my_robot.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 导入相关包 \n",
    "import random\n",
    "import numpy as np\n",
    "import torch\n",
    "from QRobot import QRobot\n",
    "from ReplayDataSet import ReplayDataSet\n",
    "from torch_py.MinDQNRobot import MinDQNRobot as TorchRobot # PyTorch版本\n",
    "import matplotlib.pyplot as plt\n",
    "from Maze import Maze\n",
    "import time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Robot(TorchRobot):\n",
    "    def __init__(self, maze):\n",
    "        \"\"\"\n",
    "        初始化 Robot 类\n",
    "        :param maze:迷宫对象\n",
    "        \"\"\"\n",
    "        super(Robot, self).__init__(maze)\n",
    "        \n",
    "        # destination 设置为和迷宫大小相关，为了在足够大的迷宫中，机器人也能「看得到」终点。\n",
    "        maze.set_reward(reward={\n",
    "            \"hit_wall\": 10.,\n",
    "            \"destination\": -maze.maze_size ** 2 * 4.,\n",
    "            \"default\": 1.,\n",
    "        })\n",
    "        self.maze = maze\n",
    "        self.epsilon = 0\n",
    "        \"\"\"开启金手指，获取全图视野\"\"\"\n",
    "        self.memory.build_full_view(maze=maze)\n",
    "        \n",
    "        # 初始化后即开始训练\n",
    "        self.loss_list = self.train()\n",
    "        \n",
    "\n",
    "    def train(self):\n",
    "        loss_list = []\n",
    "        batch_size = len(self.memory)\n",
    "        \n",
    "        start = time.time()\n",
    "        # 训练，直到能走出这个迷宫\n",
    "        while True:\n",
    "            loss = self._learn(batch=batch_size)\n",
    "            loss_list.append(loss)\n",
    "            self.reset()\n",
    "            for _ in range(self.maze.maze_size ** 2 - 1):\n",
    "                a, r = self.test_update()\n",
    "                if r == self.maze.reward[\"destination\"]:\n",
    "                    print('Training time: {:.2f} s'.format(time.time() - start))\n",
    "                    return loss_list\n",
    "\n",
    "\n",
    "    def train_update(self):\n",
    "        state = self.sense_state()\n",
    "        action = self._choose_action(state)\n",
    "        reward = self.maze.move_robot(action)\n",
    "\n",
    "        \"\"\"---update the step and epsilon---\"\"\"\n",
    "        # self.epsilon = max(0.01, self.epsilon * 0.995)\n",
    "\n",
    "        return action, reward\n",
    "    \n",
    "    \n",
    "    def test_update(self):\n",
    "        state = np.array(self.sense_state(), dtype=np.int16)\n",
    "        state = torch.from_numpy(state).float().to(self.device)\n",
    "\n",
    "        self.eval_model.eval()\n",
    "        with torch.no_grad():\n",
    "            q_value = self.eval_model(state).cpu().data.numpy()\n",
    "\n",
    "        action = self.valid_action[np.argmin(q_value).item()]\n",
    "        reward = self.maze.move_robot(action)\n",
    "        return action, reward"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Maze of size (6, 6)\n",
      "Training time: 1.72 s\n",
      "{'hit_wall': 10.0, 'destination': -144.0, 'default': 1.0}\n",
      "action: d reward: 1.0\n",
      "action: d reward: 1.0\n",
      "action: r reward: 1.0\n",
      "action: r reward: 1.0\n",
      "action: d reward: 1.0\n",
      "action: r reward: 1.0\n",
      "action: d reward: 1.0\n",
      "action: d reward: 1.0\n",
      "action: r reward: 1.0\n",
      "action: r reward: -144.0\n",
      "success\n"
     ]
    }
   ],
   "source": [
    "maze = Maze(maze_size=6) \n",
    "print(maze)\n",
    "\n",
    "robot = Robot(maze=maze)\n",
    "\n",
    "print(robot.maze.reward) # 输出最小值选择策略的reward值\n",
    "\n",
    "\"\"\"Test Robot\"\"\"\n",
    "robot.reset()\n",
    "for _ in range(maze.maze_size ** 2 - 1):\n",
    "    a, r = robot.test_update()\n",
    "    print(\"action:\", a, \"reward:\", r)\n",
    "    if r == maze.reward[\"destination\"]:\n",
    "        print(\"success\")\n",
    "        break"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x126c434d0>]"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "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": [
    "# 绘制损失曲线\n",
    "loss_list = robot.loss_list\n",
    "n = len(loss_list)\n",
    "plt.plot(range(n), loss_list)"
   ]
  },
  {
   "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": "3.7.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}