plt、nn 使用

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt

data = pd.read_csv('csv_data/test.csv')
data.head(3) # 查看前3行
data.info() # 查看类型、是否空

plt.scatter(data.num,data.debt)  # 散点图
plt.xlabel('num')
plt.ylabel('debt')

from torch import nn
import torch

# 数据预处理
x = torch.from_numpy(data.num.values.reshape(-1,1).astype(np.float32))
y = torch.from_numpy(data.debt.values.reshape(-1,1).astype(np.float32))
# 转为tensor类型

model = nn.Linear(1,1)
loss_fn = nn.MSELoss() # loss函数
opt =torch.optim.SGD(model.parameters(),lr=0.001) # 优化

for epoch in range(100):
    for a,b in zip(x,y):  
        y_pred = model(a)  # 预测
        loss=loss_fn(b,y_pred) # 损失
        opt.zero_grad() # 梯度清零
        loss.backward() # 计算梯度
        opt.step()      # 优化参数
    with torch.no_grad():
        print(epoch,'  loss: ', loss_fn(model(x),y).data.item())

plt.scatter(data.num,data.debt)
plt.plot(x.numpy(),model(x).data.numpy(), c='r')