:18<00:00, 5411.73it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 120091.27it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 82985.64it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 64953.80it/s]\n",
      "100%|██████████| 100000/100000 [00:18<00:00, 5368.55it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 122553.82it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 82725.10it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 66641.92it/s]\n",
      "100%|██████████| 100000/100000 [00:20<00:00, 4788.02it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 94266.16it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 66730.55it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 52196.40it/s]\n",
      "100%|██████████| 100000/100000 [00:22<00:00, 4468.17it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 104965.71it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 70921.90it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 58006.25it/s]\n",
      "100%|██████████| 100000/100000 [00:22<00:00, 4530.24it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 104562.06it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 71722.69it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 55597.72it/s]\n",
      "100%|██████████| 100000/100000 [00:21<00:00, 4580.62it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 123793.74it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 77736.75it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 54608.08it/s]\n",
      "100%|██████████| 100000/100000 [00:21<00:00, 4630.63it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 105365.54it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 72593.75it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 56916.53it/s]\n",
      "100%|██████████| 100000/100000 [00:21<00:00, 4591.81it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 104449.02it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 70947.79it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 60421.39it/s]\n",
      "100%|██████████| 100000/100000 [00:21<00:00, 4606.93it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 104141.50it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 71139.14it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 57183.15it/s]\n",
      "100%|██████████| 100000/100000 [00:21<00:00, 4600.02it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 111226.96it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 70982.04it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 54671.28it/s]\n",
      "100%|██████████| 100000/100000 [00:21<00:00, 4623.27it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 102536.43it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 74249.86it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 60799.22it/s]\n",
      "100%|██████████| 100000/100000 [00:21<00:00, 4586.65it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 123245.45it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 72235.59it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 55358.91it/s]\n",
      "100%|██████████| 100000/100000 [00:21<00:00, 4598.01it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 102618.52it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 69808.90it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 54965.10it/s]\n",
      "100%|██████████| 100000/100000 [00:21<00:00, 4593.71it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 104275.07it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 70504.26it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 56087.07it/s]\n",
      "100%|██████████| 100000/100000 [00:21<00:00, 4624.63it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 86509.16it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 60635.43it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 55871.17it/s]\n",
      "100%|██████████| 100000/100000 [00:21<00:00, 4566.98it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 103736.47it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 63213.80it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 56435.48it/s]\n",
      "100%|██████████| 100000/100000 [00:21<00:00, 4598.55it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 122619.96it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 70996.76it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 55064.74it/s]\n",
      "100%|██████████| 100000/100000 [00:21<00:00, 4609.54it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 104412.46it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 72994.31it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 59049.93it/s]\n",
      "100%|██████████| 100000/100000 [00:21<00:00, 4653.38it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 103824.16it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 62885.29it/s]\n",
      "100%|██████████| 100000/100000 [00:01<00:00, 55022.63it/s]\n",
      "100%|██████████| 100000/100000 [00:22<00:00, 4521.83it/s]\n",
      "100%|██████████| 100000/100000 [00:00<00:00, 102893.87it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 75165.63it/s]\n",
      "100%|██████████| 500000/500000 [00:08<00:00, 56968.06it/s]\n",
      "100%|██████████| 500000/500000 [01:49<00:00, 4556.04it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 103507.02it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 72688.56it/s]\n",
      "100%|██████████| 500000/500000 [00:08<00:00, 57039.22it/s]\n",
      "100%|██████████| 500000/500000 [01:49<00:00, 4583.16it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 102943.12it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 73901.21it/s]\n",
      "100%|██████████| 500000/500000 [00:08<00:00, 56486.14it/s]\n",
      "100%|██████████| 500000/500000 [01:50<00:00, 4539.42it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 103971.01it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 74103.23it/s]\n",
      "100%|██████████| 500000/500000 [00:09<00:00, 55335.50it/s]\n",
      "100%|██████████| 500000/500000 [01:49<00:00, 4552.91it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 104465.42it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 74315.36it/s]\n",
      "100%|██████████| 500000/500000 [00:08<00:00, 57102.72it/s]\n",
      "100%|██████████| 500000/500000 [01:49<00:00, 4573.79it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 102044.40it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 72657.50it/s]\n",
      "100%|██████████| 500000/500000 [00:08<00:00, 56878.59it/s]\n",
      "100%|██████████| 500000/500000 [01:48<00:00, 4600.44it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 107524.22it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 73132.16it/s]\n",
      "100%|██████████| 500000/500000 [00:08<00:00, 56370.29it/s]\n",
      "100%|██████████| 500000/500000 [01:48<00:00, 4609.48it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 102035.15it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 74741.58it/s]\n",
      "100%|██████████| 500000/500000 [00:08<00:00, 55914.64it/s]\n",
      "100%|██████████| 500000/500000 [01:48<00:00, 4603.14it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 106117.98it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 71783.35it/s]\n",
      "100%|██████████| 500000/500000 [00:08<00:00, 57585.73it/s]\n",
      "100%|██████████| 500000/500000 [01:49<00:00, 4568.40it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 104112.05it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 72929.65it/s]\n",
      "100%|██████████| 500000/500000 [00:08<00:00, 57153.93it/s]\n",
      "100%|██████████| 500000/500000 [01:49<00:00, 4558.94it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 106893.12it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 75255.02it/s]\n",
      "100%|██████████| 500000/500000 [00:08<00:00, 55985.29it/s]\n",
      "100%|██████████| 500000/500000 [01:49<00:00, 4571.53it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 109192.77it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 74244.32it/s]\n",
      "100%|██████████| 500000/500000 [00:08<00:00, 57035.50it/s]\n",
      "100%|██████████| 500000/500000 [01:49<00:00, 4569.01it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 107228.46it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 72236.45it/s]\n",
      "100%|██████████| 500000/500000 [00:08<00:00, 56285.50it/s]\n",
      "100%|██████████| 500000/500000 [01:49<00:00, 4569.91it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 103641.25it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 73670.48it/s]\n",
      "100%|██████████| 500000/500000 [00:08<00:00, 56863.96it/s]\n",
      "100%|██████████| 500000/500000 [01:49<00:00, 4555.64it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 104828.37it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 75057.17it/s]\n",
      "100%|██████████| 500000/500000 [00:08<00:00, 57342.69it/s]\n",
      "100%|██████████| 500000/500000 [01:50<00:00, 4545.13it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 106638.21it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 72403.37it/s]\n",
      "100%|██████████| 500000/500000 [00:08<00:00, 56600.23it/s]\n",
      "100%|██████████| 500000/500000 [01:50<00:00, 4528.46it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 102839.25it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 74319.27it/s]\n",
      "100%|██████████| 500000/500000 [00:08<00:00, 55971.96it/s]\n",
      "100%|██████████| 500000/500000 [01:48<00:00, 4616.09it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 106685.42it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 75026.80it/s]\n",
      "100%|██████████| 500000/500000 [00:08<00:00, 55906.58it/s]\n",
      "100%|██████████| 500000/500000 [01:49<00:00, 4564.49it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 104061.36it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 71538.77it/s]\n",
      "100%|██████████| 500000/500000 [00:09<00:00, 55267.10it/s]\n",
      "100%|██████████| 500000/500000 [01:50<00:00, 4531.97it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 107447.55it/s]\n",
      "100%|██████████| 500000/500000 [00:06<00:00, 74723.41it/s]\n",
      "100%|██████████| 500000/500000 [00:08<00:00, 55835.32it/s]\n",
      "100%|██████████| 500000/500000 [01:49<00:00, 4551.68it/s]\n",
      "100%|██████████| 500000/500000 [00:04<00:00, 105378.54it/s]\n",
      "100%|██████████| 1000000/1000000 [00:13<00:00, 73982.39it/s]\n",
      "100%|██████████| 1000000/1000000 [00:17<00:00, 56034.47it/s]\n",
      "100%|██████████| 1000000/1000000 [03:40<00:00, 4539.97it/s]\n",
      "100%|██████████| 1000000/1000000 [00:09<00:00, 104890.12it/s]\n",
      "100%|██████████| 1000000/1000000 [00:13<00:00, 73680.31it/s]\n",
      "100%|██████████| 1000000/1000000 [00:17<00:00, 55713.56it/s]\n",
      "100%|██████████| 1000000/1000000 [03:38<00:00, 4574.24it/s]\n",
      "100%|██████████| 1000000/1000000 [00:09<00:00, 105033.24it/s]\n",
      "100%|██████████| 1000000/1000000 [00:13<00:00, 72858.85it/s]\n",
      "100%|██████████| 1000000/1000000 [00:17<00:00, 55804.63it/s]\n",
      "100%|██████████| 1000000/1000000 [03:40<00:00, 4545.28it/s]\n",
      "100%|██████████| 1000000/1000000 [00:09<00:00, 104772.94it/s]\n",
      "100%|██████████| 1000000/1000000 [00:13<00:00, 74088.78it/s]\n",
      "100%|██████████| 1000000/1000000 [00:17<00:00, 56861.61it/s]\n",
      "100%|██████████| 1000000/1000000 [03:38<00:00, 4575.95it/s]\n",
      "100%|██████████| 1000000/1000000 [00:09<00:00, 104047.74it/s]\n",
      "100%|██████████| 1000000/1000000 [00:13<00:00, 74986.29it/s]\n",
      "100%|██████████| 1000000/1000000 [00:17<00:00, 56324.54it/s]\n",
      "100%|██████████| 1000000/1000000 [03:41<00:00, 4504.93it/s]\n",
      "100%|██████████| 1000000/1000000 [00:09<00:00, 108134.17it/s]\n",
      "100%|██████████| 1000000/1000000 [00:13<00:00, 73597.05it/s]\n",
      "100%|██████████| 1000000/1000000 [00:17<00:00, 56497.77it/s]\n",
      "100%|██████████| 1000000/1000000 [03:39<00:00, 4555.45it/s]\n",
      "100%|██████████| 1000000/1000000 [00:09<00:00, 104815.27it/s]\n",
      "100%|██████████| 1000000/1000000 [00:13<00:00, 73821.40it/s]\n",
      "100%|██████████| 1000000/1000000 [00:17<00:00, 56995.00it/s]\n",
      "100%|██████████| 1000000/1000000 [03:39<00:00, 4561.80it/s]\n",
      "100%|██████████| 1000000/1000000 [00:09<00:00, 105893.87it/s]\n",
      "100%|██████████| 1000000/1000000 [00:13<00:00, 73821.96it/s]\n",
      "100%|██████████| 1000000/1000000 [00:17<00:00, 56105.56it/s]\n",
      "100%|██████████| 1000000/1000000 [03:38<00:00, 4576.11it/s]\n",
      "100%|██████████| 1000000/1000000 [00:09<00:00, 104575.48it/s]\n",
      "100%|██████████| 1000000/1000000 [00:13<00:00, 74579.36it/s]\n",
      "100%|██████████| 1000000/1000000 [00:17<00:00, 55736.35it/s]\n",
      "100%|██████████| 1000000/1000000 [03:40<00:00, 4536.31it/s]\n",
      "100%|██████████| 1000000/1000000 [00:09<00:00, 107169.10it/s]\n",
      "100%|██████████| 1000000/1000000 [00:13<00:00, 73965.52it/s]\n",
      "100%|██████████| 1000000/1000000 [00:17<00:00, 57079.41it/s]\n",
      "100%|██████████| 1000000/1000000 [03:40<00:00, 4542.45it/s]\n",
      "100%|██████████| 1000000/1000000 [00:09<00:00, 105315.02it/s]\n",
      "100%|██████████| 1000000/1000000 [00:13<00:00, 74000.41it/s]\n",
      "100%|██████████| 1000000/1000000 [00:17<00:00, 56294.29it/s]\n",
      "100%|██████████| 1000000/1000000 [03:39<00:00, 4554.79it/s]\n",
      "100%|██████████| 1000000/1000000 [00:09<00:00, 105013.70it/s]\n",
      "100%|██████████| 1000000/1000000 [00:13<00:00, 74950.96it/s]\n",
      "100%|██████████| 1000000/1000000 [00:17<00:00, 55734.16it/s]\n",
      "100%|██████████| 1000000/1000000 [03:40<00:00, 4543.29it/s]\n",
      "100%|██████████| 1000000/1000000 [00:09<00:00, 108171.53it/s]\n",
      "100%|██████████| 1000000/1000000 [00:13<00:00, 74774.23it/s]\n",
      "100%|██████████| 1000000/1000000 [00:17<00:00, 55782.60it/s]\n",
      "100%|██████████| 1000000/1000000 [03:40<00:00, 4536.95it/s]\n",
      "100%|██████████| 1000000/1000000 [00:09<00:00, 107271.59it/s]\n",
      "100%|██████████| 1000000/1000000 [00:13<00:00, 73906.88it/s]\n",
      "100%|██████████| 1000000/1000000 [00:17<00:00, 56243.38it/s]\n",
      "100%|██████████| 1000000/1000000 [03:39<00:00, 4547.37it/s]\n",
      "100%|██████████| 1000000/1000000 [00:09<00:00, 105735.67it/s]\n",
      "100%|██████████| 1000000/1000000 [00:13<00:00, 74019.34it/s]\n",
      "100%|██████████| 1000000/1000000 [00:17<00:00, 55758.97it/s]\n",
      "100%|██████████| 1000000/1000000 [03:39<00:00, 4563.41it/s]\n",
      "100%|██████████| 1000000/1000000 [00:09<00:00, 106704.45it/s]\n",
      "100%|██████████| 1000000/1000000 [00:13<00:00, 74004.59it/s]\n",
      "100%|██████████| 1000000/1000000 [00:17<00:00, 56940.02it/s]\n",
      "100%|██████████| 1000000/1000000 [03:39<00:00, 4555.46it/s]\n",
      "100%|██████████| 1000000/1000000 [00:09<00:00, 106320.06it/s]\n",
      "100%|██████████| 1000000/1000000 [00:13<00:00, 73908.96it/s]\n",
      "100%|██████████| 1000000/1000000 [00:17<00:00, 56163.07it/s]\n",
      "100%|██████████| 1000000/1000000 [03:39<00:00, 4557.36it/s]\n",
      "100%|██████████| 1000000/1000000 [00:09<00:00, 107776.56it/s]\n",
      "100%|██████████| 1000000/1000000 [00:13<00:00, 74857.69it/s]\n",
      "100%|██████████| 1000000/1000000 [00:17<00:00, 56242.00it/s]\n",
      "100%|██████████| 1000000/1000000 [03:39<00:00, 4564.70it/s]\n",
      "100%|██████████| 1000000/1000000 [00:09<00:00, 104789.29it/s]\n",
      "100%|██████████| 1000000/1000000 [00:13<00:00, 74154.89it/s]\n",
      "100%|██████████| 1000000/1000000 [00:17<00:00, 55735.22it/s]\n",
      "100%|██████████| 1000000/1000000 [03:40<00:00, 4534.51it/s]\n",
      "100%|██████████| 1000000/1000000 [00:09<00:00, 104848.49it/s]\n",
      "100%|██████████| 1000000/1000000 [00:13<00:00, 73587.11it/s]\n",
      "100%|██████████| 1000000/1000000 [00:17<00:00, 56949.89it/s]\n",
      "100%|██████████| 1000000/1000000 [03:40<00:00, 4529.46it/s]\n",
      "100%|██████████| 1000000/1000000 [00:09<00:00, 104142.44it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 4h 53min 32s, sys: 54min 7s, total: 5h 47min 40s\n",
      "Wall time: 4h 30min 31s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "experiment_round_size = [1000, 10000, 100000, 500000, 1000000]\n",
    "n_runs_per_round_size = 20\n",
    "n_actions = 5\n",
    "dim_context = 5\n",
    "\n",
    "bandit_policies = [\n",
    "    (Random, {\"n_actions\": n_actions, \"epsilon\": 1.0, \"random_state\": 12345}),\n",
    "    (BernoulliTS, {\"n_actions\": n_actions, \"random_state\": 12345}),\n",
    "    (LinTS, {\"dim\": dim_context, \"n_actions\": n_actions, \"random_state\": 12345}),\n",
    "    (EpsilonGreedy, {\"n_actions\": n_actions, \"epsilon\": 0.1, \"random_state\": 12345}),\n",
    "]\n",
    "\n",
    "dummy_policies = [policy_class(**args) for policy_class, args in bandit_policies]\n",
    "train_rewards = {policy.policy_name: [] for policy in dummy_policies}\n",
    "eval_rewards = {f\"{IPWLearner.__name__} {policy.policy_name}\": [] for policy in dummy_policies}\n",
    "train_rewards[\"n_rounds\"] = []\n",
    "eval_rewards[\"n_rounds\"] = []\n",
    "\n",
    "# In this setting the dataset is static, so we can use the same dataset class and redraw batches\n",
    "# for different settings. When drift or delay is added the dataset object needs to be reinitiated with a new seed\n",
    "# for each experiment.\n",
    "env = BanditEnvironmentSimulator(\n",
    "    n_actions=10,\n",
    "    dim_context=5,\n",
    "    reward_type=\"binary\",\n",
    "    reward_function=logistic_sparse_reward_function,\n",
    "    random_state=12345,\n",
    ")\n",
    "\n",
    "for n_rounds in experiment_round_size:\n",
    "    for experiment in range(n_runs_per_round_size):\n",
    "        training_bandit_batch = env.next_bandit_round_batch(n_rounds=n_rounds)\n",
    "        evaluation_bandit_batch = env.next_bandit_round_batch(n_rounds=n_rounds)\n",
    "\n",
    "        # Train the bandit algorithms and get the rewards for the training and evaluation period\n",
    "        for policy_class, args in bandit_policies:\n",
    "            policy = policy_class(**args)\n",
    "\n",
    "            training_simulator = BanditPolicySimulator(policy=policy)\n",
    "\n",
    "            training_simulator.steps(batch_bandit_rounds=training_bandit_batch)\n",
    "            train_rewards[policy.policy_name].append(training_simulator.total_reward)\n",
    "\n",
    "            # Train a propensity model on the actions in the training period to get propensities per round\n",
    "            propensity_model = LogisticRegression(random_state=12345)\n",
    "            propensity_model.fit(training_simulator.contexts, training_simulator.selected_actions)\n",
    "            pscores = propensity_model.predict_proba(training_simulator.contexts)\n",
    "\n",
    "            # Train an IPW learning from the logged data and learned propensities\n",
    "            ipw_learner = IPWLearner(n_actions=env.n_actions,\n",
    "                                     base_classifier=RandomForest(n_estimators=30, min_samples_leaf=10, random_state=12345))\n",
    "\n",
    "            ipw_learner.fit(\n",
    "                context=training_simulator.contexts,\n",
    "                action=training_simulator.selected_actions,\n",
    "                reward=training_simulator.obtained_rewards,\n",
    "                pscore=np.choose(training_simulator.selected_actions, pscores.T)\n",
    "            )\n",
    "            eval_action_dists = ipw_learner.predict(\n",
    "                context=evaluation_bandit_batch.context\n",
    "            )\n",
    "\n",
    "            eval_rewards[f\"{ipw_learner.policy_name} {policy.policy_name}\"].append(\n",
    "                np.sum(eval_action_dists.squeeze(axis=2) * evaluation_bandit_batch.rewards)\n",
    "            )\n",
    "\n",
    "        train_rewards[\"n_rounds\"].append(n_rounds)\n",
    "        eval_rewards[\"n_rounds\"].append(n_rounds)\n",
    "\n"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "The plot below shows how the performance of the same IPW model setup can significantly differ when the underlying data collection process changes."
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%% md\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "outputs": [
    {
     "data": {
      "text/plain": "<Figure size 432x288 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_average_reward_per_n_rounds(eval_rewards)\n"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "outputs": [],
   "source": [],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}