Pytorch dataloader implements minibatch (incomplete)

Take the book back 《pytorch Build the simplest version of neural network 》 Not used last time miniBatch, Input all the data for training at one time .

This pass DataLoader Realization miniBatch

import torch
import numpy as np
from torch.utils.data import Dataset, DataLoader


class DiabetesDataset(Dataset):
    def __init__(self, filepath):
        xy = np.loadtxt(filepath, delimiter=',', dtype=np.float32)
        self.len = xy.shape[0]
        self.x_data = torch.from_numpy(xy[:, :-1])
        self.y_data = torch.from_numpy(xy[:, [-1]])

    def __getitem__(self, index):
        return self.x_data[index], self.y_data[index]

    def __len__(self):
        return self.len


dataset = DiabetesDataset('diabetes.csv')
train_loader = DataLoader(dataset=dataset,
                          batch_size=749,
                          shuffle=False,
                          num_workers=1)


class Model(torch.nn.Module):
    def __init__(self):
        super().__init__()
        self.linear1 = torch.nn.Linear(8, 6)
        self.linear2 = torch.nn.Linear(6, 4)
        self.linear3 = torch.nn.Linear(4, 1)
        self.sigmoid = torch.nn.Sigmoid()
        self.activate = torch.nn.ReLU()

    # __call__()  This function will be called in ！
    def forward(self, x):
        # x = self.activate(self.linear1(x))
        # x = self.activate(self.linear2(x))
        # x = self.activate(self.linear3(x))
        x = self.sigmoid(self.linear1(x))
        x = self.sigmoid(self.linear2(x))
        x = self.sigmoid(self.linear3(x))

        return x


# model Is callable ！  Realized  __call__()
model = Model()

#  Specify the loss function 
# criterion = torch.nn.MSELoss(size_average=Flase)  # True
# criterion = torch.nn.MSELoss(reduction='sum')  # sum： Sum up  mean： Averaging 
criterion = torch.nn.BCELoss(reduction='mean')  #  Two class cross entropy loss function 

# --  Specify optimizer （ In fact, it is the algorithm of gradient descent , be responsible for ）, The optimizer and model Associated 
# optimizer = torch.optim.SGD(model.parameters(), lr=0.01)
optimizer = torch.optim.Adam(model.parameters(), lr=0.01)  #  This one is very accurate , Other things are not good at all 
# optimizer = torch.optim.Rprop(model.parameters(), lr=0.01)


# for epoch in range(5000):
#     y_pred = model(x_data)  #  Directly put the whole test data into 
#     loss = criterion(y_pred, y_data)
#     print(epoch, loss.item())
#     optimizer.zero_grad()  #  It will automatically find all the w and b Clear ！ The role of the optimizer  （ Why is this put in loss.backward() You can't clear it later ？）
#     loss.backward()
#     optimizer.step()  #  It will automatically find all the w and b updated , The role of the optimizer ！
#
if __name__ == '__main__':

    for epoch in range(1000):
        for i, data in enumerate(train_loader, 0):
            # 1. Prepare data
            inputs, labels = data
            # 2. Forward
            y_pred = model(inputs)
            loss = criterion(y_pred, labels)
            print(epoch, i, loss.item())
            # 3. Backward
            optimizer.zero_grad()
            loss.backward()
            # 4. Update
            optimizer.step()


    #   test 
    xy = np.loadtxt('diabetes_test.csv', delimiter=',', dtype=np.float32)
    x_data = torch.from_numpy(xy[:, :-1])
    y_data = torch.from_numpy(xy[:, -1])

    x_test = torch.Tensor(x_data)
    y_test = model(x_test)  #  forecast 
    print('y_pred = ', y_test.data)

    #  Compare the predicted results with the real results 
    for index, i in enumerate(y_test.data.numpy()):
        if i[0] > 0.5:
            print(1, int(y_data[index].item()))
        else:
            print(0, int(y_data[index].item()))

Intercept key code ：

train_loader = DataLoader(dataset=dataset,
                          batch_size=749,
                          shuffle=False,
                          num_workers=1)

batch_size The size of each batch of training , If the total number of data pieces is 7490 strip , So at this time train_loader Will produce 10 individual batch.shuffle It's a shuffle , It is to disorder the order of data and then put the data into 10 individual batch in .

And now why do I , To put  batch_size Set to 29, And set it not to shuffle . Because I want to compare with the previous process .

Because there are only 749 data , Then there will only be one batch, One batch Inside is all the data without disordering the order . This is the same as the previous training process .

I did this because I found , Used DataLoader After that, the training process became unusually slow , After changing to the above configuration , I thought the speed would be the same , But it's actually slow 10 More than ！

What hurts more is , If  batch_size Make it smaller ,loss There is no way to converge directly , If there are too many cyclic tests, an error is reported ：

return _winapi.DuplicateHandle( PermissionError: [WinError 5] Access denied

this DataLoader How to use it ？ Which God will give you some advice ？