SAiDL_RL.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "SAiDL RL.ipynb",
      "provenance": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "cells": [
    {
      "cell_type": "code",
      "metadata": {
        "id": "ijHUgddSrUjy"
      },
      "source": [
        "import torch\n",
        "import torch.nn as nn\n",
        "from torch.utils.data import Dataset,DataLoader\n",
        "import torchvision.datasets as datasets\n",
        "import torchvision.transforms as transforms\n",
        "if (torch.cuda.is_available()):\n",
        "  device = torch.device('cuda')\n",
        "else:\n",
        "  device = torch.device('cpu')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DZflPjuZsBrk"
      },
      "source": [
        "import gym\n",
        "from gym import error, spaces, utils\n",
        "from gym.utils import seeding\n",
        "import random\n",
        "class FooEnv(gym.Env):\n",
        "  def __init__(self,test = False):\n",
        "    super().__init__()\n",
        "    self.coefficients = (torch.rand([6])-0.5)*60\n",
        "    self.x = torch.rand([1])*100\n",
        "    self.y = torch.rand([1])*100\n",
        "    self.test = test\n",
        "    if self.test:\n",
        "      self.coefficients = torch.ones([6])\n",
        "      self.x = torch.ones([1])\n",
        "      self.y = torch.ones([1])\n",
        "    self.x = self.x.to(device)\n",
        "    self.y = self.y.to(device)\n",
        "    self.coefficients = self.coefficients.to(device)\n",
        "    self.state = torch.cat([self.coefficients,self.x,self.y])\n",
        "    self.length = 50\n",
        "  def step(self, actionx, actiony):\n",
        "    netin = self.calc()\n",
        "    self.x += actionx\n",
        "    self.y += actiony\n",
        "    self.length-=1\n",
        "    netfin = self.calc()\n",
        "    info = {}\n",
        "    if (netfin<netin):\n",
        "      reward = 1\n",
        "    else:\n",
        "      reward = -5\n",
        "    self.state = torch.cat([self.coefficients,self.x,self.y])\n",
        "    return self.state,reward,self.length,info\n",
        "  def reset(self):\n",
        "    self.coefficients = (torch.rand([6])-0.5)*60\n",
        "    self.x = torch.rand([1])*100\n",
        "    self.y = torch.rand([1])*100\n",
        "    if self.test:\n",
        "      self.coefficients = torch.ones([6])\n",
        "      self.x = torch.ones([1])\n",
        "      self.y = torch.ones([1])\n",
        "    self.x = self.x.to(device)\n",
        "    self.y = self.y.to(device)\n",
        "    self.coefficients = self.coefficients.to(device)\n",
        "    self.state = torch.cat([self.coefficients,self.x,self.y])\n",
        "    self.length = 50\n",
        "  def render(self, mode='human'):\n",
        "    pass\n",
        "  def calc(self):\n",
        "    ans = self.coefficients[0]*(self.x**2)+self.coefficients[1]*(self.y**2)+self.coefficients[2]*(self.x*self.y)+self.coefficients[3]*(self.x)+self.coefficients[4]*(self.y)+self.coefficients[5]\n",
        "    return ans\n",
        "\n",
        "environment = FooEnv()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8xJt1ZqBrx9d",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "a29dd9ff-337b-48d7-d679-7b882aaf5f03"
      },
      "source": [
        "class Network(nn.Module):\n",
        "  def __init__(self):\n",
        "    super().__init__()\n",
        "    self.m = nn.Sequential(\n",
        "        nn.Linear(8,32),\n",
        "        nn.Tanh(),\n",
        "        nn.Linear(32,32),\n",
        "        nn.Tanh(),\n",
        "        nn.Linear(32,64),\n",
        "        nn.Tanh(),\n",
        "        nn.Linear(64,128),\n",
        "        nn.Tanh(),\n",
        "        nn.Linear(128,4)\n",
        "\n",
        "    )\n",
        "  def forward(self,x):\n",
        "    return self.m(x)\n",
        "\n",
        "model = Network()\n",
        "model.to(device)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "Network(\n",
              "  (m): Sequential(\n",
              "    (0): Linear(in_features=8, out_features=32, bias=True)\n",
              "    (1): Tanh()\n",
              "    (2): Linear(in_features=32, out_features=32, bias=True)\n",
              "    (3): Tanh()\n",
              "    (4): Linear(in_features=32, out_features=64, bias=True)\n",
              "    (5): Tanh()\n",
              "    (6): Linear(in_features=64, out_features=128, bias=True)\n",
              "    (7): Tanh()\n",
              "    (8): Linear(in_features=128, out_features=4, bias=True)\n",
              "  )\n",
              ")"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 14
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "znKkAkyVNXZK"
      },
      "source": [
        "def cost(prob,rewards):\n",
        "  gamma = 0.9\n",
        "  reward = torch.zeros(1,requires_grad=True)\n",
        "  loss = torch.zeros(1,requires_grad=True)\n",
        "  reward = reward.to(device)\n",
        "  loss = loss.to(device)\n",
        "  for i in range(len(rewards)):\n",
        "    reward = rewards[i] + gamma*reward\n",
        "    loss = loss - reward*prob[i]\n",
        "  return loss"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_vvWLSH5aRdA"
      },
      "source": [
        "optimizer = torch.optim.Adam(model.parameters(),lr = 0.001)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5y5ZxG8ie5uu"
      },
      "source": [
        "testenv = FooEnv(True)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "MVZmJEmzsr5Z"
      },
      "source": [
        "episode_reward = []"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "tC-nx_Olf2Bf"
      },
      "source": [
        "def reinforce(env):\n",
        "  batches = 100\n",
        "  eps = 50\n",
        "  for i in range(batches):\n",
        "    loss = torch.zeros(1,requires_grad = True)\n",
        "    loss = loss.to(device)\n",
        "    for j in range(eps):\n",
        "      length = env.length\n",
        "      state = env.state\n",
        "      reward_list = []\n",
        "      action_list = []\n",
        "      action_prob_list = []\n",
        "      while (length>0):\n",
        "        params = model(state)\n",
        "        mean = params[:2]\n",
        "        covariance = torch.diag(params[2:])\n",
        "        covariance = torch.abs(covariance)\n",
        "        covariance = covariance.to(device)\n",
        "        dist = torch.distributions.MultivariateNormal(mean,covariance)\n",
        "        action = dist.sample()\n",
        "        action_prob = dist.log_prob(action)\n",
        "        state,reward,length,_ = env.step(action[0],action[1])\n",
        "        reward_list.append(reward)\n",
        "        action_list.insert(0,action)\n",
        "        action_prob_list.insert(0,action_prob)\n",
        "      loss = loss+cost(action_prob_list,reward_list)\n",
        "      env.reset()\n",
        "      episode_reward.append(sum(reward_list))\n",
        "    print('loss : ',loss.item())\n",
        "    optimizer.zero_grad()\n",
        "    loss.backward()\n",
        "    optimizer.step()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "JjJTNG9ugn9h"
      },
      "source": [
        "def modeltest(env):\n",
        "  env.reset()\n",
        "  length = env.length\n",
        "  state = env.state\n",
        "  while (length>0):\n",
        "    with torch.no_grad():\n",
        "      params = model(state)\n",
        "      mean = params[:2]\n",
        "      covariance = torch.tensor([[params[2],0],[0,params[3:]]])\n",
        "      covariance = torch.abs(covariance)\n",
        "      covariance = covariance.to(device)\n",
        "      dist = torch.distributions.MultivariateNormal(mean,covariance)\n",
        "      action = dist.sample()\n",
        "      state,reward,length,_ = env.step(action[0],action[1])\n",
        "  return env.calc().item()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "4N_Tq9Jpibih",
        "outputId": "9f870cc4-fd35-414a-b1c4-a5b52922799f"
      },
      "source": [
        "expected = 0\n",
        "for i in range(1000):\n",
        " expected += modeltest(testenv)\n",
        "expected/=1000\n",
        "print(f'expected value : {expected}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "expected value : 16.864271780848505\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "fHXkFZbhgV3Y",
        "outputId": "ba5cf631-a2a5-4f09-9e85-3b754b4e1ce9"
      },
      "source": [
        "reinforce(environment)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "loss :  -2096.12060546875\n",
            "loss :  -22621.20703125\n",
            "loss :  -35210.2734375\n",
            "loss :  -42473.87109375\n",
            "loss :  -45246.671875\n",
            "loss :  -47237.7734375\n",
            "loss :  -43663.0390625\n",
            "loss :  -48470.3828125\n",
            "loss :  -51612.984375\n",
            "loss :  -41604.87890625\n",
            "loss :  -38096.1640625\n",
            "loss :  -33008.28515625\n",
            "loss :  -43747.12109375\n",
            "loss :  -42836.16796875\n",
            "loss :  -40347.140625\n",
            "loss :  -42601.0\n",
            "loss :  -41781.453125\n",
            "loss :  -48344.73828125\n",
            "loss :  -45596.39453125\n",
            "loss :  -50921.390625\n",
            "loss :  -49120.75390625\n",
            "loss :  -45733.4375\n",
            "loss :  -51301.484375\n",
            "loss :  -44055.13671875\n",
            "loss :  -58804.24609375\n",
            "loss :  -53730.5546875\n",
            "loss :  -60194.67578125\n",
            "loss :  -53367.61328125\n",
            "loss :  -59272.8125\n",
            "loss :  -44295.59765625\n",
            "loss :  -56567.6875\n",
            "loss :  -58784.4453125\n",
            "loss :  -55790.88671875\n",
            "loss :  -67071.28125\n",
            "loss :  -55229.8828125\n",
            "loss :  -59136.25390625\n",
            "loss :  -51189.11328125\n",
            "loss :  -55574.6328125\n",
            "loss :  -56399.578125\n",
            "loss :  -62266.984375\n",
            "loss :  -53135.69140625\n",
            "loss :  -55384.953125\n",
            "loss :  -60318.01953125\n",
            "loss :  -57204.2421875\n",
            "loss :  -59986.24609375\n",
            "loss :  -55993.3828125\n",
            "loss :  -61628.0390625\n",
            "loss :  -53541.49609375\n",
            "loss :  -57662.37890625\n",
            "loss :  -52034.34765625\n",
            "loss :  -57407.9453125\n",
            "loss :  -55029.6015625\n",
            "loss :  -56193.46484375\n",
            "loss :  -60093.3125\n",
            "loss :  -55166.36328125\n",
            "loss :  -62370.90234375\n",
            "loss :  -54046.51953125\n",
            "loss :  -53113.71875\n",
            "loss :  -63054.890625\n",
            "loss :  -54209.2734375\n",
            "loss :  -60268.71875\n",
            "loss :  -57931.2890625\n",
            "loss :  -53348.54296875\n",
            "loss :  -55641.4453125\n",
            "loss :  -62099.57421875\n",
            "loss :  -55057.83203125\n",
            "loss :  -58055.046875\n",
            "loss :  -58237.4140625\n",
            "loss :  -58844.50390625\n",
            "loss :  -59517.16015625\n",
            "loss :  -61656.7734375\n",
            "loss :  -59703.01171875\n",
            "loss :  -54612.0078125\n",
            "loss :  -58314.15234375\n",
            "loss :  -54899.109375\n",
            "loss :  -54692.03515625\n",
            "loss :  -50766.984375\n",
            "loss :  -54877.72265625\n",
            "loss :  -58648.171875\n",
            "loss :  -55015.95703125\n",
            "loss :  -53969.58984375\n",
            "loss :  -57567.5390625\n",
            "loss :  -57701.9609375\n",
            "loss :  -49400.90625\n",
            "loss :  -49817.9453125\n",
            "loss :  -53249.28125\n",
            "loss :  -53437.296875\n",
            "loss :  -60213.80078125\n",
            "loss :  -59751.2734375\n",
            "loss :  -58223.26953125\n",
            "loss :  -50791.04296875\n",
            "loss :  -60183.375\n",
            "loss :  -56485.06640625\n",
            "loss :  -56988.28125\n",
            "loss :  -55707.828125\n",
            "loss :  -58766.2890625\n",
            "loss :  -50954.18359375\n",
            "loss :  -53286.296875\n",
            "loss :  -54100.19921875\n",
            "loss :  -52555.359375\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Le8RywsQgiyE",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "cbf54618-344d-4737-9f05-4a7601551646"
      },
      "source": [
        "expected = 0\n",
        "for i in range(1000):\n",
        " expected += modeltest(testenv)\n",
        "expected/=1000\n",
        "print(f'expected value : {expected}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "expected value : 49.06428558665514\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1Gywi3NPs0M_",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 265
        },
        "outputId": "ab2522eb-5dce-4dfd-c64d-50ccf1174ebc"
      },
      "source": [
        "import matplotlib.pyplot as plt\n",
        "\n",
        "for i in range(len(episode_reward)):\n",
        "  episode_reward[i]  = episode_reward[i]\n",
        "\n",
        "moving_averages = []\n",
        "for i in range(100,len(episode_reward)):\n",
        "  moving_averages.append(sum(episode_reward[i-100:i])/100)\n",
        "plt.plot(episode_reward)\n",
        "plt.plot(moving_averages)\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    }
  ]
}