# 隨機初始化一個二維數據集，使用朋友torch訓練一個回歸模型
import numpy as np
import random
import matplotlib.pyplot as plt

x = np.arange(20)
y = np.array([5*x[i] + random.randint(1,20) for i in range(len(x))])    # random.randint(參數1,參數2)函數返回參數1和參數2之間的任意整數
print('-'*50)
# 打印數據集
print(x)
print(y)

import torch
x_train = torch.from_numpy(x).float()
y_train = torch.from_numpy(y).float()

# model
class LinearRegression(torch.nn.Module):
    def __init__(self):
        super(LinearRegression, self).__init__()
        # 輸入與輸出都是一維的
        self.linear = torch.nn.Linear(1,1)
    def forward(self,x):
        return self.linear(x)

# 新建模型，誤差函數，優化器
model = LinearRegression()
criterion = torch.nn.MSELoss()
optimizer = torch.optim.SGD(model.parameters(),0.001)
# 開始訓練
num_epoch = 20
for i in range(num_epoch):
    input_data = x_train.unsqueeze(1)
    target = y_train.unsqueeze(1)           # unsqueeze(1)在第二維增加一個維度
    out = model(input_data)
    loss = criterion(out,target)
    optimizer.zero_grad()
    loss.backward()
    optimizer.step()
    print("Eopch:[{}/{},loss:[{:.4f}]".format(i+1,num_epoch,loss.item()))
    if ((i+1)%2 == 0):
        predict = model(input_data)
        plt.plot(x_train.data.numpy(),predict.squeeze(1).data.numpy(),"r")
        loss = criterion(predict,target)
        plt.title("Loss:{:.4f}".format(loss.item()))
        plt.xlabel("X")
        plt.ylabel("Y")
        plt.scatter(x_train,y_train)
        plt.show()