update

codes/PolicyGradient/model.py

@@ -0,0 +1,30 @@
+#!/usr/bin/env python
+# coding=utf-8
+'''
+Author: John
+Email: johnjim0816@gmail.com
+Date: 2021-03-23 16:35:58
+LastEditor: John
+LastEditTime: 2021-03-23 16:36:20
+Discription: 
+Environment: 
+'''
+import torch.nn as nn
+import torch.nn.functional as F
+class MLP(nn.Module):
+    ''' 多层感知机
+        输入：state维度
+        输出：概率
+    '''
+    def __init__(self,n_states,hidden_dim = 36):
+        super(MLP, self).__init__()
+        # 24和36为hidden layer的层数，可根据state_dim, n_actions的情况来改变
+        self.fc1 = nn.Linear(n_states, hidden_dim)
+        self.fc2 = nn.Linear(hidden_dim,hidden_dim)
+        self.fc3 = nn.Linear(hidden_dim, 1)  # Prob of Left
+
+    def forward(self, x):
+        x = F.relu(self.fc1(x))
+        x = F.relu(self.fc2(x))
+        x = F.sigmoid(self.fc3(x))
+        return x