[Atari Pong] Single-Agent Reinforcement Learning PPO.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "provenance": [],
      "gpuType": "T4",
      "authorship_tag": "ABX9TyN1fNakmbF39qDg8j4kQMmJ",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/kuds/rl-atari-pong/blob/main/%5BAtari%20Pong%5D%20Single-Agent%20Reinforcement%20Learning%20PPO.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# [Atari Pong] Single-Agent Reinforcement Learning PPO"
      ],
      "metadata": {
        "id": "ob-2hGT3jNdX"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "!pip install swig"
      ],
      "metadata": {
        "id": "Fy1WJ1R-iP4V",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "8f938bde-a088-4d84-94a6-5c4af55e4918"
      },
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Requirement already satisfied: swig in /usr/local/lib/python3.10/dist-packages (4.3.0)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "!pip install stable-baselines3"
      ],
      "metadata": {
        "id": "7U_qTvGlj2oy",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "3d5baae3-fce4-444a-879d-a092276d4bcf"
      },
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Requirement already satisfied: stable-baselines3 in /usr/local/lib/python3.10/dist-packages (2.4.0)\n",
            "Requirement already satisfied: gymnasium<1.1.0,>=0.29.1 in /usr/local/lib/python3.10/dist-packages (from stable-baselines3) (1.0.0)\n",
            "Requirement already satisfied: numpy<2.0,>=1.20 in /usr/local/lib/python3.10/dist-packages (from stable-baselines3) (1.26.4)\n",
            "Requirement already satisfied: torch>=1.13 in /usr/local/lib/python3.10/dist-packages (from stable-baselines3) (2.5.1+cu121)\n",
            "Requirement already satisfied: cloudpickle in /usr/local/lib/python3.10/dist-packages (from stable-baselines3) (3.1.0)\n",
            "Requirement already satisfied: pandas in /usr/local/lib/python3.10/dist-packages (from stable-baselines3) (2.2.2)\n",
            "Requirement already satisfied: matplotlib in /usr/local/lib/python3.10/dist-packages (from stable-baselines3) (3.8.0)\n",
            "Requirement already satisfied: typing-extensions>=4.3.0 in /usr/local/lib/python3.10/dist-packages (from gymnasium<1.1.0,>=0.29.1->stable-baselines3) (4.12.2)\n",
            "Requirement already satisfied: farama-notifications>=0.0.1 in /usr/local/lib/python3.10/dist-packages (from gymnasium<1.1.0,>=0.29.1->stable-baselines3) (0.0.4)\n",
            "Requirement already satisfied: filelock in /usr/local/lib/python3.10/dist-packages (from torch>=1.13->stable-baselines3) (3.16.1)\n",
            "Requirement already satisfied: networkx in /usr/local/lib/python3.10/dist-packages (from torch>=1.13->stable-baselines3) (3.4.2)\n",
            "Requirement already satisfied: jinja2 in /usr/local/lib/python3.10/dist-packages (from torch>=1.13->stable-baselines3) (3.1.4)\n",
            "Requirement already satisfied: fsspec in /usr/local/lib/python3.10/dist-packages (from torch>=1.13->stable-baselines3) (2024.10.0)\n",
            "Requirement already satisfied: sympy==1.13.1 in /usr/local/lib/python3.10/dist-packages (from torch>=1.13->stable-baselines3) (1.13.1)\n",
            "Requirement already satisfied: mpmath<1.4,>=1.1.0 in /usr/local/lib/python3.10/dist-packages (from sympy==1.13.1->torch>=1.13->stable-baselines3) (1.3.0)\n",
            "Requirement already satisfied: contourpy>=1.0.1 in /usr/local/lib/python3.10/dist-packages (from matplotlib->stable-baselines3) (1.3.1)\n",
            "Requirement already satisfied: cycler>=0.10 in /usr/local/lib/python3.10/dist-packages (from matplotlib->stable-baselines3) (0.12.1)\n",
            "Requirement already satisfied: fonttools>=4.22.0 in /usr/local/lib/python3.10/dist-packages (from matplotlib->stable-baselines3) (4.55.3)\n",
            "Requirement already satisfied: kiwisolver>=1.0.1 in /usr/local/lib/python3.10/dist-packages (from matplotlib->stable-baselines3) (1.4.7)\n",
            "Requirement already satisfied: packaging>=20.0 in /usr/local/lib/python3.10/dist-packages (from matplotlib->stable-baselines3) (24.2)\n",
            "Requirement already satisfied: pillow>=6.2.0 in /usr/local/lib/python3.10/dist-packages (from matplotlib->stable-baselines3) (11.0.0)\n",
            "Requirement already satisfied: pyparsing>=2.3.1 in /usr/local/lib/python3.10/dist-packages (from matplotlib->stable-baselines3) (3.2.0)\n",
            "Requirement already satisfied: python-dateutil>=2.7 in /usr/local/lib/python3.10/dist-packages (from matplotlib->stable-baselines3) (2.8.2)\n",
            "Requirement already satisfied: pytz>=2020.1 in /usr/local/lib/python3.10/dist-packages (from pandas->stable-baselines3) (2024.2)\n",
            "Requirement already satisfied: tzdata>=2022.7 in /usr/local/lib/python3.10/dist-packages (from pandas->stable-baselines3) (2024.2)\n",
            "Requirement already satisfied: six>=1.5 in /usr/local/lib/python3.10/dist-packages (from python-dateutil>=2.7->matplotlib->stable-baselines3) (1.17.0)\n",
            "Requirement already satisfied: MarkupSafe>=2.0 in /usr/local/lib/python3.10/dist-packages (from jinja2->torch>=1.13->stable-baselines3) (3.0.2)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "!pip install gymnasium[atari] ale-py"
      ],
      "metadata": {
        "id": "Le7dztPUkVz4",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "0334ca66-8ef5-4aa4-8466-55fad57a5300"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Requirement already satisfied: ale-py in /usr/local/lib/python3.10/dist-packages (0.10.1)\n",
            "Requirement already satisfied: gymnasium[atari] in /usr/local/lib/python3.10/dist-packages (1.0.0)\n",
            "Requirement already satisfied: numpy>=1.21.0 in /usr/local/lib/python3.10/dist-packages (from gymnasium[atari]) (1.26.4)\n",
            "Requirement already satisfied: cloudpickle>=1.2.0 in /usr/local/lib/python3.10/dist-packages (from gymnasium[atari]) (3.1.0)\n",
            "Requirement already satisfied: typing-extensions>=4.3.0 in /usr/local/lib/python3.10/dist-packages (from gymnasium[atari]) (4.12.2)\n",
            "Requirement already satisfied: farama-notifications>=0.0.1 in /usr/local/lib/python3.10/dist-packages (from gymnasium[atari]) (0.0.4)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "import os\n",
        "import csv\n",
        "import time\n",
        "import json\n",
        "import platform\n",
        "import torch\n",
        "import numpy\n",
        "import stable_baselines3\n",
        "from stable_baselines3 import PPO\n",
        "from stable_baselines3.common.env_util import make_atari_env\n",
        "from stable_baselines3.common.vec_env import VecFrameStack\n",
        "from stable_baselines3.common.evaluation import evaluate_policy\n",
        "from stable_baselines3.common.callbacks import EvalCallback\n",
        "from stable_baselines3.common.env_util import make_vec_env\n",
        "from stable_baselines3.common.callbacks import CallbackList\n",
        "from stable_baselines3.common.callbacks import BaseCallback\n",
        "from stable_baselines3.common.callbacks import CheckpointCallback\n",
        "from stable_baselines3.common.vec_env import VecVideoRecorder\n",
        "from stable_baselines3.common.vec_env import VecTransposeImage\n",
        "import matplotlib.pyplot\n",
        "import matplotlib\n",
        "import ale_py\n",
        "import gymnasium\n",
        "from importlib.metadata import version\n",
        "from datetime import datetime\n",
        "import google.colab.drive"
      ],
      "metadata": {
        "id": "o89JlEiexh1-"
      },
      "execution_count": 3,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "print(f\"Python Version: {platform.python_version()}\")\n",
        "print(f\"Torch Version: {version('torch')}\")\n",
        "print(f\"Is Cuda Available: {torch.cuda.is_available()}\")\n",
        "print(f\"Cuda Version: {torch.version.cuda}\")\n",
        "print(f\"Gymnasium Version: {version('gymnasium')}\")\n",
        "print(f\"Numpy Version: {version('numpy')}\")\n",
        "print(f\"Stable Baselines3 Version: {version('stable_baselines3')}\")\n",
        "print(f\"ALE Version: {version('ale_py')}\")"
      ],
      "metadata": {
        "id": "KRrJSPuDiDN4",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "c5c2d0d4-00f2-493d-a04e-c52d7bbabaf5"
      },
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Python Version: 3.10.12\n",
            "Torch Version: 2.5.1+cu121\n",
            "Is Cuda Available: True\n",
            "Cuda Version: 12.1\n",
            "Gymnasium Version: 1.0.0\n",
            "Numpy Version: 1.26.4\n",
            "Stable Baselines3 Version: 2.4.0\n",
            "ALE Version: 0.10.1\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.10/dist-packages/ipykernel/ipkernel.py:283: DeprecationWarning: `should_run_async` will not call `transform_cell` automatically in the future. Please pass the result to `transformed_cell` argument and any exception that happen during thetransform in `preprocessing_exc_tuple` in IPython 7.17 and above.\n",
            "  and should_run_async(code)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "gdrive_path = \"/content/gdrive\"\n",
        "google.colab.drive.mount(gdrive_path, force_remount=True)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "8WkaauEBoH_c",
        "outputId": "57c7ea9c-cd7d-4ba2-e51d-d41ef52e695b"
      },
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Mounted at /content/gdrive\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "rl_type = \"PPO\"\n",
        "env_str =  \"PongNoFrameskip-v4\" #\"ALE/Pong-v5\"\n",
        "name_prefix = \"atari_pong_ppo\"\n",
        "log_dir = \"{}/MyDrive/Finding Theta/logs/{}/{}\".format(gdrive_path,\n",
        "                                                       env_str,\n",
        "                                                       rl_type)\n",
        "training_data_path = os.path.join(log_dir, \"training jobs\")\n",
        "time_folder = datetime.now().strftime('%Y-%m-%d %H:%M:%S')\n",
        "model_folder_path = os.path.join(log_dir, \"training jobs\", time_folder)"
      ],
      "metadata": {
        "id": "aR7w3BqcFiIO"
      },
      "execution_count": 6,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "#Create Folders\n",
        "os.makedirs(log_dir, exist_ok=True)\n",
        "os.makedirs(training_data_path, exist_ok=True)\n",
        "os.makedirs(model_folder_path, exist_ok=True)"
      ],
      "metadata": {
        "id": "vC3zncEXIuBf"
      },
      "execution_count": 7,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "env = make_atari_env(env_str, n_envs=1, seed=0)\n",
        "print(\"Observation Space Size: \", env.observation_space.shape)\n",
        "print('Actions Space: ', env.action_space)\n",
        "env.close()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "T78JvLBVdmQO",
        "outputId": "3e6ee0f4-4ad5-4c93-90bc-625bc84c51d1"
      },
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Observation Space Size:  (84, 84, 1)\n",
            "Actions Space:  Discrete(6)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "env = gymnasium.make(env_str)\n",
        "print(\"Observation Space Size: \", env.observation_space.shape)\n",
        "print('Actions Space: ', env.action_space)\n",
        "env.close()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "5vCHAlsgg3lu",
        "outputId": "54ad5a7e-1c72-4606-f4fc-13d30229ec32"
      },
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Observation Space Size:  (210, 160, 3)\n",
            "Actions Space:  Discrete(6)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Use Hyperparameters from RL Zoo\n",
        "hyperparams = {\n",
        "    \"env_str\": env_str,\n",
        "    \"rl_type\": rl_type,\n",
        "    \"eval_freq\": 50_000,\n",
        "    \"total_timesteps\": 5_000_000,\n",
        "    \"clip_range\": 0.1,\n",
        "    \"learning_rate\": 2.5e-4,\n",
        "    \"n_steps\": 128,\n",
        "    \"n_epochs\": 4,\n",
        "    \"n_envs\": 8,\n",
        "    \"batch_size\": 256,\n",
        "    \"ent_coef\": 0.01,\n",
        "    \"vf_coef\": 0.5,\n",
        "}\n",
        "\n",
        "# https://github.com/DLR-RM/rl-baselines3-zoo/blob/master/hyperparams/ppo.yml\n",
        "# atari:\n",
        "#   env_wrapper:\n",
        "#     - stable_baselines3.common.atari_wrappers.AtariWrapper\n",
        "#   frame_stack: 4\n",
        "#   policy: 'CnnPolicy'\n",
        "#   n_envs: 8\n",
        "#   n_steps: 128\n",
        "#   n_epochs: 4\n",
        "#   batch_size: 256\n",
        "#   n_timesteps: !!float 1e7\n",
        "#   learning_rate: lin_2.5e-4\n",
        "#   clip_range: lin_0.1\n",
        "#   vf_coef: 0.5\n",
        "#   ent_coef: 0.01"
      ],
      "metadata": {
        "id": "V4PtNPT9Iye3"
      },
      "execution_count": 10,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Evaluation Environment Wrapper Arguments\n",
        "# Disable Frame Skipping\n",
        "# Disable Noop Max\n",
        "# https://danieltakeshi.github.io/2016/11/25/frame-skipping-and-preprocessing-for-deep-q-networks-on-atari-2600-games/\n",
        "# https://stable-baselines3.readthedocs.io/en/master/common/atari_wrappers.html\n",
        "\n",
        "# Training Aruguments\n",
        "env_wrap_args = {\"frame_skip\": 4}\n",
        "env_kwargs = {\"render_mode\": \"rgb_array\"}\n",
        "\n",
        "# Evaluation Aruguments\n",
        "env_val_wrap_args = {\"frame_skip\": 4, \"noop_max\": 30}\n",
        "env_val_kwargs = {\"render_mode\": \"rgb_array\"}\n",
        "\n",
        "if(env_str == \"ALE/Pong-v5\"):\n",
        "  env_wrap_args[\"frame_skip\"] = 0\n",
        "  env_val_kwargs[\"frameskip\"] = 32\n",
        "  env_val_kwargs[\"repeat_action_probability\"] = 0.0"
      ],
      "metadata": {
        "id": "ldok-8n-JZJe"
      },
      "execution_count": 11,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "with open(os.path.join(model_folder_path, 'parameters.json'), 'w') as fp:\n",
        "    parameters = {\"hyperparams\": hyperparams,\n",
        "                  \"env_wrap_args\": env_wrap_args,\n",
        "                  \"env_kwargs\": env_kwargs,\n",
        "                  \"env_val_wrap_args\": env_val_wrap_args,\n",
        "                  \"env_val_kwargs\": env_val_kwargs}\n",
        "    json.dump(parameters, fp, indent=4)"
      ],
      "metadata": {
        "id": "PXxtJfrvJbFM"
      },
      "execution_count": 12,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "class VideoRecordCallback(BaseCallback):\n",
        "    def __init__(\n",
        "        self,\n",
        "        save_path: str,\n",
        "        video_length: int,\n",
        "        save_freq: int = 5_000,\n",
        "        name_prefix: str =\"rl_model\",\n",
        "        verbose: int = 0):\n",
        "\n",
        "        super().__init__(verbose)\n",
        "        self.save_freq = save_freq\n",
        "        self.video_length = video_length\n",
        "        self.save_path = save_path\n",
        "        self.name_prefix = name_prefix\n",
        "        # Those variables will be accessible in the callback\n",
        "        # (they are defined in the base class)\n",
        "        # The RL model\n",
        "        # self.model = None  # type: BaseAlgorithm\n",
        "        # An alias for self.model.get_env(), the environment used for training\n",
        "        # self.training_env # type: VecEnv\n",
        "        # Number of time the callback was called\n",
        "        # self.n_calls = 0  # type: int\n",
        "        # num_timesteps = n_envs * n times env.step() was called\n",
        "        # self.num_timesteps = 0  # type: int\n",
        "        # local and global variables\n",
        "        # self.locals = {}  # type: Dict[str, Any]\n",
        "        # self.globals = {}  # type: Dict[str, Any]\n",
        "        # The logger object, used to report things in the terminal\n",
        "        # self.logger # type: stable_baselines3.common.logger.Logger\n",
        "        # Sometimes, for event callback, it is useful\n",
        "        # to have access to the parent object\n",
        "        # self.parent = None  # type: Optional[BaseCallback]\n",
        "\n",
        "    def _on_step(self) -> bool:\n",
        "        if self.n_calls % self.save_freq == 0:\n",
        "\n",
        "          name_prefix = f\"{self.name_prefix}_{self.num_timesteps}\"\n",
        "\n",
        "          # Record video of the best model playing Atari's Environment\n",
        "          rec_val = make_atari_env(\n",
        "              env_str,\n",
        "              n_envs=1,\n",
        "              seed=1)\n",
        "          rec_val = VecFrameStack(rec_val, n_stack=4)\n",
        "          rec_val = VecTransposeImage(rec_val)\n",
        "          rec_val = VecVideoRecorder(rec_val,\n",
        "                                    self.save_path,\n",
        "                                    video_length=self.video_length,\n",
        "                                    record_video_trigger=lambda x: x == 0,\n",
        "                                    name_prefix=name_prefix)\n",
        "\n",
        "          obs = rec_val.reset()\n",
        "          for _ in range(self.video_length):\n",
        "              action, _states = self.model.predict(obs)\n",
        "              obs, rewards, dones, info = rec_val.step(action)\n",
        "              rec_val.render()\n",
        "              if dones:\n",
        "                break\n",
        "\n",
        "          rec_val.close()\n",
        "        return True"
      ],
      "metadata": {
        "id": "HJ6lOtRoJc6f"
      },
      "execution_count": 13,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Create the Training Atari's Pong environment with appropriate wrappers\n",
        "env = make_atari_env(env_str,\n",
        "                     n_envs=hyperparams[\"n_envs\"],\n",
        "                     seed=0,\n",
        "                     monitor_dir=os.path.join(model_folder_path, \"monitor\"))\n",
        "env = VecFrameStack(env, n_stack=4)\n",
        "env = VecTransposeImage(env)\n",
        "\n",
        "# Create the Evaluation Atari's Pong environment with appropriate wrappers\n",
        "env_val = make_atari_env(env_str,\n",
        "                         n_envs=1,\n",
        "                         seed=1)\n",
        "env_val = VecFrameStack(env_val, n_stack=4)\n",
        "env_val = VecTransposeImage(env_val)"
      ],
      "metadata": {
        "id": "tYT5M7HxJfXg"
      },
      "execution_count": 14,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Create Callbacks\n",
        "# Create Evaluation Callback\n",
        "eval_callback = EvalCallback(\n",
        "    env_val,\n",
        "    best_model_save_path=model_folder_path,\n",
        "    log_path=model_folder_path,\n",
        "    eval_freq=hyperparams[\"eval_freq\"],\n",
        "    render=False,\n",
        "    deterministic=True,\n",
        "    n_eval_episodes=5)\n",
        "\n",
        "# Create Checkpoint Callback\n",
        "checkpoint_callback = CheckpointCallback(\n",
        "    save_freq=hyperparams[\"eval_freq\"],\n",
        "    save_path=os.path.join(model_folder_path, \"checkpoints\"),\n",
        "    name_prefix=name_prefix,\n",
        "    save_replay_buffer=False,\n",
        "    save_vecnormalize=False,\n",
        ")\n",
        "\n",
        "video_record_callback = VideoRecordCallback(\n",
        "    save_path=os.path.join(model_folder_path, \"videos\"),\n",
        "    video_length=2_500,\n",
        "    save_freq=hyperparams[\"eval_freq\"],\n",
        "    name_prefix=name_prefix)\n",
        "\n",
        "# Create the callback list\n",
        "callbackList = CallbackList([checkpoint_callback,\n",
        "                             video_record_callback,\n",
        "                             eval_callback])"
      ],
      "metadata": {
        "id": "1AAugcZGJhdn"
      },
      "execution_count": 15,
      "outputs": []
    },
    {
      "cell_type": "code",
      "execution_count": 16,
      "metadata": {
        "id": "SvPXWTCHiO69",
        "outputId": "025abd48-4c26-47d8-ff63-d5331f6aa9ed",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.10/dist-packages/moviepy/config_defaults.py:1: DeprecationWarning: invalid escape sequence '\\P'\n",
            "  \"\"\"\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Moviepy - Building video /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_400000-step-0-to-step-2500.mp4.\n",
            "Moviepy - Writing video /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_400000-step-0-to-step-2500.mp4\n",
            "\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": []
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Moviepy - Done !\n",
            "Moviepy - video ready /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_400000-step-0-to-step-2500.mp4\n",
            "Eval num_timesteps=400000, episode_reward=-21.00 +/- 0.00\n",
            "Episode length: 3056.00 +/- 0.00\n",
            "New best mean reward!\n",
            "Moviepy - Building video /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_800000-step-0-to-step-2500.mp4.\n",
            "Moviepy - Writing video /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_800000-step-0-to-step-2500.mp4\n",
            "\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": []
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Moviepy - Done !\n",
            "Moviepy - video ready /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_800000-step-0-to-step-2500.mp4\n",
            "Eval num_timesteps=800000, episode_reward=-12.00 +/- 4.00\n",
            "Episode length: 7438.40 +/- 831.20\n",
            "New best mean reward!\n",
            "Moviepy - Building video /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_1200000-step-0-to-step-2500.mp4.\n",
            "Moviepy - Writing video /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_1200000-step-0-to-step-2500.mp4\n",
            "\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": []
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Moviepy - Done !\n",
            "Moviepy - video ready /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_1200000-step-0-to-step-2500.mp4\n",
            "Eval num_timesteps=1200000, episode_reward=-9.40 +/- 14.22\n",
            "Episode length: 6580.80 +/- 145.15\n",
            "New best mean reward!\n",
            "Moviepy - Building video /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_1600000-step-0-to-step-2500.mp4.\n",
            "Moviepy - Writing video /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_1600000-step-0-to-step-2500.mp4\n",
            "\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": []
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Moviepy - Done !\n",
            "Moviepy - video ready /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_1600000-step-0-to-step-2500.mp4\n",
            "Eval num_timesteps=1600000, episode_reward=10.00 +/- 7.01\n",
            "Episode length: 9601.80 +/- 2345.16\n",
            "New best mean reward!\n",
            "Moviepy - Building video /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_2000000-step-0-to-step-2500.mp4.\n",
            "Moviepy - Writing video /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_2000000-step-0-to-step-2500.mp4\n",
            "\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": []
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Moviepy - Done !\n",
            "Moviepy - video ready /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_2000000-step-0-to-step-2500.mp4\n",
            "Eval num_timesteps=2000000, episode_reward=16.60 +/- 8.80\n",
            "Episode length: 7547.60 +/- 1874.22\n",
            "New best mean reward!\n",
            "Moviepy - Building video /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_2400000-step-0-to-step-2500.mp4.\n",
            "Moviepy - Writing video /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_2400000-step-0-to-step-2500.mp4\n",
            "\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": []
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Moviepy - Done !\n",
            "Moviepy - video ready /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_2400000-step-0-to-step-2500.mp4\n",
            "Eval num_timesteps=2400000, episode_reward=21.00 +/- 0.00\n",
            "Episode length: 6677.80 +/- 79.64\n",
            "New best mean reward!\n",
            "Moviepy - Building video /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_2800000-step-0-to-step-2500.mp4.\n",
            "Moviepy - Writing video /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_2800000-step-0-to-step-2500.mp4\n",
            "\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": []
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Moviepy - Done !\n",
            "Moviepy - video ready /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_2800000-step-0-to-step-2500.mp4\n",
            "Eval num_timesteps=2800000, episode_reward=21.00 +/- 0.00\n",
            "Episode length: 6674.60 +/- 82.30\n",
            "Moviepy - Building video /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_3200000-step-0-to-step-2500.mp4.\n",
            "Moviepy - Writing video /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_3200000-step-0-to-step-2500.mp4\n",
            "\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": []
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Moviepy - Done !\n",
            "Moviepy - video ready /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_3200000-step-0-to-step-2500.mp4\n",
            "Eval num_timesteps=3200000, episode_reward=21.00 +/- 0.00\n",
            "Episode length: 6608.60 +/- 9.33\n",
            "Moviepy - Building video /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_3600000-step-0-to-step-2500.mp4.\n",
            "Moviepy - Writing video /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_3600000-step-0-to-step-2500.mp4\n",
            "\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": []
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Moviepy - Done !\n",
            "Moviepy - video ready /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_3600000-step-0-to-step-2500.mp4\n",
            "Eval num_timesteps=3600000, episode_reward=21.00 +/- 0.00\n",
            "Episode length: 6681.80 +/- 76.10\n",
            "Moviepy - Building video /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_4000000-step-0-to-step-2500.mp4.\n",
            "Moviepy - Writing video /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_4000000-step-0-to-step-2500.mp4\n",
            "\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": []
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Moviepy - Done !\n",
            "Moviepy - video ready /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_4000000-step-0-to-step-2500.mp4\n",
            "Eval num_timesteps=4000000, episode_reward=20.20 +/- 1.60\n",
            "Episode length: 6816.20 +/- 400.47\n",
            "Moviepy - Building video /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_4400000-step-0-to-step-2500.mp4.\n",
            "Moviepy - Writing video /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_4400000-step-0-to-step-2500.mp4\n",
            "\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": []
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Moviepy - Done !\n",
            "Moviepy - video ready /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_4400000-step-0-to-step-2500.mp4\n",
            "Eval num_timesteps=4400000, episode_reward=21.00 +/- 0.00\n",
            "Episode length: 6620.20 +/- 0.98\n",
            "Moviepy - Building video /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_4800000-step-0-to-step-2500.mp4.\n",
            "Moviepy - Writing video /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_4800000-step-0-to-step-2500.mp4\n",
            "\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": []
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Moviepy - Done !\n",
            "Moviepy - video ready /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/atari_pong_ppo_4800000-step-0-to-step-2500.mp4\n",
            "Eval num_timesteps=4800000, episode_reward=21.00 +/- 0.00\n",
            "Episode length: 6616.20 +/- 7.65\n"
          ]
        }
      ],
      "source": [
        "# Initialize PPO\n",
        "model = PPO('CnnPolicy',\n",
        "            env,\n",
        "            verbose=0,\n",
        "            batch_size=hyperparams[\"batch_size\"],\n",
        "            learning_rate=hyperparams[\"learning_rate\"],\n",
        "            clip_range=hyperparams[\"clip_range\"],\n",
        "            n_steps=hyperparams[\"n_steps\"],\n",
        "            n_epochs=hyperparams[\"n_epochs\"],\n",
        "            tensorboard_log=os.path.join(log_dir, \"tensorboard\"))\n",
        "\n",
        "# Train the model\n",
        "model.learn(total_timesteps=hyperparams[\"total_timesteps\"],\n",
        "            progress_bar=False,\n",
        "            callback=callbackList)\n",
        "\n",
        "# Save the trained model\n",
        "model.save(os.path.join(model_folder_path, \"final_model\"))\n",
        "\n",
        "env.close()\n",
        "env_val.close()"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# # Get model path from last training job (uncomment if training job interrupted)\n",
        "# # List all entries in the directory\n",
        "# entries = os.listdir(training_data_path)\n",
        "\n",
        "# # Filter out only directories\n",
        "# folders = [entry for entry in entries if os.path.isdir(os.path.join(training_data_path, entry))]\n",
        "\n",
        "# # Sort the folders alphabetically\n",
        "# folders.sort()\n",
        "\n",
        "# # Get the last folder\n",
        "# model_folder_path = os.path.join(training_data_path, folders[-1])\n",
        "# print(model_folder_path)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "blB9bYzUJrlD",
        "outputId": "9bc9a100-bd40-4174-e537-3c17a6effa16"
      },
      "execution_count": 17,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.10/dist-packages/ipykernel/ipkernel.py:283: DeprecationWarning: `should_run_async` will not call `transform_cell` automatically in the future. Please pass the result to `transformed_cell` argument and any exception that happen during thetransform in `preprocessing_exc_tuple` in IPython 7.17 and above.\n",
            "  and should_run_async(code)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Create the Evaluation Atari's Pong environment with appropriate wrappers\n",
        "env_val = make_atari_env(env_str,\n",
        "                         n_envs=1,\n",
        "                         seed=1)\n",
        "env_val = VecFrameStack(env_val, n_stack=4)\n",
        "env_val = VecTransposeImage(env_val)\n",
        "\n",
        "# Load the best model\n",
        "best_model_path = os.path.join(model_folder_path, \"best_model\")\n",
        "best_model = PPO.load(best_model_path, env=env_val)\n",
        "\n",
        "mean_reward, std_reward = evaluate_policy(\n",
        "    best_model,\n",
        "    env_val,\n",
        "    n_eval_episodes=5)\n",
        "print(f\"Best Model - Mean reward: {mean_reward:.2f} +/- {std_reward:.2f}\")\n",
        "\n",
        "best_metrics_path = os.path.join(log_dir, \"best_model_metrics.csv\")\n",
        "\n",
        "# Create Best Model Metrics file if not there\n",
        "if(not os.path.isfile(best_metrics_path)):\n",
        "  with open(best_metrics_path, 'w', newline='') as csvfile:\n",
        "    writer = csv.writer(csvfile)\n",
        "    writer.writerow([\"run_date\",\n",
        "                     \"batch_size\",\n",
        "                     \"ent_coef\",\n",
        "                     \"learning_rate\",\n",
        "                     \"num_timesteps\",\n",
        "                     \"mean_reward\",\n",
        "                     \"std_reward\",\n",
        "                     \"n_steps\",\n",
        "                     \"n_envs\",\n",
        "                     \"gamma\",\n",
        "                     \"gae_lambda\",\n",
        "                     \"clip_range_vf\",\n",
        "                     \"n_epochs\"])\n",
        "\n",
        "new_data = [os.path.basename(os.path.normpath(model_folder_path)),\n",
        "            best_model.batch_size,\n",
        "            best_model.ent_coef,\n",
        "            best_model.learning_rate,\n",
        "            best_model.num_timesteps,\n",
        "            mean_reward,\n",
        "            std_reward,\n",
        "            best_model.n_steps,\n",
        "            best_model.n_envs,\n",
        "            best_model.gamma,\n",
        "            best_model.gae_lambda,\n",
        "            best_model.clip_range_vf,\n",
        "            best_model.n_epochs]\n",
        "\n",
        "with open(best_metrics_path, 'a', newline='') as csvfile:\n",
        "    writer = csv.writer(csvfile)\n",
        "    writer.writerow(new_data)\n",
        "\n",
        "# Record video of the best model playing Atari's Pong\n",
        "rec_val = VecVideoRecorder(env_val,\n",
        "                           os.path.join(model_folder_path, \"videos\"),\n",
        "                           video_length=5000,\n",
        "                           record_video_trigger=lambda x: x == 0,\n",
        "                           name_prefix=\"best_model_{}\".format(name_prefix))\n",
        "\n",
        "obs = rec_val.reset()\n",
        "for _ in range(5000):\n",
        "    action, _states = best_model.predict(obs)\n",
        "    obs, rewards, dones, info = rec_val.step(action)\n",
        "    rec_val.render()\n",
        "    if dones:\n",
        "      break\n",
        "\n",
        "env_val.close()\n",
        "rec_val.close()"
      ],
      "metadata": {
        "id": "OqwJT0bhcOu0",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "bdf04730-6036-4577-d1df-bb7136959091"
      },
      "execution_count": 18,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Best Model - Mean reward: 21.00 +/- 0.00\n",
            "Moviepy - Building video /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/best_model_atari_pong_ppo-step-0-to-step-5000.mp4.\n",
            "Moviepy - Writing video /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/best_model_atari_pong_ppo-step-0-to-step-5000.mp4\n",
            "\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "                                                                  "
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Moviepy - Done !\n",
            "Moviepy - video ready /content/gdrive/MyDrive/Finding Theta/logs/PongNoFrameskip-v4/PPO/training jobs/2024-12-17 21:33:55/videos/best_model_atari_pong_ppo-step-0-to-step-5000.mp4\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "\r"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Print Model\n",
        "print(best_model.policy)"
      ],
      "metadata": {
        "id": "2zN6WKnxGRSa",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "62db8e26-d581-4189-d12b-3b585e7a1433"
      },
      "execution_count": 19,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "ActorCriticCnnPolicy(\n",
            "  (features_extractor): NatureCNN(\n",
            "    (cnn): Sequential(\n",
            "      (0): Conv2d(4, 32, kernel_size=(8, 8), stride=(4, 4))\n",
            "      (1): ReLU()\n",
            "      (2): Conv2d(32, 64, kernel_size=(4, 4), stride=(2, 2))\n",
            "      (3): ReLU()\n",
            "      (4): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1))\n",
            "      (5): ReLU()\n",
            "      (6): Flatten(start_dim=1, end_dim=-1)\n",
            "    )\n",
            "    (linear): Sequential(\n",
            "      (0): Linear(in_features=3136, out_features=512, bias=True)\n",
            "      (1): ReLU()\n",
            "    )\n",
            "  )\n",
            "  (pi_features_extractor): NatureCNN(\n",
            "    (cnn): Sequential(\n",
            "      (0): Conv2d(4, 32, kernel_size=(8, 8), stride=(4, 4))\n",
            "      (1): ReLU()\n",
            "      (2): Conv2d(32, 64, kernel_size=(4, 4), stride=(2, 2))\n",
            "      (3): ReLU()\n",
            "      (4): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1))\n",
            "      (5): ReLU()\n",
            "      (6): Flatten(start_dim=1, end_dim=-1)\n",
            "    )\n",
            "    (linear): Sequential(\n",
            "      (0): Linear(in_features=3136, out_features=512, bias=True)\n",
            "      (1): ReLU()\n",
            "    )\n",
            "  )\n",
            "  (vf_features_extractor): NatureCNN(\n",
            "    (cnn): Sequential(\n",
            "      (0): Conv2d(4, 32, kernel_size=(8, 8), stride=(4, 4))\n",
            "      (1): ReLU()\n",
            "      (2): Conv2d(32, 64, kernel_size=(4, 4), stride=(2, 2))\n",
            "      (3): ReLU()\n",
            "      (4): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1))\n",
            "      (5): ReLU()\n",
            "      (6): Flatten(start_dim=1, end_dim=-1)\n",
            "    )\n",
            "    (linear): Sequential(\n",
            "      (0): Linear(in_features=3136, out_features=512, bias=True)\n",
            "      (1): ReLU()\n",
            "    )\n",
            "  )\n",
            "  (mlp_extractor): MlpExtractor(\n",
            "    (policy_net): Sequential()\n",
            "    (value_net): Sequential()\n",
            "  )\n",
            "  (action_net): Linear(in_features=512, out_features=6, bias=True)\n",
            "  (value_net): Linear(in_features=512, out_features=1, bias=True)\n",
            ")\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Load the evaluations.npz file\n",
        "data = numpy.load(os.path.join(model_folder_path, \"evaluations.npz\"))\n",
        "\n",
        "# Extract the relevant data\n",
        "timesteps = data['timesteps']\n",
        "results = data['results']\n",
        "\n",
        "# Calculate the mean and standard deviation of the results\n",
        "mean_results = numpy.mean(results, axis=1)\n",
        "std_results = numpy.std(results, axis=1)\n",
        "\n",
        "# Plot the results\n",
        "matplotlib.pyplot.figure()\n",
        "matplotlib.pyplot.plot(timesteps, mean_results)\n",
        "matplotlib.pyplot.fill_between(timesteps,\n",
        "                               mean_results - std_results,\n",
        "                               mean_results + std_results,\n",
        "                               alpha=0.3)\n",
        "\n",
        "matplotlib.pyplot.xlabel('Timesteps')\n",
        "matplotlib.pyplot.ylabel('Mean Reward')\n",
        "matplotlib.pyplot.title(f\"{rl_type} Performance on {env_str}\")\n",
        "matplotlib.pyplot.show()"
      ],
      "metadata": {
        "id": "Xisgp9SfcONO",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 472
        },
        "outputId": "39df3787-5e34-446c-97c0-cbd0151d1c38"
      },
      "execution_count": 20,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    }
  ]
}