ch20 Level5_Pytorch.py

Add GRU model

A-基础教程/A2-神经网络基本原理简明教程/SourceCode/ch20-RNNModel/Level5_Pytorch.py

@@ -14,6 +14,7 @@
 import torch.nn as nn
 import torch
 from torch.utils.data import TensorDataset, DataLoader
+from torch.nn.init import xavier_normal
 
 
 train_file = "../../data/ch19.train_minus.npz"
@@ -34,10 +35,36 @@ def process_data(dr):               # 统一输入单词的长度，不足补0
 
     return dr.XTrain, dr.XTest, dr.YTrain, dr.YTest             # 这是为了统一一个表达方式，和之前的章节一样
 
+def weights_init(m):
+    classname=m.__class__.__name__
+    if classname.find('Conv') != -1:
+        xavier_normal(m.weight.data)
+        xavier_normal(m.bias.data)
 
-class RNN(nn.Module):
+class LSTM(nn.Module):
     def __init__(self):
-        super(RNN, self).__init__()
+        super(LSTM, self).__init__()
+        self.rnn = nn.LSTM(
+            input_size=2,          # character num.
+            hidden_size=4,         # RNN or LSTM hidden layer, 设置的稍大一些可能效果更佳，此处仅作对比
+            num_layers=1,
+            batch_first=True,
+            bidirectional=True,      # 双向LSTM, 若设置为False,对应hidden_size增大两倍
+
+        )
+        self.softmax = nn.Softmax()         # classification, softmax
+        self.fc = nn.Linear(8, 2)
+        self.sigmoid = nn.Sigmoid()
+
+    def forward(self, x):
+        r_out, (h_n, h_c) = self.rnn(x, None)           # 多对多的序列            (2, 4, 8)
+        out = self.fc(r_out)                            # 0和1是二分类问题       （2, 4, 2)
+        out = self.sigmoid(out)
+        return out
+
+class GRU(nn.Module):
+    def __init__(self):
+        super(GRU, self).__init__()
         self.rnn = nn.LSTM(
             input_size=2,          # character num.
             hidden_size=4,         # RNN or LSTM hidden layer, 设置的稍大一些可能效果更佳，此处仅作对比
@@ -77,7 +104,9 @@ def accracy_score(pred, label):
     max_epoch = 100      # hyper-parameters
     lr = 1e-2
     batch_size = 2
-    rnn = RNN()
+    # rnn = LSTM()        # LSTM  model
+    rnn = GRU()         # GRU model
+    rnn.apply(weights_init)
 
     # Data processing
     dataReader = load_data()