Pytorch 实现逻辑回归

1 导入实验需要的包

import numpy as np
import pandas as pd 
import torch
from torch import nn
import matplotlib.pyplot as plt
from torch.autograd import Variable
from torch.utils.data import DataLoader,TensorDataset
import os
os.environ["KMP_DUPLICATE_LIB_OK"]  =  "TRUE"

2 加载数据集

data =pd.read_csv("./dataset/credit-a.csv")
X = data.iloc[:,:-1]  
Y = data.iloc[:,-1]  

3 数据转换

Y = Y.replace(-1,0)  #替换标签 -1 ，改成 0 
Y.unique() #查看有几种标签数据
#数据转换为Tensor
X = torch.from_numpy(X.values).type(torch.float32)
Y = torch.from_numpy(Y.values.reshape(-1,1)).type(torch.float32)

4 设置迭代器，将数据转换在cuda下

def load_data(X,Y,batch_size):
    X = torch.autograd.Variable(X).cuda()
    Y = torch.autograd.Variable(Y).cuda()
    print("type x",X)
    data = TensorDataset(X,Y)
    data_loader = DataLoader(data,batch_size,shuffle=True)
#     data_loader = DataLoader(data,batch_size,shuffle=False)
    return data_loader

5 参数初始化及读取数据

batchs_size = 16
data_iter = load_data(X,Y,batchs_size)
#最后的代码用到了，cpu（）那行
X = torch.autograd.Variable(X).cuda()
Y = torch.autograd.Variable(Y).cuda()

6 设置模型

model = nn.Sequential()
model.add_module('mylinear1',nn.Linear(15,1))
model.add_module('mysigmoid',nn.Sigmoid())
if torch.cuda.is_available():
    model.cuda()

7 设置损失函数

loss_fn = nn.BCELoss()

8 设置优化器

optimizer = torch.optim.Adam(model.parameters(),lr=0.0001)
# optimizer = torch.optim.SGD(model.parameters(),lr=0.001)

9 训练模型

epochs = 150
for epoch in range(epochs):
    correct = 0
    for x,y in data_iter:
        y_hat = model(x)
        loss = loss_fn(y_hat,y)
#         print(y_hat.ge(0.5))
        out = y_hat.ge(0.5).float()  # 以0.5为阈值进行分类
        correct += (out == y).sum()  # 计算正确预测的样本个数
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()
    print("epoch  %d ,current acc = %.4f"%(epoch+1,correct/Y.shape[0]))
    print("epoch  %d ,current loss = %.4f"%(epoch+1,loss))

10 其他

# ((model(X).data.numpy()> 0.5).astype('int')==Y.numpy()).mean()
((model(X).data.cpu().numpy()> 0.5).astype('int')==Y.cpu().numpy()).mean()

model(X).data.cpu()