new test file

Reinforcement Learning/LunarLander/_.ipynb

@@ -0,0 +1,109 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import gym\n",
+    "import torch\n",
+    "\n"
+   ]
+  },
+  {
+   "attachments": {},
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Define Policy Network"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "class PolicyNetwork(torch.nn.Module):\n",
+    "    def __init__(self, n=4, in_dim=128):\n",
+    "        super(PolicyNetwork, self).__init__()\n",
+    "        self.fc1 = torch.nn.Linear(in_dim, 128)\n",
+    "        self.fc2 = torch.nn.Linear(128, 128)\n",
+    "        self.fc3 = torch.nn.Linear(128, 128)\n",
+    "        self.fc4 = torch.nn.Linear(128, n) \n",
+    "        self.l_relu = torch.nn.LeakyReLU(0.1)\n",
+    "\n",
+    "    def forward(self, x):        \n",
+    "        x = self.l_relu(self.fc1(x))\n",
+    "        x = self.l_relu(self.fc2(x))\n",
+    "        x = self.l_relu(self.fc3(x))       \n",
+    "        y = self.fc4(x)       \n",
+    "        y = F.softmax(y, dim=-1)        \n",
+    "        return y    \n",
+    "\n",
+    "    def sample_action(self, state):\n",
+    "        if not state is torch.Tensor:\n",
+    "            state = torch.from_numpy(state).float().to(device)\n",
+    "        if len(state.size()) == 1:\n",
+    "            state = state.unsqueeze(0)       \n",
+    "        y = self(state)\n",
+    "        dist = Categorical(y)\n",
+    "        action = dist.sample()\n",
+    "        log_probability = dist.log_prob(action)\n",
+    "        return action.item(), log_probability.item()\n",
+    "\n",
+    "    def best_action(self, state):        \n",
+    "        if not state is torch.Tensor:\n",
+    "            state = torch.from_numpy(state).float().to(device)        \n",
+    "        if len(state.size()) == 1:\n",
+    "            state = state.unsqueeze(0)        \n",
+    "        y = self(state).squeeze()        \n",
+    "        action = torch.argmax(y)        \n",
+    "        return action.item()    \n",
+    "\n",
+    "    def evaluate_actions(self, states, actions):\n",
+    "        y = self(states)        \n",
+    "        dist = Categorical(y)        \n",
+    "        entropy = dist.entropy()        \n",
+    "        log_probabilities = dist.log_prob(actions) \n",
+    "        return log_probabilities, entropy\n",
+    "\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "base",
+   "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": "3.9.12"
+  },
+  "orig_nbformat": 4,
+  "vscode": {
+   "interpreter": {
+    "hash": "4f523f7c76dd18e7ed336217f32f6f704c23c323644912475b9d3570cf04b060"
+   }
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}