mxnet_ppo_lunarlander-v2.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "mxnet_ppo_test.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_ppo_lunarlander-v2.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9FiaX_ohLvWD",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#!pip install mxnet\n",
        "import numpy as np \n",
        "import mxnet as mx \n",
        "from mxnet import gluon, autograd, nd\n",
        "import mxnet.ndarray as F\n",
        "from mxnet.gluon import nn\n",
        "from mxnet.gluon.loss import L2Loss\n",
        "import gym\n",
        "from mxnet import init\n",
        "import matplotlib.pyplot as plt"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "NET-eqwPgkq_",
        "colab_type": "text"
      },
      "source": [
        "# some function"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "uakkFrEad24J",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def nd_entropy(probs):\n",
        "  return -nd.sum(probs*nd.log(probs+1e-8),axis=1)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6LIRe6hjgmjV",
        "colab_type": "text"
      },
      "source": [
        "# memory"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Y6KC6RytdufZ",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "class Memory:\n",
        "  def __init__(self):\n",
        "    self.actions = []\n",
        "    self.states = []\n",
        "    self.logprobs = []\n",
        "    self.rewards = []\n",
        "    self.is_terminals = []\n",
        "  def clear_memory(self):\n",
        "    del self.actions[:]\n",
        "    del self.states[:]\n",
        "    del self.logprobs[:]\n",
        "    del self.rewards[:]\n",
        "    del self.is_terminals[:]"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Npwuiud-gn71",
        "colab_type": "text"
      },
      "source": [
        "# model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yFy1rzTIL5is",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "class ActorCritic_Discrete(gluon.Block):\n",
        "  def __init__(self,action_dim):\n",
        "    super(ActorCritic_Discrete, self).__init__()\n",
        "    self.action_dim = action_dim\n",
        "    with self.name_scope():\n",
        "      self.probs = nn.Sequential()\n",
        "      self.probs.add(nn.Dense(64,activation='relu'))\n",
        "      self.probs.add(nn.Dense(64,activation='relu'))\n",
        "      self.probs.add(nn.Dense(64,activation='relu'))\n",
        "      self.probs.add(nn.Dense(action_dim))\n",
        "      \n",
        "      self.value = nn.Sequential()\n",
        "      self.value.add(nn.Dense(64,activation='relu'))\n",
        "      self.value.add(nn.Dense(64,activation='relu'))\n",
        "      self.value.add(nn.Dense(64,activation='relu'))\n",
        "      self.value.add(nn.Dense(1))       \n",
        "  \n",
        "  def choose_action(self, state, memory):\n",
        "    state = nd.array(state)\n",
        "    probs = nd.softmax(self.probs(state))\n",
        "    action, log_prob = nd.sample_multinomial(probs, get_prob = True)\n",
        "    memory.states.append(list(nd.squeeze(state).asnumpy())) \n",
        "    memory.actions.append(action.asnumpy()[0])\n",
        "    memory.logprobs.append(log_prob.asnumpy()[0])\n",
        "    return action.asnumpy()[0]\n",
        "  \n",
        "  def evaluate(self, state, action):\n",
        "    state = nd.array(state)\n",
        "    action = nd.array(action)\n",
        "    probs = nd.softmax(self.probs(state))\n",
        "    action_logprobs = nd.pick(probs.log(),action,axis=1)\n",
        "    state_value = self.value(state)\n",
        "    return action_logprobs, nd.squeeze(state_value), nd_entropy(probs) "
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "QxNAy1W-g5gm",
        "colab_type": "text"
      },
      "source": [
        "# ppo"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "AjBVLp_6YhaR",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "class PPO_Discrete(object):\n",
        "  def __init__(self, action_dim, lr, gamma):\n",
        "    \n",
        "    # initialize policy\n",
        "    self.policy = ActorCritic_Discrete(action_dim)\n",
        "    self.policy.probs.initialize(init=init.Xavier())\n",
        "    self.policy.value.initialize(init=init.Xavier())\n",
        "    \n",
        "    # initialize old policy\n",
        "    self.policy_old = ActorCritic_Discrete(action_dim)\n",
        "    self.policy_old.probs.initialize(init=init.Xavier())\n",
        "    self.policy_old.value.initialize(init=init.Xavier())\n",
        "\n",
        "    # initialize forward 4 net\n",
        "    random_x = nd.random_uniform(shape=(1,state_dim))\n",
        "    _ = self.policy.probs(random_x)\n",
        "    _ = self.policy.value(random_x)\n",
        "    _ = self.policy_old.probs(random_x)\n",
        "    _ = self.policy_old.value(random_x)\n",
        "    \n",
        "    # Copy policy params to old policy:\n",
        "    self.policy.probs.save_parameters(\"temp_probs.params\")\n",
        "    self.policy.value.save_parameters(\"temp_value.params\")\n",
        "    self.policy_old.probs.load_parameters(\"temp_probs.params\")\n",
        "    self.policy_old.value.load_parameters(\"temp_value.params\")\n",
        "    print('initialize copy params to old policy')\n",
        "    \n",
        "    self.lr = lr\n",
        "    self.gamma = gamma\n",
        "    self.trainer = gluon.Trainer(self.policy.collect_params(),'adam',{'learning_rate':self.lr})\n",
        "\n",
        "  def update(self, memory): \n",
        "    \n",
        "    # Monte Carlo estimate of state rewards:\n",
        "    rewards = []\n",
        "    discounted_reward = 0\n",
        "    for reward, is_terminal in zip(reversed(memory.rewards), reversed(memory.is_terminals)):\n",
        "      if is_terminal:\n",
        "        discounted_reward = 0\n",
        "      discounted_reward = reward + (self.gamma * discounted_reward)\n",
        "      rewards.insert(0, discounted_reward)\n",
        "    \n",
        "    # Normalizing the rewards:\n",
        "    rewards = np.array(rewards)\n",
        "    rewards = (rewards - rewards.mean()) / (rewards.std() + 1e-5)\n",
        "    \n",
        "    # sample batch data from memory\n",
        "    old_states = memory.states\n",
        "    old_actions = memory.actions\n",
        "    old_logprobs = memory.logprobs\n",
        "\n",
        "    # to ndarray\n",
        "    rewards = nd.array(rewards)\n",
        "    old_logprobs = nd.array(old_logprobs)\n",
        "\n",
        "    # Optimize policy for 4 epochs:\n",
        "    for _ in range(4):\n",
        "      with autograd.record():    \n",
        "        logprobs, state_values, dist_entropy = self.policy.evaluate(old_states, old_actions)\n",
        "        ratios = nd.exp(logprobs - old_logprobs)\n",
        "        advantages = rewards - state_values\n",
        "        surr1 = ratios * advantages\n",
        "        surr2 = nd.clip(ratios, 0.8, 1.2)*advantages\n",
        "        loss = -nd.minimum(surr1,surr2) + 0.5*(advantages)**2 - 0.01*dist_entropy\n",
        "        loss = nd.mean(loss)\n",
        "      # loss backward and trainer step\n",
        "      loss.backward()\n",
        "      self.trainer.step(1)\n",
        "    \n",
        "    # Copy policy params to old policy:\n",
        "    self.policy.probs.save_parameters(\"temp_probs.params\")\n",
        "    self.policy.value.save_parameters(\"temp_value.params\")\n",
        "    self.policy_old.probs.load_parameters(\"temp_probs.params\")\n",
        "    self.policy_old.value.load_parameters(\"temp_value.params\")\n",
        "    print('copy params to old policy')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "E2gVFfDErj2r",
        "colab_type": "text"
      },
      "source": [
        "# env and params"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "obr88nc6d-9i",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#!pip install box2d-py"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "JDpJoQBAfU0b",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "env_name = \"LunarLander-v2\"\n",
        "env = gym.make(env_name)\n",
        "state_dim = env.observation_space.shape[0]\n",
        "action_dim = env.action_space.n\n",
        "solved_reward = 200         # stop training if avg_reward > solved_reward\n",
        "log_interval = 20           # print avg reward in the interval\n",
        "max_episodes = 50000        # max training episodes\n",
        "max_timesteps = 300         # max timesteps in one episode\n",
        "update_timestep = 2000      # update policy every n timesteps\n",
        "lr = 0.002\n",
        "gamma = 0.99                # discount factor"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "oNlWFrt0qOOH",
        "colab_type": "text"
      },
      "source": [
        "# main"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "hgMq9T-lEkXq",
        "colab_type": "code",
        "outputId": "ac90ed8d-605a-4552-e448-5e6125330128",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "memory = Memory()\n",
        "ppo = PPO_Discrete(action_dim,lr=0.001,gamma=0.99)"
      ],
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "initialize copy params to old policy\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3Y0fjmcimvg1",
        "colab_type": "code",
        "outputId": "55c6759d-9567-4bd7-e708-ce81f90ca227",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "all_rewards = []\n",
        "avg_rewards = []\n",
        "time_step = 0    \n",
        "\n",
        "# run 50000 episode\n",
        "for i_episode in range(max_episodes):\n",
        "  # reset env and R(total_reward)\n",
        "  state = env.reset()\n",
        "  R = 0\n",
        "  # game start \n",
        "  for t in range(max_timesteps):\n",
        "    time_step +=1\n",
        "    # transform state to nd_array\n",
        "    state = nd.array(np.expand_dims(state, 0))\n",
        "    # choose action\n",
        "    action = ppo.policy_old.choose_action(state, memory)\n",
        "    # do action on env\n",
        "    state, reward, done, _ = env.step(action)\n",
        "    \n",
        "    # append to memory\n",
        "    R += reward\n",
        "    memory.rewards.append(reward)\n",
        "    memory.is_terminals.append(done)\n",
        "    \n",
        "    # every update_timestep ppo learn\n",
        "    if time_step % update_timestep == 0:\n",
        "      ppo.update(memory)\n",
        "      memory.clear_memory()\n",
        "      time_step = 0\n",
        "    \n",
        "    # game over\n",
        "    if done:\n",
        "      break            \n",
        "  \n",
        "  # every episode record R\n",
        "  all_rewards.append(R)\n",
        "  avg_rewards.append(np.mean(all_rewards[-20:]))\n",
        "  \n",
        "  # if solved!\n",
        "  if (all_rewards[-1] >= solved_reward)&(avg_rewards[-1] >= solved_reward*0.9):\n",
        "    print(\"########## Solved! ##########\")\n",
        "    print(i_episode,all_rewards[-1],avg_rewards[-1])\n",
        "    break\n",
        "  \n",
        "  # print infomation\n",
        "  if i_episode % log_interval == 0:\n",
        "    print(i_episode,avg_rewards[-1])"
      ],
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "0 -366.2525621862741\n",
            "20 -195.18966624749888\n",
            "copy params to old policy\n",
            "40 -170.09886888417105\n",
            "copy params to old policy\n",
            "60 -147.45566287692998\n",
            "copy params to old policy\n",
            "80 -157.30789764110796\n",
            "copy params to old policy\n",
            "100 -156.81572798974437\n",
            "copy params to old policy\n",
            "120 -157.06864624312112\n",
            "copy params to old policy\n",
            "140 -134.73091970344313\n",
            "copy params to old policy\n",
            "160 -117.42445366828277\n",
            "180 -106.65086888358269\n",
            "copy params to old policy\n",
            "200 -108.54087362746013\n",
            "copy params to old policy\n",
            "220 -128.0439613674883\n",
            "copy params to old policy\n",
            "240 -126.32195430801843\n",
            "copy params to old policy\n",
            "260 -106.84009733270291\n",
            "copy params to old policy\n",
            "280 -118.51429764073882\n",
            "copy params to old policy\n",
            "300 -117.75476145680143\n",
            "copy params to old policy\n",
            "320 -95.99123377636461\n",
            "copy params to old policy\n",
            "340 -90.4985236759626\n",
            "copy params to old policy\n",
            "360 -80.71354676117645\n",
            "copy params to old policy\n",
            "380 -121.33168555563529\n",
            "copy params to old policy\n",
            "400 -108.85206838306677\n",
            "copy params to old policy\n",
            "420 -85.52837733857103\n",
            "copy params to old policy\n",
            "440 -88.91846799296613\n",
            "copy params to old policy\n",
            "460 -105.46816316438287\n",
            "copy params to old policy\n",
            "480 -87.56474833809376\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "500 -83.50519486624596\n",
            "copy params to old policy\n",
            "520 -68.10689141732429\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "540 -47.13096916334409\n",
            "copy params to old policy\n",
            "560 -50.84582660520407\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "580 -41.82504021742943\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "600 4.907144188353863\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "620 13.088976585549307\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "640 7.464323988932942\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "660 27.272040305882946\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "680 52.20704255086436\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "700 67.378024867471\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "720 75.88930133608639\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "740 53.4019494932269\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "760 81.53692860516428\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "780 95.69526561083141\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "800 50.35224682893475\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "820 58.165443104702135\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "840 66.76961986042699\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "860 102.92328454381845\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "880 86.2998415008142\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "900 84.21071910234227\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "920 62.627321728269365\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "940 109.03477389359414\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "960 106.97245760361336\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "980 127.63301256955285\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "1000 107.79731854801238\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "1020 133.15515754711768\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "1040 119.39990786324461\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "1060 105.30227754585894\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "1080 103.82569766195427\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "1100 109.12769066808846\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "1120 101.60538916424181\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "1140 104.87725301576063\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "1160 138.43386294397115\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "1180 157.72909784174345\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "copy params to old policy\n",
            "1200 178.10139166368197\n",
            "########## Solved! ##########\n",
            "1204 290.2452440672386 180.39743314588424\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "tkHdkPFtUZOp",
        "colab_type": "text"
      },
      "source": [
        "# plot"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Wm8DVdTB2Lmy",
        "colab_type": "code",
        "outputId": "d9eab4b4-2597-44fa-c6ee-f406f22543b8",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 279
        }
      },
      "source": [
        "plt.plot(all_rewards)\n",
        "plt.plot(avg_rewards)\n",
        "plt.xlabel('episode')\n",
        "plt.show()"
      ],
      "execution_count": 10,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAX8AAAEGCAYAAACNaZVuAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAzn4m5l1IAd/M7MO5OBvZtaB/h+rZ93bPsfj7wAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "VKpVyJNrhev_",
        "colab_type": "text"
      },
      "source": [
        "# test"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lWl3S8NAhe6j",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 185
        },
        "outputId": "501e75f4-fbb3-4c47-f45f-b078da31bb93"
      },
      "source": [
        "for i_episode in range(10):\n",
        "  state = env.reset()\n",
        "  R = 0\n",
        "  for t in range(1000):\n",
        "    state = nd.array(np.expand_dims(state, 0))\n",
        "    action = ppo.policy_old.choose_action(state, memory)\n",
        "    state, reward, done, _ = env.step(action)\n",
        "    R += reward\n",
        "    if done:\n",
        "      break            \n",
        "  print(i_episode,R)"
      ],
      "execution_count": 11,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "0 238.35654848244386\n",
            "1 237.83271118369947\n",
            "2 238.39163854443348\n",
            "3 265.74826492192176\n",
            "4 305.79812009942265\n",
            "5 66.93421217443262\n",
            "6 223.20954969688802\n",
            "7 264.4166658820544\n",
            "8 286.6863565646854\n",
            "9 21.416109780089286\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-lNsvJstUo7l",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    }
  ]
}