mxnet_DDQN_hybridize.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "mxnet_dqn.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/skywalker0803r/mxnet_course/blob/master/mxnet_DDQN_hybridize.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "RFBepfyUG8x1",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#!pip install mxnet\n",
        "import mxnet as mx\n",
        "import mxnet.ndarray as nd\n",
        "import mxnet.gluon as gluon\n",
        "from mxnet.gluon.loss import L2Loss\n",
        "import numpy as np\n",
        "from mxnet import init\n",
        "import mxnet.gluon.nn as nn\n",
        "import gym\n",
        "import matplotlib.pyplot as plt"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "EeauIatW2GBB",
        "colab_type": "text"
      },
      "source": [
        "# Hyper params"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Js1WdQzOzxMK",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "BATCH_SIZE = 256          \n",
        "LR = 0.001                                                       \n",
        "EPSILON = 0.95                                                \n",
        "GAMMA = 0.95                             \n",
        "TARGET_REPLACE_ITER = 1000                          \n",
        "MEMORY_CAPACITY = 10000\n",
        "l2loss = L2Loss(batch_axis=0)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "vjVqpXrb2IVA",
        "colab_type": "text"
      },
      "source": [
        "# env"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zgM-jjAg0Mm7",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "env_name = 'CartPole-v0'\n",
        "env = gym.make(env_name)                 \n",
        "N_ACTIONS = env.action_space.n                       \n",
        "N_STATES = env.observation_space.shape[0]    "
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "VfFZuUf32Jkn",
        "colab_type": "text"
      },
      "source": [
        "# Q_net"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "J1Ol2ir1M6Ty",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def build_Q_net(net,num_actions):\n",
        "  with net.name_scope():\n",
        "    net.add(nn.Dense(128,activation='relu'))\n",
        "    net.add(nn.Dense(128,activation='relu'))\n",
        "    net.add(nn.Dense(num_actions))\n",
        "  return net"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "v29u65XpNQrv",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "state = nd.random.normal(shape=(1,4))"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yDn8zU4666o5",
        "colab_type": "code",
        "outputId": "7dda4ad7-b77f-4781-e98a-c7263ce299fb",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 70
        }
      },
      "source": [
        "q = build_Q_net(nn.HybridSequential(),N_ACTIONS)\n",
        "q.initialize(init=init.Xavier(),force_reinit=True)\n",
        "q(state)"
      ],
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "\n",
              "[[-0.20866524 -0.40591076]]\n",
              "<NDArray 1x2 @cpu(0)>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 6
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "pXt_8Brv40VK",
        "colab_type": "text"
      },
      "source": [
        "# copy params"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "xWROrpMOCuwA",
        "colab_type": "code",
        "outputId": "ad6f4b1c-a2be-4154-93b5-5c7a8b101f47",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 230
        }
      },
      "source": [
        "q1 = build_Q_net(nn.HybridSequential(),N_ACTIONS)\n",
        "q1.initialize(init=init.Xavier())\n",
        "q2 = build_Q_net(nn.HybridSequential(),N_ACTIONS)\n",
        "q2.initialize(init=init.Xavier())\n",
        "\n",
        "# different\n",
        "print(q1(state))\n",
        "print(q2(state))\n",
        "\n",
        "# copy params\n",
        "q1.save_parameters(\"temp.params\")\n",
        "q2.load_parameters(\"temp.params\")\n",
        "\n",
        "# same\n",
        "print(q1(state))\n",
        "print(q2(state))"
      ],
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\n",
            "[[-0.31712627  0.0712393 ]]\n",
            "<NDArray 1x2 @cpu(0)>\n",
            "\n",
            "[[ 0.169029   -0.11568238]]\n",
            "<NDArray 1x2 @cpu(0)>\n",
            "\n",
            "[[-0.31712627  0.0712393 ]]\n",
            "<NDArray 1x2 @cpu(0)>\n",
            "\n",
            "[[-0.31712627  0.0712393 ]]\n",
            "<NDArray 1x2 @cpu(0)>\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "D5phbbLiGZTw",
        "colab_type": "text"
      },
      "source": [
        "# memory"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Ho5du-EZGDao",
        "colab_type": "code",
        "outputId": "19cd0b7a-1ada-40c2-c314-fb390f5a5322",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "s = [1,1]\n",
        "r = 1\n",
        "a = 1\n",
        "d = 1\n",
        "s_ = [1,1]\n",
        "transition = np.hstack((s,[a,r,d],s_))\n",
        "transition"
      ],
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([1, 1, 1, 1, 1, 1, 1])"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 8
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "vGiqYeS19N_0",
        "colab_type": "text"
      },
      "source": [
        "# class DQN"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yp4Us4jk7Kdj",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "class DQN(object):\n",
        "  def __init__(self,N_ACTIONS):\n",
        "    # eval_net\n",
        "    self.eval_net = build_Q_net(nn.HybridSequential(),N_ACTIONS)\n",
        "    self.eval_net.hybridize()\n",
        "    self.eval_net.initialize()#init=init.Xavier())\n",
        "    \n",
        "    # target_net\n",
        "    self.target_net = build_Q_net(nn.HybridSequential(),N_ACTIONS)\n",
        "    self.target_net.hybridize()\n",
        "    self.target_net.initialize()#init=init.Xavier())\n",
        "    \n",
        "    # initialize forward\n",
        "    random_x = nd.random_uniform(shape=(1,N_STATES))\n",
        "    _ = self.eval_net(random_x)\n",
        "    _ = self.target_net(random_x)\n",
        "\n",
        "    # initialize copy params\n",
        "    print('initialize copy params to target_net')\n",
        "    self.eval_net.save_parameters(\"temp.params\")\n",
        "    self.target_net.load_parameters(\"temp.params\")\n",
        "\n",
        "    # initialize memory\n",
        "    self.memory_counter = 0\n",
        "    self.memory = np.zeros(shape=(MEMORY_CAPACITY, N_STATES*2 + 3)) # state,[action,reward ,done],next_state N_STATES*2 +3\n",
        "    \n",
        "    # initialize trainer\n",
        "    self.trainer = gluon.Trainer(self.eval_net.collect_params(),'Adam',{'learning_rate':LR})\n",
        "\n",
        "    # record history\n",
        "    self.learn_step_counter = 0\n",
        "  \n",
        "  def choose_action(self,x,test=False):\n",
        "    # select best action\n",
        "    if (np.random.uniform() < EPSILON)|(test == True):\n",
        "      x = nd.array([x])\n",
        "      action_value = self.eval_net(x)\n",
        "      action = nd.argmax(action_value,axis=1).asscalar()\n",
        "      return int(action)\n",
        "    # select random action\n",
        "    else:\n",
        "      return np.random.randint(0,N_ACTIONS)\n",
        "  \n",
        "  def store_transition(self,s,a,r,d,s_):\n",
        "    # add transition to memory\n",
        "    transition = np.hstack((s,[a,r,d],s_))\n",
        "    index = self.memory_counter % MEMORY_CAPACITY\n",
        "    self.memory[index,:] = transition\n",
        "    self.memory_counter += 1\n",
        "  \n",
        "  def learn(self):\n",
        "    # record how many times of eval_net update\n",
        "    self.learn_step_counter += 1\n",
        "\n",
        "    # update target_net every TARGET_REPLACE_ITER\n",
        "    if self.learn_step_counter % TARGET_REPLACE_ITER == 0:\n",
        "      print('copy params to target_net')\n",
        "      self.eval_net.save_parameters(\"temp.params\")\n",
        "      self.target_net.load_parameters(\"temp.params\")\n",
        "    \n",
        "    # from memory sample a mini_batch_data\n",
        "    sample_index = np.random.choice(MEMORY_CAPACITY,BATCH_SIZE)\n",
        "    b_memory = self.memory[sample_index,:]\n",
        "    b_s = nd.array(b_memory[:,:N_STATES])\n",
        "    b_a = nd.array(b_memory[:,N_STATES:N_STATES+1])\n",
        "    b_r = nd.array(b_memory[:,N_STATES+1:N_STATES+2])\n",
        "    b_r = nd.squeeze(b_r)\n",
        "    b_d = nd.array(b_memory[:,N_STATES+2:N_STATES+3])\n",
        "    b_d = nd.squeeze(b_d)\n",
        "    b_s_= nd.array(b_memory[:,-N_STATES:])\n",
        "    \n",
        "    # caculate gradient\n",
        "    with mx.autograd.record():\n",
        "      # 1.target_net input b_s_(next_state) output q_sp\n",
        "      Q_sp = nd.max(self.target_net(b_s_),axis=1) # the max next_q_value\n",
        "      Q_sp = Q_sp * (nd.ones(BATCH_SIZE) - b_d) # if is done not have Q_sp\n",
        "      # 2.eval_net input b_s(state) output q_s_array\n",
        "      Q_s_array = self.eval_net(b_s) # Q_s_array\n",
        "      Q_s = nd.pick(Q_s_array,b_a,1) # the choose action's q_value\n",
        "      # count l2_loss\n",
        "      loss = nd.mean(l2loss(Q_s,(b_r + GAMMA*Q_sp)))\n",
        "    \n",
        "    # loss backward and trainer step\n",
        "    loss.backward()\n",
        "    self.trainer.step(BATCH_SIZE)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "obZskD3E-Y6B",
        "colab_type": "code",
        "outputId": "c44e3a8d-bc0d-420a-882b-f2f2eace0d79",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "all_rewards = []\n",
        "avg_rewards = []\n",
        "dqn = DQN(N_ACTIONS)\n",
        "for episode in range(1000):\n",
        "  s = env.reset()\n",
        "  rewards = []\n",
        "  while True:\n",
        "    # choose action\n",
        "    a = dqn.choose_action(s)\n",
        "    # do action on env\n",
        "    s_ ,r ,done ,info = env.step(a)\n",
        "    # store\n",
        "    rewards.append(r) \n",
        "    dqn.store_transition(s,a,r,int(done),s_)\n",
        "    # if have collect a lot of memory can learn\n",
        "    if dqn.memory_counter > BATCH_SIZE:\n",
        "      dqn.learn()\n",
        "    # if env done\n",
        "    if done:\n",
        "      break\n",
        "    # if env not done\n",
        "    else:\n",
        "      s = s_\n",
        "  # record this episode infomation \n",
        "  all_rewards.append(np.sum(rewards))\n",
        "  avg_rewards.append(np.mean(all_rewards[-10:]))\n",
        "  if episode % 10 == 0:\n",
        "    print(episode,avg_rewards[-1:])\n",
        "  if (all_rewards[-1] >= 200)&(avg_rewards[-1] >= 195):\n",
        "    print(episode,all_rewards[-1],avg_rewards[-1])\n",
        "    best_episode = episode\n",
        "    break"
      ],
      "execution_count": 10,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "initialize copy params to target_net\n",
            "0 [12.0]\n",
            "10 [26.4]\n",
            "20 [13.5]\n",
            "30 [10.5]\n",
            "40 [10.9]\n",
            "50 [12.3]\n",
            "60 [17.1]\n",
            "70 [17.8]\n",
            "copy params to target_net\n",
            "80 [18.1]\n",
            "90 [11.7]\n",
            "100 [13.5]\n",
            "110 [13.8]\n",
            "120 [13.0]\n",
            "130 [14.9]\n",
            "140 [15.3]\n",
            "150 [14.7]\n",
            "copy params to target_net\n",
            "160 [13.7]\n",
            "170 [23.2]\n",
            "180 [23.3]\n",
            "190 [21.3]\n",
            "copy params to target_net\n",
            "200 [30.5]\n",
            "210 [73.9]\n",
            "copy params to target_net\n",
            "220 [92.0]\n",
            "copy params to target_net\n",
            "copy params to target_net\n",
            "230 [145.2]\n",
            "copy params to target_net\n",
            "240 [173.8]\n",
            "copy params to target_net\n",
            "copy params to target_net\n",
            "250 [189.0]\n",
            "copy params to target_net\n",
            "copy params to target_net\n",
            "260 [164.1]\n",
            "copy params to target_net\n",
            "copy params to target_net\n",
            "270 [170.4]\n",
            "copy params to target_net\n",
            "280 [175.6]\n",
            "copy params to target_net\n",
            "copy params to target_net\n",
            "290 [184.4]\n",
            "copy params to target_net\n",
            "copy params to target_net\n",
            "300 [178.1]\n",
            "copy params to target_net\n",
            "copy params to target_net\n",
            "310 [144.8]\n",
            "copy params to target_net\n",
            "320 [117.9]\n",
            "copy params to target_net\n",
            "330 [131.7]\n",
            "copy params to target_net\n",
            "340 [133.3]\n",
            "copy params to target_net\n",
            "copy params to target_net\n",
            "350 [128.3]\n",
            "copy params to target_net\n",
            "360 [130.4]\n",
            "copy params to target_net\n",
            "370 [104.7]\n",
            "copy params to target_net\n",
            "380 [83.2]\n",
            "copy params to target_net\n",
            "390 [92.7]\n",
            "copy params to target_net\n",
            "400 [128.1]\n",
            "copy params to target_net\n",
            "410 [136.4]\n",
            "copy params to target_net\n",
            "copy params to target_net\n",
            "420 [129.9]\n",
            "copy params to target_net\n",
            "430 [172.1]\n",
            "copy params to target_net\n",
            "copy params to target_net\n",
            "440 [168.6]\n",
            "copy params to target_net\n",
            "copy params to target_net\n",
            "448 200.0 196.6\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "fItQxcRAW5Fr",
        "colab_type": "code",
        "outputId": "a40d7d52-498b-45d7-c8b6-956760dc5f1d",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 278
        }
      },
      "source": [
        "plt.plot(all_rewards)\n",
        "plt.plot(avg_rewards)\n",
        "plt.xlabel('Episode')\n",
        "plt.show()"
      ],
      "execution_count": 11,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXcAAAEGCAYAAACevtWaAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nA48M9qcAAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6nj4yla4y6wv",
        "colab_type": "text"
      },
      "source": [
        "# save model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "WqSgxz54kPjq",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "1c8a875b-27ef-4f02-8428-7f00a55d1857"
      },
      "source": [
        "save_path = 'env_{}_epoch_{}.params'.format(env_name,best_episode)\n",
        "dqn.eval_net.save_parameters(save_path)\n",
        "del dqn\n",
        "print('saved!')"
      ],
      "execution_count": 12,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "saved!\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "IUZZ2CjZztHI",
        "colab_type": "text"
      },
      "source": [
        "# load model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7s9PNTWnzgQc",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 52
        },
        "outputId": "30d30b97-0b21-4a0f-f28f-d3d5606a75ab"
      },
      "source": [
        "dqn = DQN(N_ACTIONS)\n",
        "dqn.eval_net.load_parameters(save_path)\n",
        "print('loaded!')"
      ],
      "execution_count": 13,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "initialize copy params to target_net\n",
            "loaded!\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "oVst45-Gzw-k",
        "colab_type": "text"
      },
      "source": [
        "# test model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pkAkAHi-zp3A",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 194
        },
        "outputId": "bed41c55-b39f-4686-f8c7-174a75c0d508"
      },
      "source": [
        "all_rewards = []\n",
        "for episode in range(10):\n",
        "  s = env.reset()\n",
        "  rewards = []\n",
        "  while True:\n",
        "    a = dqn.choose_action(s,test=True)\n",
        "    s_ ,r ,done ,info = env.step(a)\n",
        "    rewards.append(r) \n",
        "    if done:\n",
        "      break\n",
        "    s = s_ \n",
        "  all_rewards.append(np.sum(rewards))\n",
        "  print(episode,'score',all_rewards[-1:])"
      ],
      "execution_count": 16,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "0 score [200.0]\n",
            "1 score [200.0]\n",
            "2 score [200.0]\n",
            "3 score [200.0]\n",
            "4 score [200.0]\n",
            "5 score [200.0]\n",
            "6 score [200.0]\n",
            "7 score [200.0]\n",
            "8 score [200.0]\n",
            "9 score [200.0]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_Zp-5mKPz2Bb",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    }
  ]
}