B站-刘老师-PyTorch深度学习实践-处理多维特征的输入课后练习-学习笔记（5）

        本次课程老师讲述关于多维的输入值如何训练，通过多层神经网络对原输入值进行线性处理最后得到需要的结果

  视频如下：《PyTorch深度学习实践》完结合集_哔哩哔哩_bilibili

  代码如下：

import torch
import numpy as np
import matplotlib.pyplot as plt

# 得到数据集
def getDataset(path):
    xy = np.loadtxt(path, delimiter=',',
                    dtype=np.float32)
    x_data = torch.from_numpy(xy[:, :-1])  
    y_data = torch.from_numpy(xy[:, [-1]])
    return x_data,y_data

# 构建神经网络
class NNModel(torch.nn.Module):

    def __init__(self,name):
        super(NNModel, self).__init__()
        self.name = name
        self.linear1 = torch.nn.Linear(8,6)
        self.linear2 = torch.nn.Linear(6,4)
        self.linear3 = torch.nn.Linear(4,1)
        if (self.name == 'ELU'):
            self.active = torch.nn.ELU()
            self.sigmoid = torch.nn.Sigmoid()
        elif (self.name == 'Hardswish'):
            self.active = torch.nn.Hardswish()
            self.sigmoid = torch.nn.Sigmoid()
        elif (self.name == 'ReLU6'):
            self.active = torch.nn.ReLU6()
            self.sigmoid = torch.nn.Sigmoid()
        elif (self.name == 'Softplus'):
            self.active = torch.nn.Softplus()
            self.sigmoid = torch.nn.Sigmoid()
        elif (self.name == 'Tanf'):
            self.active = torch.nn.Tanh()
            self.sigmoid = torch.nn.Sigmoid()
        else:
            self.active = torch.nn.Sigmoid()
            self.sigmoid = torch.nn.Sigmoid()

    def forward(self,x):
        x = self.active(self.linear1(x))
        x = self.active(self.linear2(x))
        # 判断矩阵中是否有小于0或者大于1的值
        return self.judge(x)

    # 当计算y_hat时，如果有值小于0 或者 大于1，就进行sigmoid计算
    def judge(self,x):
        if (torch.where(x < 0) or torch.where(x > 1)):
            return self.sigmoid(self.linear3(x))
        else:
            return self.active(self.linear3(x))

# 绘制epoch-loss图
def draw_image_epoch_loss(name,loss):
    epoch = range(1,len(loss)+1)
    plt.plot(epoch, loss)
    plt.xlabel('Epoch')
    plt.ylabel('Loss')
    plt.title(name)
    plt.savefig('./'+name +'.jpg', dpi=300)

def run(path,name):
    l = []

    x_data,y_data = getDataset(path)
    model = NNModel(name)

    # 构建loss值和优化器
    criterion = torch.nn.BCELoss(size_average=False)
    optimizer = torch.optim.SGD(model.parameters(),lr = 0.01)

    for epoch in range(200):
        # Forward
        y_pred = model(x_data)
        loss = criterion(y_pred,y_data)
        l.append(loss.item())

        # Backward
        optimizer.zero_grad()
        loss.backward()

        # Update
        optimizer.step()

    # 将loss变化图保存
    draw_image_epoch_loss(name,l)

if __name__ == '__main__':
    name = 'sigmoid'
    # name = 'Hardswish'
    # name = 'ReLU6'
    # name = 'ELU'
    # name = 'Softplus'
    # name = 'Tanf'
    run('diabetes.csv',name)

从老师的PPT中所列举的激活函数进行训练，这里只列举几个函数(本人才疏学浅，不理解有些激活函数让Loss值居高不降？如果有大佬指错，本人长揖到地表示感谢)。

各类激活函数：

 （侵权删）

运行的效果图: