当我将数据加载器张量转换为 numpy 数组时，Pytorch 张量数据受到干扰

Question

I am using a simple train loop for a regression task. To make sure that regression ground-truth values are the same as what I expect in the training loop, I decided to plot each batch of data. However, I see that when I convert the data loader's tensor to numpy array and plot it, it is disturbed. I am using myTensor.data.cpu().numpy() for the conversion.

My code is as below:

 train_ds = TensorDataset(x_train, y_train) train_dl = DataLoader(train_ds, batch_size = 32, shuffle = True, num_workers = 0, drop_last = True) for epoch in range(epochs): model.train() for i, (x, y) in enumerate(train_dl): x = x.cuda() y = y.cuda() yy = y.data.cpu().numpy() pyplot.plot(yy[0: 32, 0]) pyplot.show()

Answer 1

I think it is because I set shuffle = True in data loader. If I set it to false, it is fine. However, How can I shuffle training batches after each epoch if I set shuffle = False in data loader then?