说明：跟教程学搞了一个神经网络，一共283个样本，每个样本144个特征（样本数据data_trans.csv），样本标签是10分类（标签文件data_label_10.csv），样本都是csv格式的，模型是随便抄的很简单的cnn。
问题1：在Dataloader中设置的batch_size=10，为什么在Conv1卷积层中in_channel设为1会出bug，设置为batch_size一样的10才行，样本本身就是1维的啊。
问题2：Linear层out_features设为10会报错，“mat1 and mat2 shapes cannot be multiplied (5x144 and 720x10)”，是哪个参数设置错误了啊？
问题3：请帮忙处理bug，能跑通就行。
代码和数据csv文件都在Github里，链接https://github.com/sizimiya/CSDN-Question.git
有用记得采纳，batch_size=随便改
# -*- coding: UTF-8 -*-
"""
@项目名称：简单cnn网络_csv数据集_bug解决.py
@作 者：陆地起飞全靠浪
@创建日期：2022-08-31-09:20
https://ask.csdn.net/questions/7779923
https://blog.csdn.net/weixin_41944061?type=ask
"""
import pandas as pd
import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import Dataset, DataLoader
from tqdm import tqdm
from datetime import datetime
class CsvDataset(Dataset):
def __init__(self):
super(CsvDataset, self).__init__()
self.feature_path = 'data_trans.csv'
self.label_path = 'data_label_10.csv'
feature_df_ = pd.read_csv(self.feature_path)
label_df_ = pd.read_csv(self.label_path)
assert feature_df_.columns.tolist()[1:] == label_df_[label_df_.columns[0]].tolist(), \
'feature name does not match label name'
self.feature = [feature_df_[i].tolist() for i in feature_df_.columns[1:]]
self.label = label_df_[label_df_.columns[1]]
assert len(self.feature) == len(self.label)
self.length = len(self.feature)
def __getitem__(self, index):
x = self.feature[index]
x = torch.Tensor(x)
x = x.reshape(1,12, 12)
y = self.label[index]
return x, y
def __len__(self):
return self.length
train_dataset = CsvDataset()
train_loader = DataLoader(dataset=train_dataset, batch_size=2, shuffle=False)
class SimpleModel(nn.Module):
def __init__(self):
super(SimpleModel, self).__init__()
self.conv1 = nn.Conv2d(in_channels=1, out_channels=3, kernel_size=(1, 1)) # ***************************
self.conv2 = nn.Conv2d(in_channels=3, out_channels=5, kernel_size=(1, 1))
self.relu = nn.ReLU(inplace=True)
self.flatten = nn.Flatten(start_dim=1, end_dim=-1) # (B, C, H ,W)
self.linear = nn.Linear(in_features=5 * 12 * 12, out_features=10, bias=False) # ****************************
def forward(self, x):
x = self.conv1(x)
x = self.relu(x)
x = self.conv2(x)
x = self.relu(x)
print("[before flatten] x.shape: {}".format(x.shape)) # torch.Size([1, 5, 12, 12])
x = self.flatten(x)
print("[after flatten] x.shape: {}".format(x.shape)) # torch.Size([1, 720])
x = self.linear(x)
x = self.relu(x)
return x
model = SimpleModel()
optimizer = optim.SGD(params=model.parameters(), lr=0.0001, momentum=0.9)
loss_fn = nn.CrossEntropyLoss()
for epoch in range(2):
with tqdm(train_loader, desc='EPOCH:{}'.format(epoch)) as train_bar:
for (x, y) in train_bar:
optimizer.zero_grad()
loss = loss_fn(model(x), y)
loss.backward()
optimizer.step()
print('epoch:{}, loss:{:.6f}'.format(epoch, loss))
time = str(datetime.now()).split('')[0].replace('-', '_')
torch.save(model.state_dict(), 'model_{}.pth'.format(time))