RL.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-08-17T05:36:38.934238Z",
     "start_time": "2019-08-17T05:36:38.438787Z"
    }
   },
   "outputs": [],
   "source": [
    "from __future__ import print_function\n",
    "\n",
    "import csv\n",
    "import math\n",
    "import sys\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import matplotlib.patches as patches\n",
    "import random\n",
    "\n",
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "import torch.optim as optim\n",
    "import torchvision.transforms as T\n",
    "from scipy.stats import kde\n",
    "\n",
    "from hp import HP\n",
    "from rl_memory import Memory\n",
    "\n",
    "%matplotlib inline\n",
    "\n",
    "dev = \"cuda:0\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-08-17T05:36:38.971752Z",
     "start_time": "2019-08-17T05:36:38.935588Z"
    }
   },
   "outputs": [],
   "source": [
    "class Net(nn.Module):\n",
    "    \n",
    "    # Input is one hot encoding of the protein and its conformation\n",
    "    # Output is the value of changing a given angle to another angle\n",
    "    \n",
    "    def __init__(self,n):\n",
    "        super(Net,self).__init__()\n",
    "        self.n = n\n",
    "        self.fc1 = nn.Linear(5*n,2048)\n",
    "        self.fc2 = nn.Linear(2048,256)\n",
    "        self.fc3 = nn.Linear(256,3*n)\n",
    "        self.relu = nn.ReLU()\n",
    "        self.dropout = nn.Dropout(p=0.5)\n",
    "\n",
    "    def forward(self,x):\n",
    "        x = x.view(-1,5*self.n)\n",
    "        x = self.dropout(x)\n",
    "        x = self.relu(self.fc1(x))\n",
    "        x = self.dropout(x)\n",
    "        x = self.relu(self.fc2(x))\n",
    "        x = self.dropout(x)\n",
    "        x = self.fc3(x)\n",
    "        x = x.view(-1,self.n,3)\n",
    "        return x\n",
    "    \n",
    "    def value(self,x):\n",
    "        return self.forward(x).view(-1,3*self.n).max(1).values        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-08-17T05:36:38.984478Z",
     "start_time": "2019-08-17T05:36:38.973668Z"
    }
   },
   "outputs": [],
   "source": [
    "def main_loop():\n",
    "    \n",
    "    energies = []\n",
    "    \n",
    "    for i in range(NUM_EPISODES):\n",
    "        \n",
    "        hp = HP.random(30)\n",
    "        state = hp.make_state(dev=dev)\n",
    "        \n",
    "        for j in range(NUM_STEPS):\n",
    "            \n",
    "            if random.random() < EXPLORE:\n",
    "                action = random_action()\n",
    "            else:\n",
    "                action = policy_net(state)\n",
    "                \n",
    "            next_state, reward = perform_action(state, action)\n",
    "            memory.push(state, action, next_state, reward)\n",
    "            state = next_state\n",
    "            optimize_model()\n",
    "            \n",
    "        hp = state2hp(state)\n",
    "        e = hp.energy()\n",
    "        energies.append(e)\n",
    "        sys.stdout.write(\"\\r%d, %s, %s, %f     \" % (i, hp.sequence, hp.conformation, e))\n",
    "                        \n",
    "        value_net.load_state_dict(policy_net.state_dict())\n",
    "        \n",
    "    return energies\n",
    "            \n",
    "def state2hp(state):\n",
    "    return HP.create_from_one_hots(state[:,:2], state[:PROTEIN_LENGTH-1,2:5]) \n",
    "\n",
    "def random_action():\n",
    "    a = torch.zeros(1,PROTEIN_LENGTH,3)\n",
    "    a = a.to(dev)\n",
    "    i = random.randint(0,29)\n",
    "    j = random.randint(0,2)\n",
    "    a[0][i][j] = 1\n",
    "    return a\n",
    "\n",
    "def perform_action(state, action):\n",
    "    n = len(action[0])\n",
    "    a = action.view(-1,3*n).argmax(-1)\n",
    "    i, j = int(a/3), int(a%3)\n",
    "    new_state = state.clone()\n",
    "    new_state[i][2] = 0\n",
    "    new_state[i][3] = 0\n",
    "    new_state[i][4] = 0    \n",
    "    new_state[i][j+2] = 1\n",
    "    return new_state, state2hp(state).energy() - state2hp(new_state).energy()\n",
    "\n",
    "def optimize_model():\n",
    "    \n",
    "    if len(memory) < BATCH_SIZE:\n",
    "        return\n",
    "    \n",
    "    states, next_states, actions, rewards = memory.sample(BATCH_SIZE)\n",
    "    estimates = policy_net.value(states)\n",
    "    targets = GAMMA * value_net.value(next_states) + rewards\n",
    "    loss = loss_function(estimates, targets)\n",
    "    optimizer.zero_grad()\n",
    "    loss.backward()\n",
    "    torch.nn.utils.clip_grad_norm_(policy_net.parameters(), GRAD_CLIP)\n",
    "    optimizer.step()\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-08-17T23:52:27.532464Z",
     "start_time": "2019-08-17T05:36:38.986299Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "3999, PHHPPPHPHPHPPPHHPHPPPHHPPHHHHP, SLSLLSRLLLSRSSRSSSLSSSLRSLSSL, 497.000000      "
     ]
    }
   ],
   "source": [
    "PROTEIN_LENGTH = 30\n",
    "NUM_EPISODES = 4000\n",
    "NUM_STEPS = 500\n",
    "MEM_SIZE = 200000\n",
    "BATCH_SIZE = 128\n",
    "GAMMA = 0.9999\n",
    "LR = 0.0001\n",
    "GRAD_CLIP = 0.25\n",
    "EXPLORE = 0.05\n",
    "\n",
    "policy_net = Net(PROTEIN_LENGTH)\n",
    "policy_net.to(dev)\n",
    "value_net = Net(PROTEIN_LENGTH)\n",
    "value_net.to(dev)\n",
    "\n",
    "memory = Memory(MEM_SIZE,dev)\n",
    "\n",
    "loss_function = nn.MSELoss()\n",
    "optimizer = optim.Adam(policy_net.parameters(),  lr=LR)\n",
    "\n",
    "energies = main_loop()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-08-18T05:13:15.253537Z",
     "start_time": "2019-08-18T05:13:15.119198Z"
    }
   },
   "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"
    }
   ],
   "source": [
    "M = 25\n",
    "x = [sum(energies[M*i:M*(i+1)])/M for i in range(int(len(energies)/M))]\n",
    "plt.plot(x);"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-08-18T05:13:30.445822Z",
     "start_time": "2019-08-18T05:13:30.364341Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor(5834, device='cuda:0')"
      ]
     },
     "execution_count": 46,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "states, next_states, actions, rewards = memory.sample(50*BATCH_SIZE)\n",
    "rewards.argmax()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-08-18T05:13:31.729207Z",
     "start_time": "2019-08-18T05:13:31.539511Z"
    }
   },
   "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"
    }
   ],
   "source": [
    "state2hp(states[22]).show(plt.gca())"
   ]
  },
  {
   "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": "3.6.8"
  },
  "toc": {
   "base_numbering": 1,
   "nav_menu": {},
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": false,
   "title_cell": "Table of Contents",
   "title_sidebar": "Contents",
   "toc_cell": false,
   "toc_position": {},
   "toc_section_display": true,
   "toc_window_display": false
  },
  "varInspector": {
   "cols": {
    "lenName": 16,
    "lenType": 16,
    "lenVar": 40
   },
   "kernels_config": {
    "python": {
     "delete_cmd_postfix": "",
     "delete_cmd_prefix": "del ",
     "library": "var_list.py",
     "varRefreshCmd": "print(var_dic_list())"
    },
    "r": {
     "delete_cmd_postfix": ") ",
     "delete_cmd_prefix": "rm(",
     "library": "var_list.r",
     "varRefreshCmd": "cat(var_dic_list()) "
    }
   },
   "types_to_exclude": [
    "module",
    "function",
    "builtin_function_or_method",
    "instance",
    "_Feature"
   ],
   "window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}