ap_rllib_experiment_analysis/notebooks/merge_exported_policy_pairs.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# take to .csv files (from eval and 5adv analyses) and merge them into one with all pairs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /scratch/sergei/miniconda3/envs/adv-tf2/lib/python3.8/site-packages/tensorflow/python/compat/v2_compat.py:96: disable_resource_variables (from tensorflow.python.ops.variable_scope) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "non-resource variables are not supported in the long term\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "from tqdm.auto import tqdm\n",
    "from ap_rllib.config import get_trainer, get_config_by_name\n",
    "import os\n",
    "import pandas as pd\n",
    "import ray"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "files = ['/tmp/to_eval_5adv.csv', '/tmp/to_eval_1adv_eval.csv']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "dfs = [pd.read_csv(f) for f in files]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2020-12-08 06:29:35,865\tINFO services.py:1090 -- View the Ray dashboard at \u001b[1m\u001b[32mhttp://127.0.0.1:8265\u001b[39m\u001b[22m\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "{'node_ip_address': '128.32.175.10',\n",
       " 'raylet_ip_address': '128.32.175.10',\n",
       " 'redis_address': '128.32.175.10:13157',\n",
       " 'object_store_address': '/tmp/ray/session_2020-12-08_06-29-35_005056_15499/sockets/plasma_store',\n",
       " 'raylet_socket_name': '/tmp/ray/session_2020-12-08_06-29-35_005056_15499/sockets/raylet',\n",
       " 'webui_url': '127.0.0.1:8265',\n",
       " 'session_dir': '/tmp/ray/session_2020-12-08_06-29-35_005056_15499',\n",
       " 'metrics_export_port': 57385,\n",
       " 'node_id': '74a9a479e1656975c8a2071146e35101816b64d7'}"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# need ray for parallel evaluation\n",
    "ray.shutdown()\n",
    "ray.init(num_cpus=28, ignore_reinit_error=True, log_to_driver=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2020-12-08 06:29:40,586\tINFO trainer.py:588 -- Executing eagerly, with eager_tracing=False\n",
      "2020-12-08 06:29:40,588\tINFO trainer.py:1064 -- `_use_trajectory_view_api` only supported for PyTorch so far! Will run w/o.\n",
      "2020-12-08 06:29:40,589\tINFO trainer.py:617 -- Current log_level is WARN. For more information, set 'log_level': 'INFO' / 'DEBUG' or use the -v and -vv flags.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Creating agent humanoid_blocker\n",
      "Reading agent XML from: /scratch/sergei/better-adversarial-defenses/multiagent-competition/gym_compete/new_envs/assets/humanoid_body.xml\n",
      "Creating agent humanoid\n",
      "Reading agent XML from: /scratch/sergei/better-adversarial-defenses/multiagent-competition/gym_compete/new_envs/assets/humanoid_body.xml\n",
      "Scene XML path: /scratch/sergei/better-adversarial-defenses/multiagent-competition/gym_compete/new_envs/assets/world_body.humanoid_body.humanoid_body.xml\n",
      "Created Scene with agents\n",
      "Creating agent humanoid_blocker\n",
      "Reading agent XML from: /scratch/sergei/better-adversarial-defenses/multiagent-competition/gym_compete/new_envs/assets/humanoid_body.xml\n",
      "Creating agent humanoid\n",
      "Reading agent XML from: /scratch/sergei/better-adversarial-defenses/multiagent-competition/gym_compete/new_envs/assets/humanoid_body.xml\n",
      "Scene XML path: /scratch/sergei/better-adversarial-defenses/multiagent-competition/gym_compete/new_envs/assets/world_body.humanoid_body.humanoid_body.xml\n",
      "Created Scene with agents\n",
      "Setting NN weights\n",
      "Loading normalization\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2020-12-08 06:29:42,060\tWARNING catalog.py:333 -- Custom ModelV2 should accept all custom options as **kwargs, instead of expecting them in config['custom_model_config']!\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Setting NN weights\n",
      "Loading normalization\n",
      "It looks like variables [\"<tf.Varaible player_1/param/clip:0 shape=(1,) <dtype: 'float32'>>\", \"<tf.Varaible dense_1/kernel:0 shape=(64, 64) <dtype: 'float32'>>\", \"<tf.Varaible dense_1/bias:0 shape=(64,) <dtype: 'float32'>>\", \"<tf.Varaible player_1/param/mean:0 shape=(380,) <dtype: 'float32'>>\", \"<tf.Varaible dense/kernel:0 shape=(380, 64) <dtype: 'float32'>>\", \"<tf.Varaible player_1/param/std:0 shape=(380,) <dtype: 'float32'>>\", \"<tf.Varaible mean/bias:0 shape=(17,) <dtype: 'float32'>>\", \"<tf.Varaible player_1/param/mean:0 shape=(1,) <dtype: 'float32'>>\", \"<tf.Varaible player_1/param/mean:0 shape=(380,) <dtype: 'float32'>>\", \"<tf.Varaible player_1/param/clip:0 shape=(1,) <dtype: 'float32'>>\", \"<tf.Varaible player_1/sequential/h2/bias:0 shape=(64,) <dtype: 'float32'>>\", \"<tf.Varaible player_1/sequential/value/kernel:0 shape=(64, 1) <dtype: 'float32'>>\", \"<tf.Varaible mean/kernel:0 shape=(64, 17) <dtype: 'float32'>>\", \"<tf.Varaible player_1/param/std:0 shape=(380,) <dtype: 'float32'>>\", \"<tf.Varaible player_1/sequential/h2/kernel:0 shape=(64, 64) <dtype: 'float32'>>\", \"<tf.Varaible player_1/sequential/value/bias:0 shape=(1,) <dtype: 'float32'>>\", \"<tf.Varaible dense/bias:0 shape=(64,) <dtype: 'float32'>>\", \"<tf.Varaible player_1/var/std:0 shape=(1, 17) <dtype: 'float32'>>\", \"<tf.Varaible player_1/param/std:0 shape=(1,) <dtype: 'float32'>>\", \"<tf.Varaible player_1/sequential/h1/kernel:0 shape=(380, 64) <dtype: 'float32'>>\", \"<tf.Varaible player_1/sequential/h1/bias:0 shape=(64,) <dtype: 'float32'>>\"] were created as part of <gym_compete_rllib.gym_compete_to_rllib.GymCompetePretrainedModel object at 0x7f385ae41040> but does not appear in model.variables() ([\"<tf.Varaible player_1/param/clip:0 shape=(1,) <dtype: 'float32'>>\", \"<tf.Varaible player_1/param/std:0 shape=(380,) <dtype: 'float32'>>\", \"<tf.Varaible player_1/param/mean:0 shape=(380,) <dtype: 'float32'>>\", \"<tf.Varaible dense_3/kernel:0 shape=(64, 64) <dtype: 'float32'>>\", \"<tf.Varaible dense_2/kernel:0 shape=(380, 64) <dtype: 'float32'>>\", \"<tf.Varaible player_1/param/std:0 shape=(380,) <dtype: 'float32'>>\", \"<tf.Varaible dense_3/bias:0 shape=(64,) <dtype: 'float32'>>\", \"<tf.Varaible player_1/sequential_1/h2/kernel:0 shape=(64, 64) <dtype: 'float32'>>\", \"<tf.Varaible player_1/var/std:0 shape=(1, 17) <dtype: 'float32'>>\", \"<tf.Varaible player_1/param/mean:0 shape=(380,) <dtype: 'float32'>>\", \"<tf.Varaible player_1/sequential_1/h1/kernel:0 shape=(380, 64) <dtype: 'float32'>>\", \"<tf.Varaible player_1/sequential_1/h2/bias:0 shape=(64,) <dtype: 'float32'>>\", \"<tf.Varaible player_1/param/mean:0 shape=(1,) <dtype: 'float32'>>\", \"<tf.Varaible player_1/param/clip:0 shape=(1,) <dtype: 'float32'>>\", \"<tf.Varaible dense_2/bias:0 shape=(64,) <dtype: 'float32'>>\", \"<tf.Varaible mean_1/bias:0 shape=(17,) <dtype: 'float32'>>\", \"<tf.Varaible player_1/sequential_1/h1/bias:0 shape=(64,) <dtype: 'float32'>>\", \"<tf.Varaible player_1/sequential_1/value/kernel:0 shape=(64, 1) <dtype: 'float32'>>\", \"<tf.Varaible player_1/sequential_1/value/bias:0 shape=(1,) <dtype: 'float32'>>\", \"<tf.Varaible mean_1/kernel:0 shape=(64, 17) <dtype: 'float32'>>\", \"<tf.Varaible player_1/param/std:0 shape=(1,) <dtype: 'float32'>>\"]). Did you forget to call model.register_variables() on the variables in question?\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2020-12-08 06:29:43,304\tWARNING catalog.py:333 -- Custom ModelV2 should accept all custom options as **kwargs, instead of expecting them in config['custom_model_config']!\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Setting NN weights\n",
      "Loading normalization\n",
      "Setting NN weights\n",
      "Loading normalization\n",
      "It looks like variables [\"<tf.Varaible player_2/param/std:0 shape=(380,) <dtype: 'float32'>>\", \"<tf.Varaible player_2/sequential_2/h1/bias:0 shape=(64,) <dtype: 'float32'>>\", \"<tf.Varaible player_2/sequential_2/h2/kernel:0 shape=(64, 64) <dtype: 'float32'>>\", \"<tf.Varaible player_2/param/mean:0 shape=(380,) <dtype: 'float32'>>\", \"<tf.Varaible player_2/sequential_2/h2/bias:0 shape=(64,) <dtype: 'float32'>>\", \"<tf.Varaible player_2/param/mean:0 shape=(380,) <dtype: 'float32'>>\", \"<tf.Varaible player_2/param/mean:0 shape=(1,) <dtype: 'float32'>>\", \"<tf.Varaible player_2/sequential_2/value/bias:0 shape=(1,) <dtype: 'float32'>>\", \"<tf.Varaible dense_4/bias:0 shape=(64,) <dtype: 'float32'>>\", \"<tf.Varaible dense_5/kernel:0 shape=(64, 64) <dtype: 'float32'>>\", \"<tf.Varaible player_2/sequential_2/value/kernel:0 shape=(64, 1) <dtype: 'float32'>>\", \"<tf.Varaible player_2/param/clip:0 shape=(1,) <dtype: 'float32'>>\", \"<tf.Varaible player_2/param/std:0 shape=(380,) <dtype: 'float32'>>\", \"<tf.Varaible dense_4/kernel:0 shape=(380, 64) <dtype: 'float32'>>\", \"<tf.Varaible dense_5/bias:0 shape=(64,) <dtype: 'float32'>>\", \"<tf.Varaible mean_2/kernel:0 shape=(64, 17) <dtype: 'float32'>>\", \"<tf.Varaible mean_2/bias:0 shape=(17,) <dtype: 'float32'>>\", \"<tf.Varaible player_2/var/std:0 shape=(1, 17) <dtype: 'float32'>>\", \"<tf.Varaible player_2/param/clip:0 shape=(1,) <dtype: 'float32'>>\", \"<tf.Varaible player_2/param/std:0 shape=(1,) <dtype: 'float32'>>\", \"<tf.Varaible player_2/sequential_2/h1/kernel:0 shape=(380, 64) <dtype: 'float32'>>\"] were created as part of <gym_compete_rllib.gym_compete_to_rllib.GymCompetePretrainedModel object at 0x7f3850527b80> but does not appear in model.variables() ([\"<tf.Varaible player_2/sequential_3/h1/kernel:0 shape=(380, 64) <dtype: 'float32'>>\", \"<tf.Varaible player_2/param/mean:0 shape=(1,) <dtype: 'float32'>>\", \"<tf.Varaible player_2/sequential_3/h1/bias:0 shape=(64,) <dtype: 'float32'>>\", \"<tf.Varaible player_2/param/clip:0 shape=(1,) <dtype: 'float32'>>\", \"<tf.Varaible player_2/param/std:0 shape=(380,) <dtype: 'float32'>>\", \"<tf.Varaible player_2/param/mean:0 shape=(380,) <dtype: 'float32'>>\", \"<tf.Varaible player_2/var/std:0 shape=(1, 17) <dtype: 'float32'>>\", \"<tf.Varaible player_2/sequential_3/h2/kernel:0 shape=(64, 64) <dtype: 'float32'>>\", \"<tf.Varaible dense_6/bias:0 shape=(64,) <dtype: 'float32'>>\", \"<tf.Varaible dense_7/kernel:0 shape=(64, 64) <dtype: 'float32'>>\", \"<tf.Varaible player_2/sequential_3/h2/bias:0 shape=(64,) <dtype: 'float32'>>\", \"<tf.Varaible player_2/param/std:0 shape=(1,) <dtype: 'float32'>>\", \"<tf.Varaible dense_6/kernel:0 shape=(380, 64) <dtype: 'float32'>>\", \"<tf.Varaible dense_7/bias:0 shape=(64,) <dtype: 'float32'>>\", \"<tf.Varaible mean_3/kernel:0 shape=(64, 17) <dtype: 'float32'>>\", \"<tf.Varaible player_2/param/std:0 shape=(380,) <dtype: 'float32'>>\", \"<tf.Varaible mean_3/bias:0 shape=(17,) <dtype: 'float32'>>\", \"<tf.Varaible player_2/param/clip:0 shape=(1,) <dtype: 'float32'>>\", \"<tf.Varaible player_2/sequential_3/value/kernel:0 shape=(64, 1) <dtype: 'float32'>>\", \"<tf.Varaible player_2/param/mean:0 shape=(380,) <dtype: 'float32'>>\", \"<tf.Varaible player_2/sequential_3/value/bias:0 shape=(1,) <dtype: 'float32'>>\"]). Did you forget to call model.register_variables() on the variables in question?\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2020-12-08 06:29:43,684\tWARNING util.py:40 -- Install gputil for GPU system monitoring.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Creating agent humanoid_blocker\n",
      "Reading agent XML from: /scratch/sergei/better-adversarial-defenses/multiagent-competition/gym_compete/new_envs/assets/humanoid_body.xml\n",
      "Creating agent humanoid\n",
      "Reading agent XML from: /scratch/sergei/better-adversarial-defenses/multiagent-competition/gym_compete/new_envs/assets/humanoid_body.xml\n",
      "Scene XML path: /scratch/sergei/better-adversarial-defenses/multiagent-competition/gym_compete/new_envs/assets/world_body.humanoid_body.humanoid_body.xml\n",
      "Created Scene with agents\n"
     ]
    }
   ],
   "source": [
    "config = 'external_test'\n",
    "config_ = get_config_by_name(config)\n",
    "config_['_verbose'] = False\n",
    "config_['_policies'] = [None, 'pretrained', 'pretrained']\n",
    "trainer = get_trainer(config_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "import logging\n",
    "logging.basicConfig(level=logging.WARNING)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "policies = ['player_1', 'player_2']\n",
    "data = {p: [] for p in policies} # format: p -> [(name, hash, weights)]\n",
    "\n",
    "for f in tqdm(files):\n",
    "    for _, r in tqdm(pd.read_csv(f).iterrows(), leave=False):\n",
    "        trainer.restore(r.checkpoint_rllib)\n",
    "        for p in policies:\n",
    "            name = r[p]\n",
    "            weights = trainer.get_weights()[p]\n",
    "            hash_ = sum([np.sum(t) for t in weights])\n",
    "            data[p].append((name, hash_, weights))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'player_1': {0,\n",
       "  1,\n",
       "  2,\n",
       "  3,\n",
       "  4,\n",
       "  5,\n",
       "  6,\n",
       "  7,\n",
       "  8,\n",
       "  9,\n",
       "  10,\n",
       "  11,\n",
       "  12,\n",
       "  13,\n",
       "  14,\n",
       "  15,\n",
       "  16,\n",
       "  17,\n",
       "  18,\n",
       "  19,\n",
       "  20,\n",
       "  21,\n",
       "  22,\n",
       "  23,\n",
       "  24,\n",
       "  25,\n",
       "  26,\n",
       "  27,\n",
       "  28,\n",
       "  29,\n",
       "  30,\n",
       "  31,\n",
       "  32,\n",
       "  33,\n",
       "  34,\n",
       "  35,\n",
       "  36,\n",
       "  37,\n",
       "  38,\n",
       "  39,\n",
       "  40,\n",
       "  41,\n",
       "  42,\n",
       "  43,\n",
       "  44,\n",
       "  45,\n",
       "  5100,\n",
       "  5102,\n",
       "  5103,\n",
       "  5104,\n",
       "  5105,\n",
       "  5106,\n",
       "  5107,\n",
       "  5108,\n",
       "  5109,\n",
       "  5110,\n",
       "  5111,\n",
       "  5112,\n",
       "  5200,\n",
       "  5202,\n",
       "  5203,\n",
       "  5204,\n",
       "  5205,\n",
       "  5206,\n",
       "  5207,\n",
       "  5208,\n",
       "  5209,\n",
       "  5210,\n",
       "  5211,\n",
       "  5212,\n",
       "  5300,\n",
       "  5302,\n",
       "  5303,\n",
       "  5304,\n",
       "  5305,\n",
       "  5306,\n",
       "  5307,\n",
       "  5308,\n",
       "  5309,\n",
       "  5310,\n",
       "  5311,\n",
       "  5312,\n",
       "  5400,\n",
       "  5402,\n",
       "  5403,\n",
       "  5404,\n",
       "  5405,\n",
       "  5406,\n",
       "  5407,\n",
       "  5408,\n",
       "  5409,\n",
       "  5410,\n",
       "  5411,\n",
       "  5412,\n",
       "  5500,\n",
       "  5502,\n",
       "  5503,\n",
       "  5504,\n",
       "  5505,\n",
       "  5506,\n",
       "  5507,\n",
       "  5508,\n",
       "  5509,\n",
       "  5510,\n",
       "  5511,\n",
       "  5512},\n",
       " 'player_2': {34,\n",
       "  36,\n",
       "  37,\n",
       "  38,\n",
       "  40,\n",
       "  41,\n",
       "  42,\n",
       "  43,\n",
       "  44,\n",
       "  45,\n",
       "  5000,\n",
       "  5002,\n",
       "  5003,\n",
       "  5004,\n",
       "  5005,\n",
       "  5006,\n",
       "  5007,\n",
       "  5008,\n",
       "  5009,\n",
       "  5010,\n",
       "  5011,\n",
       "  5012}}"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "unique_names"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "unique_names = {p: set(d[0] for d in data[p]) for p in policies}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "# checking that hashes match for the same name\n",
    "for p in policies:\n",
    "    for n in unique_names[p]:\n",
    "        vals = [d for d in data[p] if d[0] == n]\n",
    "        hashes = [v[1] for v in vals]\n",
    "        assert all([hashes[0] == h for h in hashes])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "# only unique names\n",
    "data_u = {p: [] for p in policies}\n",
    "for p in policies:\n",
    "    for n in unique_names[p]:\n",
    "        val = [d for d in data[p] if d[0] == n][0]\n",
    "        data_u[p].append(val)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "2332"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(data_u['player_1']) * len(data_u['player_2'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "# checking that got all the agents\n",
    "total_players_df = {p: sum([len(set(df[p])) for df in dfs]) for p in policies}\n",
    "for p in policies:\n",
    "    assert len(data_u[p]) == total_players_df[p]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "from uuid import uuid1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "# saving the output!\n",
    "ds = []\n",
    "for p1name, _, p1w in data_u['player_1']:\n",
    "    for p2name, _, p2w in data_u['player_2']:\n",
    "        d = {'player_1': p1name, 'player_2': p2name}\n",
    "        trainer.set_weights({'player_1': p1w, 'player_2': p2w})\n",
    "        ckpt = trainer.save(f\"/tmp/{uuid1()}\")\n",
    "        w_hash = sum([sum([np.sum(t) for t in y]) for y in trainer.get_weights().values()])\n",
    "        d['checkpoint_rllib'] = ckpt\n",
    "        d['hash'] = w_hash\n",
    "#         print(p1name, p2name)\n",
    "    \n",
    "        for d_other in ds:\n",
    "            if d_other['hash'] == d['hash']:\n",
    "                print(\"Potential collision\", d, d_other)\n",
    "                \n",
    "        ds.append(d)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "hashes = [x['hash'] for x in ds]\n",
    "assert len(hashes) == len(set(hashes)), \"Some pairs are the same\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "fn = \"/tmp/to_eval_all.csv\"\n",
    "df_to_eval = pd.DataFrame(ds)\n",
    "df_to_eval.to_csv(fn, index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>player_1</th>\n",
       "      <th>player_2</th>\n",
       "      <th>checkpoint_rllib</th>\n",
       "      <th>hash</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>5000</td>\n",
       "      <td>/tmp/e72400bc-3961-11eb-85ce-1831bf455a80/chec...</td>\n",
       "      <td>4420.587047</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0</td>\n",
       "      <td>5002</td>\n",
       "      <td>/tmp/e72400bd-3961-11eb-85ce-1831bf455a80/chec...</td>\n",
       "      <td>4389.690206</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0</td>\n",
       "      <td>5003</td>\n",
       "      <td>/tmp/e72400be-3961-11eb-85ce-1831bf455a80/chec...</td>\n",
       "      <td>4372.916389</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0</td>\n",
       "      <td>5004</td>\n",
       "      <td>/tmp/e72400bf-3961-11eb-85ce-1831bf455a80/chec...</td>\n",
       "      <td>4412.166457</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0</td>\n",
       "      <td>5005</td>\n",
       "      <td>/tmp/e72400c0-3961-11eb-85ce-1831bf455a80/chec...</td>\n",
       "      <td>4418.325421</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2327</th>\n",
       "      <td>5112</td>\n",
       "      <td>41</td>\n",
       "      <td>/tmp/faf8b9ff-3961-11eb-85ce-1831bf455a80/chec...</td>\n",
       "      <td>4451.015669</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2328</th>\n",
       "      <td>5112</td>\n",
       "      <td>42</td>\n",
       "      <td>/tmp/faf8ba00-3961-11eb-85ce-1831bf455a80/chec...</td>\n",
       "      <td>4470.394853</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2329</th>\n",
       "      <td>5112</td>\n",
       "      <td>43</td>\n",
       "      <td>/tmp/faf8ba01-3961-11eb-85ce-1831bf455a80/chec...</td>\n",
       "      <td>4435.065287</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2330</th>\n",
       "      <td>5112</td>\n",
       "      <td>44</td>\n",
       "      <td>/tmp/faf8ba02-3961-11eb-85ce-1831bf455a80/chec...</td>\n",
       "      <td>4465.460510</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2331</th>\n",
       "      <td>5112</td>\n",
       "      <td>45</td>\n",
       "      <td>/tmp/faf8ba03-3961-11eb-85ce-1831bf455a80/chec...</td>\n",
       "      <td>4454.926560</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>2332 rows × 4 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "      player_1  player_2                                   checkpoint_rllib  \\\n",
       "0            0      5000  /tmp/e72400bc-3961-11eb-85ce-1831bf455a80/chec...   \n",
       "1            0      5002  /tmp/e72400bd-3961-11eb-85ce-1831bf455a80/chec...   \n",
       "2            0      5003  /tmp/e72400be-3961-11eb-85ce-1831bf455a80/chec...   \n",
       "3            0      5004  /tmp/e72400bf-3961-11eb-85ce-1831bf455a80/chec...   \n",
       "4            0      5005  /tmp/e72400c0-3961-11eb-85ce-1831bf455a80/chec...   \n",
       "...        ...       ...                                                ...   \n",
       "2327      5112        41  /tmp/faf8b9ff-3961-11eb-85ce-1831bf455a80/chec...   \n",
       "2328      5112        42  /tmp/faf8ba00-3961-11eb-85ce-1831bf455a80/chec...   \n",
       "2329      5112        43  /tmp/faf8ba01-3961-11eb-85ce-1831bf455a80/chec...   \n",
       "2330      5112        44  /tmp/faf8ba02-3961-11eb-85ce-1831bf455a80/chec...   \n",
       "2331      5112        45  /tmp/faf8ba03-3961-11eb-85ce-1831bf455a80/chec...   \n",
       "\n",
       "             hash  \n",
       "0     4420.587047  \n",
       "1     4389.690206  \n",
       "2     4372.916389  \n",
       "3     4412.166457  \n",
       "4     4418.325421  \n",
       "...           ...  \n",
       "2327  4451.015669  \n",
       "2328  4470.394853  \n",
       "2329  4435.065287  \n",
       "2330  4465.460510  \n",
       "2331  4454.926560  \n",
       "\n",
       "[2332 rows x 4 columns]"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_to_eval"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "def weights_equal(w1, w2):\n",
    "    \"\"\"List of weights == another list of weights?\"\"\"\n",
    "    assert len(w1) == len(w2)\n",
    "    for a, b in zip(w1, w2):\n",
    "        if not np.allclose(a, b):\n",
    "            return False\n",
    "    return True"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "from matplotlib import pyplot as plt\n",
    "import seaborn as sns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "for p in policies:\n",
    "    plt.title(f\"Policy weight equality {p}\")\n",
    "    weq = [[weights_equal(x[2], y[2]) for x in data_u[p]] for y in data_u[p]]\n",
    "    sns.heatmap(weq)\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}