How to make dataloader return tensor?

Here, my_dataloader is returning lists instead of torch tensor. please help.

tensor_x = torch.tensor(Mat)      #Mat is a 2d numpy array
my_dataset = TensorDataset(tensor_x)

my_dataloader = DataLoader(my_dataset,batch_size=8, shuffle = True)

Hello, have you tried:

for i, (data) in enumerate(my_dataloader):

What is the output of this?

1 Like

If your data loader is returning a list of tensors, and you want a single tensor you could use torch.stack to stack all elements within the list to a single Tensor.

list = [ torch.randn(5,5) for _ in range(10) ] 
tensor = torch.stack(list)
tensor.shape() #returns torch.Size([10, 5, 5])
1 Like

Tried what you said. It gave this : AttributeError: 'list' object has no attribute 'type'
What should I do now? Please help

I want my dataloader to return a batch of tensors

Does your list object contain 2 tensors? One for the input and one for the target/label?

1 Like

Just inputs , no targets as it is unsupervised learning

But the object returned by your Dataloader seems to be a list, so perhaps it’s just a list of a single Tensor? Or perhaps it’s return a batch of lists?

1 Like


You should change the collate_fn, something like this

import torch
from import TensorDataset, DataLoader

def collate_fn(batch):
    batch =[sample[0].unsqueeze(0) for sample in batch], dim=0)
    return batch

tensor_x = torch.tensor(Mat)      #Mat is a 2d numpy array
dataset = TensorDataset(tensor_x)
loader = DataLoader(dataset, batch_size = 8, shuffle=True, collate_fn=collate_fn)