Pytorch 学习（三）sigmoid 逻辑回归预测

1. 数据预处理  处理多特征样本

import pandas as pd
 
data = pd.read_csv('./....csv')

# 划分数据  iloc[] 第一个参数 行 第二个参数 列
X = data.iloc[:, :-1]

# 标签为-1/1， 讲-1 替换为 0 
Y = data.iloc[:, -1].replace(-1, 0)

# numpy 转换为 tensor
X = torch.form_numpy(X.values)).float()
Y = torch.from_numpy(Y.values.reshape(-1, 1)).float() 

2. 创建模型

# nn.Sequential 连接多层
model = nn.Sequential(
            nn.Linear(15, 1)
            nn.Sigmoid()            # 激活
)


loss_fun = nn.BCELoss()        # 二元交叉熵损失

opt = torch.optim.Adam(model.parameters(), lr = 0.0001)    # 优化器

3. Train

# batch 批次训练
batches = 16 
num_of_batch = 653//16

epoches = 1000

for epoch in range(epoches):
    for batch in range(num_of_batch):
        start = batches * batch
        end = start + batches
        
        x = X[start:end]
        y = Y[start:end]
        y_pred = model(x)

        loss = loss_fun(y_pred, y)

        opt.zerp_grad()

        loss.backward()
    
        opt.step()

4. 测试

(model(X).data.numpy > 0.5).astype('int')