pytorch解决鸢尾花分类

骑猪看日落 2021-11-27 01:46 269阅读 0赞

半年前用numpy写了个鸢尾花分类200行。。每一步计算都是手写的 [python构建bp神经网络\_鸢尾花分类][python_bp]

现在用pytorch简单写一遍，pytorch语法解释请看上一篇[pytorch搭建简单网络][pytorch]

1 import pandas as pd
     2 import torch.nn as nn
     3 import torch
     4 
     5 
     6 class MyNet(nn.Module):
     7     def __init__(self):
     8         super(MyNet, self).__init__()
     9         self.fc = nn.Sequential(
    10             nn.Linear(4, 3),
    11             nn.Sigmoid(),
    12             nn.Linear(3, 3),
    13             nn.Sigmoid(),
    14             nn.Linear(3, 1),
    15         )
    16         self.mls = nn.MSELoss()
    17         self.opt = torch.optim.Adam(params=self.parameters(), lr=0.001)
    18 
    19     def get_data(self):
    20         inputs = []
    21         labels = []
    22         with open('flower.csv') as file:
    23             df = pd.read_csv(file, header=None)
    24             x = df.iloc[:, 0:4].values
    25             y = df.iloc[:, 4].values
    26             for i in range(len(x)):
    27                 inputs.append(x[i])
    28             for j in range(len(y)):
    29                 a = []
    30                 a.append(y[j])
    31                 labels.append(a)
    32 
    33         return inputs, labels
    34 
    35     def forward(self, inputs):
    36         out = self.fc(inputs)
    37         return out
    38 
    39     def train(self, x, label):
    40         out = self.forward(x)
    41         loss = self.mls(out, label)
    42         self.opt.zero_grad()
    43         loss.backward()
    44         self.opt.step()
    45 
    46     def test(self, x):
    47         return self.fc(x)
    48 
    49 
    50 if __name__ == '__main__':
    51     net = MyNet()
    52     inputs, labels = net.get_data()
    53     for i in range(1000):
    54         for index, input in enumerate(inputs):
    55             # 这里不加.float()会报错，可能是数据格式的问题吧
    56             input = torch.from_numpy(input).float()
    57             label = torch.Tensor(labels[index])
    58             net.train(input, label)
    59     # 简单测试一下
    60     c = torch.Tensor([[5.6, 2.7, 4.2, 1.3]])
    61     print(net.test(c))

运行结果趋近于0.5 正确，单纯练一下pytorch，就没有分训练集，测试集

1 tensor([[0.5392]], grad_fn=<AddmmBackward>)

不用手写反向传播和梯度下降 是多么幸福一件事~

转载于:https://www.cnblogs.com/MC-Curry/p/10109138.html

[python_bp]: https://www.cnblogs.com/MC-Curry/p/9105908.html
[pytorch]: https://www.cnblogs.com/MC-Curry/p/10107380.html