notebooks/dqn.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "41aabc2e-d846-46b3-aebb-ffcf04ece369",
   "metadata": {},
   "source": [
    "# Discrete DQN\n",
    "\n",
    "This is the implementation of DQN for discrete action space using environment information inputs."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "eb08f0a9-12b3-495a-b31c-3de241fa906e",
   "metadata": {
    "tags": []
   },
   "source": [
    "## Setup arguments"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "d55e070b-7ed3-463f-b04c-90b5cab363d1",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-10-10T22:05:49.556642Z",
     "iopub.status.busy": "2022-10-10T22:05:49.556364Z",
     "iopub.status.idle": "2022-10-10T22:05:49.963499Z",
     "shell.execute_reply": "2022-10-10T22:05:49.962837Z",
     "shell.execute_reply.started": "2022-10-10T22:05:49.556590Z"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "%load_ext autoreload\n",
    "%autoreload 2\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "04c139fc-ac39-4efa-92da-23e771cbed31",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-10-10T22:05:49.965441Z",
     "iopub.status.busy": "2022-10-10T22:05:49.965165Z",
     "iopub.status.idle": "2022-10-10T22:05:50.802455Z",
     "shell.execute_reply": "2022-10-10T22:05:50.801748Z",
     "shell.execute_reply.started": "2022-10-10T22:05:49.965421Z"
    },
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "20db73f2ff4142ab9aa845f05ac71d90",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "IntText(value=1, description='seed', tooltip='seed of the experiment')"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "4a7693978f6f41cb911e2b0dc494cd75",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Checkbox(value=True, description='torch-deterministic', indent=False, tooltip='if toggled, `torch.backends.cud…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "baa2c5fdb91443e693e3f801893bbcdc",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Checkbox(value=True, description='cuda', indent=False, tooltip='if toggled, cuda will be enabled by default')"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "07b3b1e9317946ffa389bd0c0653d87d",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Checkbox(value=False, description='capture-video', indent=False, tooltip='whether to capture videos of the age…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "713d136b44e14263b60b68140c6bb607",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Text(value='CartPole-v1', description='env-id', tooltip='the id of the environment')"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "97e91e3311044e4dbd93f828d9bc5c3c",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "IntText(value=500000, description='total-timesteps', tooltip='total timesteps of the experiments')"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "e150e187495f43ff90e992d0babce497",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "FloatText(value=0.00025, description='learning-rate', tooltip='the learning rate of the optimizer')"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "b9badbbe27a64810877056265afab839",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "IntText(value=10000, description='buffer-size', tooltip='the replay memory buffer size')"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "915434a6f2224a8d8bf7529ed30ce53c",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "FloatText(value=0.99, description='gamma', tooltip='the discount factor gamma')"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "bc04812fbfdd491e826d8a7cb2b32a93",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "IntText(value=500, description='target-network-frequency', tooltip='the timesteps it takes to update the targe…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "13094260b82540deac17718c3935c116",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "IntText(value=128, description='batch-size', tooltip='the batch size of sample from the reply memory')"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "600c7d368f15488cbc5d540d7863bdc0",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "FloatText(value=1.0, description='start-e', tooltip='the starting epsilon for exploration')"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "6f74bdd6b12e4b75a2480c007a9606d1",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "FloatText(value=0.05, description='end-e', tooltip='the ending epsilon for exploration')"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "134252e334e74c9a93ed3d9dc83b6e5e",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "FloatText(value=0.5, description='exploration-fraction', tooltip='the fraction of `total-timesteps` it takes f…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "11de4cfd9efc4e068905657380d76002",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "IntText(value=10000, description='learning-starts', tooltip='timestep to start learning')"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "0bdd174fdbd24fec8b9751a40b9782fd",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "IntText(value=10, description='train-frequency', tooltip='the frequency of training')"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import ipywidgets as widgets\n",
    "from IPython.display import display, clear_output, Image\n",
    "\n",
    "import time\n",
    "\n",
    "import torch\n",
    "import random\n",
    "\n",
    "from torch_rl.jupyter_utils import (\n",
    "    WidgetParser,\n",
    "    plot_metrics_notebook\n",
    ")\n",
    "\n",
    "from torch_rl.utils import (\n",
    "    seed_everything,\n",
    "    handle_terminal_ob,\n",
    "    mp4_to_gif,\n",
    "    MetricStore,\n",
    "    SPS\n",
    ")\n",
    "\n",
    "\n",
    "parser = WidgetParser()\n",
    "parser.add_widget(widgets.IntText(\n",
    "    description='seed',\n",
    "    value=1,\n",
    "    tooltip=\"seed of the experiment\"\n",
    "))\n",
    "\n",
    "parser.add_widget(widgets.Checkbox(\n",
    "    description=\"torch-deterministic\",\n",
    "    value=True,\n",
    "    disabled=False,\n",
    "    indent=False,\n",
    "    tooltip=\"if toggled, `torch.backends.cudnn.deterministic=False`\"\n",
    "))\n",
    "\n",
    "parser.add_widget(widgets.Checkbox(\n",
    "    description=\"cuda\",\n",
    "    value=True,\n",
    "    disabled=False,\n",
    "    indent=False,\n",
    "    tooltip=\"if toggled, cuda will be enabled by default\"\n",
    "))\n",
    "\n",
    "parser.add_widget(widgets.Checkbox(\n",
    "    description=\"capture-video\",\n",
    "    value=False,\n",
    "    disabled=False,\n",
    "    indent=False,\n",
    "    tooltip=\"whether to capture videos of the agent performances (check out `videos` folder)\"\n",
    "))\n",
    "\n",
    "parser.add_widget(widgets.Text(\n",
    "    description=\"env-id\",\n",
    "    value=\"CartPole-v1\",\n",
    "    tooltip=\"the id of the environment\"\n",
    "))\n",
    "\n",
    "parser.add_widget(widgets.IntText(\n",
    "    description='total-timesteps',\n",
    "    value=500000,\n",
    "    tooltip=\"total timesteps of the experiments\"\n",
    "))\n",
    "\n",
    "parser.add_widget(widgets.FloatText(\n",
    "    description='learning-rate',\n",
    "    value=2.5e-4,\n",
    "    tooltip=\"the learning rate of the optimizer\"\n",
    "))\n",
    "\n",
    "parser.add_widget(widgets.IntText(\n",
    "    description='buffer-size',\n",
    "    value=10000,\n",
    "    tooltip=\"the replay memory buffer size\"\n",
    "))\n",
    "\n",
    "parser.add_widget(widgets.FloatText(\n",
    "    description='gamma',\n",
    "    value=0.99,\n",
    "    tooltip=\"the discount factor gamma\"\n",
    "))\n",
    "\n",
    "parser.add_widget(widgets.IntText(\n",
    "    description='target-network-frequency',\n",
    "    value=500,\n",
    "    tooltip=\"the timesteps it takes to update the target network\"\n",
    "))\n",
    "\n",
    "parser.add_widget(widgets.IntText(\n",
    "    description='batch-size',\n",
    "    value=128,\n",
    "    tooltip=\"the batch size of sample from the reply memory\"\n",
    "))\n",
    "\n",
    "parser.add_widget(widgets.FloatText(\n",
    "    description='start-e',\n",
    "    value=1.0,\n",
    "    tooltip=\"the starting epsilon for exploration\"\n",
    "))\n",
    "\n",
    "parser.add_widget(widgets.FloatText(\n",
    "    description='end-e',\n",
    "    value=0.05,\n",
    "    tooltip=\"the ending epsilon for exploration\"\n",
    "))\n",
    "\n",
    "parser.add_widget(widgets.FloatText(\n",
    "    description='exploration-fraction',\n",
    "    value=0.5,\n",
    "    tooltip=\"the fraction of `total-timesteps` it takes from start-e to go end-e\"\n",
    "))\n",
    "\n",
    "parser.add_widget(widgets.IntText(\n",
    "    description='learning-starts',\n",
    "    value=10000,\n",
    "    tooltip=\"timestep to start learning\"\n",
    "))\n",
    "\n",
    "parser.add_widget(widgets.IntText(\n",
    "    description='train-frequency',\n",
    "    value=10,\n",
    "    tooltip=\"the frequency of training\"\n",
    "))\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f93fd2dc-1acd-4a98-85ab-40141ecff47b",
   "metadata": {},
   "source": [
    "### Parse inputs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "1b417f3a-206d-4624-85fb-62789374bb76",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-10-10T22:05:50.803507Z",
     "iopub.status.busy": "2022-10-10T22:05:50.803220Z",
     "iopub.status.idle": "2022-10-10T22:05:50.822476Z",
     "shell.execute_reply": "2022-10-10T22:05:50.821737Z",
     "shell.execute_reply.started": "2022-10-10T22:05:50.803487Z"
    },
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'using device cuda'"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "args = parser.parse()\n",
    "\n",
    "run_name = f\"{args.env_id}__dqn__{args.seed}__{int(time.time())}\"\n",
    "device = torch.device(\"cuda\" if torch.cuda.is_available() and args.cuda else \"cpu\")\n",
    "seed_everything(args.seed, args.torch_deterministic)\n",
    "display(f\"using device {device}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "49de7812-47cd-4b73-a015-fa3bc6decb72",
   "metadata": {},
   "source": [
    "## Setup environment"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "eebc8f65-d3ba-40f5-8c0e-a6486e3e5074",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-10-10T22:05:50.823720Z",
     "iopub.status.busy": "2022-10-10T22:05:50.823390Z",
     "iopub.status.idle": "2022-10-10T22:05:51.038049Z",
     "shell.execute_reply": "2022-10-10T22:05:51.037343Z",
     "shell.execute_reply.started": "2022-10-10T22:05:50.823673Z"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "import gym\n",
    "\n",
    "import numpy as np\n",
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "\n",
    "\n",
    "def make_env(env_id, seed, idx, capture_video, run_name):\n",
    "    def thunk():\n",
    "        env = gym.make(env_id)\n",
    "        env = gym.wrappers.RecordEpisodeStatistics(env)\n",
    "        if capture_video:\n",
    "            if idx == 0:\n",
    "                env = gym.wrappers.RecordVideo(env, f\"videos/{run_name}\")\n",
    "        env.seed(seed)\n",
    "        env.action_space.seed(seed)\n",
    "        env.observation_space.seed(seed)\n",
    "        return env\n",
    "\n",
    "    return thunk"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "72fffc90-d838-4b3d-96f0-683ea7ddd01d",
   "metadata": {},
   "source": [
    "### Ensure environment compatibility"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "cb7cba57-366e-412e-8eb6-42b9c7fec5ad",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-10-10T22:05:51.039221Z",
     "iopub.status.busy": "2022-10-10T22:05:51.038963Z",
     "iopub.status.idle": "2022-10-10T22:05:51.060346Z",
     "shell.execute_reply": "2022-10-10T22:05:51.059661Z",
     "shell.execute_reply.started": "2022-10-10T22:05:51.039202Z"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "envs = gym.vector.SyncVectorEnv([make_env(args.env_id, args.seed, 0, args.capture_video, run_name)])\n",
    "assert isinstance(envs.single_action_space, gym.spaces.Discrete), \"only discrete action space is supported\""
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6d229b6b-efac-43bd-818f-74c15f118f9b",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-10-03T03:17:34.254488Z",
     "iopub.status.busy": "2022-10-03T03:17:34.254047Z",
     "iopub.status.idle": "2022-10-03T03:17:34.270359Z",
     "shell.execute_reply": "2022-10-03T03:17:34.269512Z",
     "shell.execute_reply.started": "2022-10-03T03:17:34.254464Z"
    }
   },
   "source": [
    "### Environment examples"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "f888da2e-94d1-4af7-992d-4e45face59bb",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-10-10T22:05:51.061386Z",
     "iopub.status.busy": "2022-10-10T22:05:51.061158Z",
     "iopub.status.idle": "2022-10-10T22:05:51.573220Z",
     "shell.execute_reply": "2022-10-10T22:05:51.572584Z",
     "shell.execute_reply.started": "2022-10-10T22:05:51.061367Z"
    },
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "tmp_env = make_env(args.env_id, args.seed, 0, False, None)()\n",
    "ob = tmp_env.reset()\n",
    "\n",
    "plt.title(args.env_id)\n",
    "plt.imshow(tmp_env.render(\"rgb_array\"))\n",
    "plt.axis('off')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "1546dbcd-5820-40df-9b7d-528b791659ed",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-10-10T22:05:51.577919Z",
     "iopub.status.busy": "2022-10-10T22:05:51.577659Z",
     "iopub.status.idle": "2022-10-10T22:05:51.602310Z",
     "shell.execute_reply": "2022-10-10T22:05:51.601517Z",
     "shell.execute_reply.started": "2022-10-10T22:05:51.577896Z"
    },
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(1, array([ 0.03076804,  0.19700098, -0.03151444, -0.33357874], dtype=float32))"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "action = tmp_env.action_space.sample()\n",
    "ob, reward, done, info = tmp_env.step(action)\n",
    "action, ob"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "dbff4d3f-31f5-4dd6-91fe-088bc6252fd0",
   "metadata": {},
   "source": [
    "## Modeling"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "441336c2-97c3-49c9-b12a-21cfd4a06df3",
   "metadata": {},
   "source": [
    "### Define Neural Network"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "ba819a6e-0552-4d22-bcca-768c22651999",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-10-10T22:05:51.603453Z",
     "iopub.status.busy": "2022-10-10T22:05:51.603189Z",
     "iopub.status.idle": "2022-10-10T22:05:51.623291Z",
     "shell.execute_reply": "2022-10-10T22:05:51.622458Z",
     "shell.execute_reply.started": "2022-10-10T22:05:51.603433Z"
    }
   },
   "outputs": [],
   "source": [
    "class QNetwork(nn.Module):\n",
    "    def __init__(self, env):\n",
    "        super().__init__()\n",
    "        self.network = nn.Sequential(\n",
    "            nn.Linear(np.array(env.single_observation_space.shape).prod(), 120),\n",
    "            nn.ReLU(),\n",
    "            nn.Linear(120, 84),\n",
    "            nn.ReLU(),\n",
    "            nn.Linear(84, env.single_action_space.n),\n",
    "        )\n",
    "\n",
    "    def forward(self, x):\n",
    "        return self.network(x)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "398bf469-74ef-4f94-9218-8a94369e5f70",
   "metadata": {},
   "source": [
    "### DQN training Wrapper"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "b44085d8-b1d6-47a6-9c8a-11af1566a1e0",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-10-10T22:05:51.624543Z",
     "iopub.status.busy": "2022-10-10T22:05:51.624317Z",
     "iopub.status.idle": "2022-10-10T22:05:53.103239Z",
     "shell.execute_reply": "2022-10-10T22:05:53.102457Z",
     "shell.execute_reply.started": "2022-10-10T22:05:51.624522Z"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "from torch_rl.dqn import (\n",
    "    DQN,\n",
    "    DQNTrainingWrapper\n",
    ")\n",
    "\n",
    "\n",
    "dqn = DQN(\n",
    "    model_func=lambda: QNetwork(envs),\n",
    "    device=device\n",
    ")\n",
    "\n",
    "wrapper = DQNTrainingWrapper(\n",
    "    model = dqn,\n",
    "    env=envs,\n",
    "    device=device\n",
    ")\n",
    "\n",
    "optimizer = wrapper.get_optimizer(lr=args.learning_rate)\n",
    "\n",
    "rb = wrapper.get_replay_buffer(\n",
    "    buffer_size=args.buffer_size,\n",
    "    optimize_memory_usage=True\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ce5fbef7-e6ed-4a75-8874-2c22e784abbe",
   "metadata": {},
   "source": [
    "### Setup plots"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "89f00ff7-9e01-4fe9-90e0-4222d4e2a59e",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-10-10T22:05:53.104295Z",
     "iopub.status.busy": "2022-10-10T22:05:53.104069Z",
     "iopub.status.idle": "2022-10-10T22:05:53.125642Z",
     "shell.execute_reply": "2022-10-10T22:05:53.123883Z",
     "shell.execute_reply.started": "2022-10-10T22:05:53.104275Z"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "from torch_rl.utils import SPS\n",
    "def log_env_metrics(ms, step, infos):\n",
    "    for info in infos:\n",
    "        if \"episode\" in info.keys():\n",
    "            ms.log_metric(\"episodic_return\", info[\"episode\"][\"r\"], step=step)\n",
    "            ms.log_metric(\"episodic_length\", info[\"episode\"][\"l\"], step=step)\n",
    "\n",
    "\n",
    "def plot_metric(global_step, ms, sps):\n",
    "    metric_list = [\"episodic_return\", \"episodic_length\", \"q_values\", \"loss\"]\n",
    "    color_list = [\"blue\", \"black\", \"green\", \"red\"]\n",
    "    sps.step(global_step)\n",
    "    plot_metrics_notebook(ms, metric_list, color_list, sps=sps)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6525ae9e-6d5b-49d0-896e-3379315f2477",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-10-06T13:56:17.044309Z",
     "iopub.status.busy": "2022-10-06T13:56:17.043895Z",
     "iopub.status.idle": "2022-10-06T13:56:17.065030Z",
     "shell.execute_reply": "2022-10-06T13:56:17.064326Z",
     "shell.execute_reply.started": "2022-10-06T13:56:17.044259Z"
    },
    "tags": []
   },
   "source": [
    "### DQN training loop"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "684d7d00-db34-41d0-851a-e1f0ff1705e0",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-10-10T22:05:53.128867Z",
     "iopub.status.busy": "2022-10-10T22:05:53.127794Z",
     "iopub.status.idle": "2022-10-10T22:09:52.985649Z",
     "shell.execute_reply": "2022-10-10T22:09:52.984982Z",
     "shell.execute_reply.started": "2022-10-10T22:05:53.128803Z"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "100% processed at 2089 SPS. Remaining: 00 second(s) \n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 640x480 with 4 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "ms = MetricStore()\n",
    "sps = SPS(args.total_timesteps)\n",
    "obs = envs.reset()\n",
    "for global_step in range(args.total_timesteps):\n",
    "    epsilon = wrapper.linear_schedule(\n",
    "        args.start_e,\n",
    "        args.end_e,\n",
    "        args.exploration_fraction * args.total_timesteps,\n",
    "        global_step\n",
    "    )\n",
    "\n",
    "    if random.random() < epsilon:\n",
    "        actions = np.array([envs.single_action_space.sample()\n",
    "                            for _ in range(envs.num_envs)])\n",
    "    else:\n",
    "        actions = dqn(torch.Tensor(obs).to(device)).cpu().numpy()\n",
    "\n",
    "    next_obs, rewards, dones, infos = envs.step(actions)\n",
    "\n",
    "    next_real_obs = handle_terminal_ob(next_obs, dones, infos)\n",
    "    rb.add(obs, next_real_obs, actions, rewards, dones)\n",
    "    obs = next_obs\n",
    "\n",
    "    log_env_metrics(ms, global_step, infos)\n",
    "\n",
    "    if global_step > args.learning_starts and global_step % args.train_frequency == 0:\n",
    "        data = rb.sample(args.batch_size)\n",
    "        q_values, loss = wrapper.train_step(data, gamma=args.gamma)\n",
    "\n",
    "        if global_step % 100 == 0:\n",
    "            ms.log_metric(\"q_values\", q_values.mean().item(), step=global_step)\n",
    "            ms.log_metric(\"loss\", loss, step=global_step)\n",
    "\n",
    "        if global_step % 10000 == 0:\n",
    "            plot_metric(global_step, ms, sps)\n",
    "\n",
    "        optimizer.zero_grad()\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "\n",
    "        if global_step % args.target_network_frequency == 0:\n",
    "            dqn.update_target_network()\n",
    "\n",
    "plot_metric(global_step, ms, sps)\n",
    "envs.close()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "bd5908d7-ddb1-4f54-bebc-280b7e438fbf",
   "metadata": {},
   "source": [
    "## Evaluation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "becc5b71-4e81-448b-bc84-6e6c823a2bf5",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-10-10T22:09:52.986758Z",
     "iopub.status.busy": "2022-10-10T22:09:52.986502Z",
     "iopub.status.idle": "2022-10-10T22:09:56.695495Z",
     "shell.execute_reply": "2022-10-10T22:09:56.694680Z",
     "shell.execute_reply.started": "2022-10-10T22:09:52.986739Z"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "episodic_return=500.0, episodic_length=500\n"
     ]
    }
   ],
   "source": [
    "run_test_name = f\"test-{run_name}\"\n",
    "test_env = make_env(args.env_id, args.seed, 0, True, run_test_name)()\n",
    "\n",
    "ob = test_env.reset()\n",
    "while True:\n",
    "    action = dqn(torch.Tensor(np.array([ob])).to(device)).cpu().numpy()[0]\n",
    "    ob, reward, done, info = test_env.step(action)\n",
    "    if done:\n",
    "        print(f'episodic_return={info[\"episode\"][\"r\"]}, episodic_length={info[\"episode\"][\"l\"]}')\n",
    "        break\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "075734a3-6145-43dc-90a9-7fe28d62eb43",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-10-04T22:06:05.802908Z",
     "iopub.status.busy": "2022-10-04T22:06:05.802340Z",
     "iopub.status.idle": "2022-10-04T22:06:05.827471Z",
     "shell.execute_reply": "2022-10-04T22:06:05.826771Z",
     "shell.execute_reply.started": "2022-10-04T22:06:05.802857Z"
    }
   },
   "source": [
    "### Display full evaluation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "d74d7689-6cbd-400e-bb7f-547e9c2d73fa",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-10-10T22:09:56.696948Z",
     "iopub.status.busy": "2022-10-10T22:09:56.696680Z",
     "iopub.status.idle": "2022-10-10T22:10:09.339288Z",
     "shell.execute_reply": "2022-10-10T22:10:09.338673Z",
     "shell.execute_reply.started": "2022-10-10T22:09:56.696924Z"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Writing frame 500\n",
      "Finalizing...\n",
      "Done.\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<img src=\"images/dqn_test.gif\" width=\"500\" height=\"500\"/>"
      ],
      "text/plain": [
       "<IPython.core.display.Image object>"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mp4_path = f\"videos/{run_test_name}/rl-video-episode-0.mp4\"\n",
    "gif_path = f\"images/dqn_test.gif\"\n",
    "\n",
    "mp4_to_gif(mp4_path, gif_path)\n",
    "Image(url=gif_path, width=500, height=500)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "145ce381-a373-48bf-b9a6-c17bfe021dfb",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.8.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}