How to import our own data using torch instead of pandas?

surya_teja · November 4, 2017, 1:58pm

As we are using pandas to import the data till now. How can we import data using pytorch without using pd.read_csv()?

smth · November 5, 2017, 12:31am

whyy not use pandas, and what do you mean by “import data using pytorch”. pytorch is just a python package.

surya_teja · November 7, 2017, 2:24pm

I generally do with pandas. I am trying to explore Pytorch.

I tried this

df = torch.utils.data.DataLoader(‘train.csv’)

I am in working directory with train.csv as my data and i am trying to laod that with pytorch and i am getting an error:

TypeError: __init __() missing 1 required positional argument: ‘target_tensor’

alexis-jacq · November 7, 2017, 2:34pm

Dataloader works with directories containing data files (for example images), in your case you just want to open a csv, so you should keep using pandas. After this you can still convert your pandas arrays into pytorch tensors.

SimonW · November 7, 2017, 3:03pm

There is comprehensive document on DataLoader… You can’t just pass in something without seeing what its inputs are and expect it to work. http://pytorch.org/docs/0.2.0/data.html#torch.utils.data.DataLoader