tests/gail_test.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The autoreload extension is already loaded. To reload it, use:\n",
      "  %reload_ext autoreload\n"
     ]
    }
   ],
   "source": [
    "import sys; sys.path.insert(0, '..')\n",
    "\n",
    "import gym\n",
    "import sys\n",
    "import torch\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from IPython.display import clear_output, Javascript\n",
    "\n",
    "import pytorch_drl.models.actor_critic_models as models\n",
    "import pytorch_drl.models.gail_models as gail_models\n",
    "from notebook_utils import plot, mean_filter\n",
    "from pytorch_drl.algs.gail import GAIL\n",
    "\n",
    "%matplotlib inline\n",
    "%load_ext autoreload\n",
    "%autoreload 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Device: cuda:0\n"
     ]
    }
   ],
   "source": [
    "device =torch.device(\"cuda:0\" if torch.cuda.is_available() else \"cpu\")\n",
    "print(\"Device:\", device)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Create environment"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "State size: 4 \n",
      "Action size: 2\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/selo/Envs/ml-agent_/gym/gym/logger.py:30: UserWarning: \u001b[33mWARN: Box bound precision lowered by casting to float32\u001b[0m\n",
      "  warnings.warn(colorize('%s: %s'%('WARN', msg % args), 'yellow'))\n"
     ]
    }
   ],
   "source": [
    "env_name = \"CartPole-v0\"\n",
    "env = gym.make(env_name)\n",
    "env.seed(0)\n",
    "\n",
    "state_size =  env.observation_space.shape[0]\n",
    "action_size = env.action_space.n\n",
    "\n",
    "print(\"State size:\", state_size, \"\\nAction size:\", action_size)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## GAIL Test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "expert_trajectories = np.load(\"experts/ppo_{}\".format(env_name))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(300000, 6)\n"
     ]
    }
   ],
   "source": [
    "print(expert_trajectories.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Avg score is 172.09, i: 1999                   "
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "actor_critic = models.ActorCriticMLP(state_size, action_size, env.action_space, H=64) # 64\n",
    "\n",
    "discriminator = gail_models.GAILDiscriminator(state_size, action_size)\n",
    "\n",
    "tmax = 20\n",
    "n_traj = 2000\n",
    "n_env = 16\n",
    "ppo_epochs = 4\n",
    "batch_size = 64\n",
    "\n",
    "# init agent:\n",
    "agent = GAIL(actor_critic,\n",
    "             discriminator,\n",
    "             expert_trajectories,\n",
    "             env_name,\n",
    "             action_size,\n",
    "             gamma=0.99, \n",
    "             gail_epochs=1,\n",
    "             ppo_epochs=ppo_epochs,\n",
    "             lr_ppo=3e-3, \n",
    "             lr_discriminator=3e-3,\n",
    "             gae_tau=0.95,\n",
    "             n_env=n_env,\n",
    "             device=device,\n",
    "             max_grad_norm=0.5,\n",
    "             critic_coef=0.5,\n",
    "             entropy_coef=0.01,\n",
    "             mini_batch_size=batch_size,\n",
    "             )\n",
    "   \n",
    "    \n",
    "# train the agent\n",
    "max_score = 195.\n",
    "alg_name = \"gail_\".format(env_name)\n",
    "scores, losses = agent.train(tmax, n_traj,  env, max_score, alg_name)\n",
    "\n",
    "# plot the training:\n",
    "x = np.arange(len(scores))\n",
    "scores = mean_filter(scores, 50)\n",
    "plt.plot(x, scores, label = \"scores\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Trained Agent Demonstration"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "200.0\n",
      "200.0\n",
      "159.0\n",
      "200.0\n",
      "197.0\n"
     ]
    }
   ],
   "source": [
    "agent.test(env, render=True, n_episodes=5)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python (ml-agents)",
   "language": "python",
   "name": "ml-agents"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}