2023年5月17日,备份下，这是以前g改动的

main.py

@@ -4,6 +4,7 @@ import torch.nn as nn
 import torch.optim as optim
 import numpy as np
 from torch.utils.data import Dataset, DataLoader
+from sklearn.preprocessing import MinMaxScaler, LabelEncoder
 
 # Define MLP model
 class MLP(nn.Module):
@@ -13,71 +14,77 @@ class MLP(nn.Module):
         self.relu1 = nn.ReLU()
         self.fc2 = nn.Linear(hidden_size, output_size)
         self.sigmoid = nn.Sigmoid()
+        self.softmax = nn.Softmax(dim=1)
         
     def forward(self, x):
         out = self.fc1(x)
         out = self.relu1(out)
         out = self.fc2(out)
-        out = self.sigmoid(out)
         return out
 
 # Define custom dataset
 class PsychologyDataset(Dataset):
     def __init__(self, data_file):
-        self.data = pd.read_excel(data_file)
+        data = pd.read_excel(data_file)
+        src_features = data.iloc[:, 34:44].values.astype(np.float32)
+        src_labels = data.iloc[:, -1].values
+        # 数据预处理
+        scaler = MinMaxScaler(feature_range=(0, 5))
+        #self.opt_features = scaler.fit_transform(src_features)
+        self.opt_features = src_features/5
+        # 标签编码
+        label_encoder = LabelEncoder()
+        self.opt_labels = label_encoder.fit_transform(src_labels)
+
         
     def __len__(self):
-        return len(self.data)
+        return len(self.opt_features)
     
     def __getitem__(self, idx):
-        features = self.data.iloc[idx, 36:43].values.astype(np.float32)
-        str = self.data.iloc[idx, -1]
-        #print(idx,str,self.data.iloc[0, 0])
-        label = -1
-        if(str=="是"):
-            label = 1
-        else:
-            label = 0
-        #print(features)
-        label = np.float32(label)
-        #return torch.tensor(features, dtype=torch.float), label
-        return features, label
+        return self.opt_features[idx], self.opt_labels[idx]
 
 # Set hyperparameters
-input_size = 7
-hidden_size = 16
+input_size = 10
+hidden_size = 128
 output_size = 1
-lr = 0.01
+lr = 0.001
 num_epochs = 100
 
 # Load data
-dataset = PsychologyDataset("data/data_src.xlsx")
-dataloader = DataLoader(dataset, batch_size=1, shuffle=False)
+dataset = PsychologyDataset("/home/wcl/psychological_prediction/data/data_src.xlsx")
+dataloader = DataLoader(dataset, batch_size=32, shuffle=False)
 
 # Instantiate model, loss function, and optimizer
 model = MLP(input_size, hidden_size, output_size)
-criterion = nn.BCELoss()
+criterion = nn.BCEWithLogitsLoss()
+
 optimizer = optim.Adam(model.parameters(), lr=lr)
+#optimizer = optim.SGD(model.parameters(), lr=0.01)
 
 # Train model
 for epoch in range(num_epochs):
     running_loss = 0.0
     train_corrects = 0
-    #print(type(dataloader))
     for i, data in enumerate(dataloader):
-        #print("数据序号：", i, data)
-        #continue
         inputs, labels = data
         optimizer.zero_grad()
         outputs = model(inputs)
-        loss = criterion(outputs, labels.unsqueeze(1))
+        loss = criterion(outputs, labels.view(-1,1).to(torch.float))
+        #loss = criterion(outputs, labels)
         loss.backward()
         optimizer.step()
-        running_loss += loss.item()
-        _, preds = torch.max(outputs, 1)
-        train_corrects += torch.sum(preds == labels.data)
-    print('Epoch [%d/%d], Loss: %.4f' % (epoch+1, num_epochs, running_loss / len(dataloader)))
-    train_acc = train_corrects.double() / len(dataloader)
+        running_loss += loss.item() * inputs.size(0)
+        predicted = torch.round(torch.sigmoid(outputs))
+        #print((predicted == labels.view(-1,1)).sum().item())
+        assert(outputs.sum().item()!=0), {outputs, predicted, labels}
+        #train_corrects += torch.sum(predicted == labels.data)
+        correct = (predicted == labels.view(-1,1)).sum().item()
+        #print(correct, labels.size(0))
+        train_corrects += correct
+    
+    print('Epoch [%d/%d], Loss: %.4f' % (epoch+1, num_epochs, running_loss / len(dataloader.dataset)))
+    print(len(dataloader.dataset), len(dataloader))
+    train_acc = float(train_corrects) / len(dataloader.dataset)
     print('Epoch [%d/%d], ACC: %.4f' % (epoch+1, num_epochs, train_acc))
 
 # Save trained model

val.py

@@ -37,9 +37,10 @@ class MMLP(nn.Module):
             self.layers.append(nn.ReLU())
             input_size = h
         self.layers.append(nn.Linear(input_size, output_size))
-        # 最后一层加入激活函数后，会严重影响收敛情况，原因待分析
+        # 最后一层加入激活函数后，会严重影响收敛情况，原因待分析,softmax不需要加，因为nn.CrossEntropyLoss()损失函数自带softmax运算，加了会连续两次指数运算，容易溢出；至于Sigmoid，原因是输入数据要做（0,1）的归一化
         #self.layers.append(nn.Sigmoid())
         #self.layers.append(nn.Softmax(dim=1))
+        #self.layers.append(nn.LogSoftmax(dim=1))
 
     def forward(self, x):
         for layer in self.layers:
@@ -204,6 +205,7 @@ for fold, (train_idx, test_idx) in enumerate(skf.split(src_features, src_labels)
     criterion = nn.CrossEntropyLoss()
     #criterion = nn.BCELoss() # 使用sigmoid损失函数再使用
     optimizer = torch.optim.Adam(model.parameters(), lr=lr)
+    #optimizer = torch.optim.SGD(model.parameters(), lr=lr) # 这个优化器只有lr一个超参数，适用于小规模网络结构
 
     # 训练 MLP 模型
     train_model(model, train_loader, criterion, optimizer, num_epochs)