How to predict only one test sample in pytorch model?

1

I want to give an input to the model and predict the class label.How can I write my code?Please answer.Thanks in advance!

2

You can use the batch size as 1 if you want to do single sample using pytorch test data loader.

If not you can create tensor of same size and shape as u r input and just pass it as model(input)

3

Thanks a lot for answering! I get :
predict is: tensor([[0.1559, 0.4340, 0.3785, 0.0226, 0.0032, 0.0519, 0.1933]],
device=‘cuda:0’, grad_fn=)
When tensor is converted to numpy: predict is: [[-0.2556236 0.03545919 -0.38335416 0.3578361 -0.42080054 0.3241588
0.24587154]] .I want to get class label,and how can I predict class label?Thanks in advance!

4

Try using pred_out, pred_index = torch.max(pred, 1)

Also, try wrapping your inference code with the wrapper: with torch.no_grad():

5

Thanks a lot! pred_out is: tensor([0.6067])
pred_index is: tensor([1])
The outputs are not same when I run again and again.Why this happens?Please answer how to get same result when it is run again and again.Thanks in advance!

6

Sure.

For getting deterministic results, you need to set the seed.

Refer to this post: Reproducibility with all the bells and whistles

1 Like
7

can you explain why:

torch.max(output.data, 1)

works?

I was reading the docs (https://pytorch.org/docs/stable/generated/torch.max.html) and it’s still not entirely clear to me. This is the sentence that doesn’t make sense to me:

Returns a namedtuple (values, indices) where values is the maximum value of each row of the input tensor in the given dimension dim.

I don’t understand what the max value of a row mean given a specific slice of a tensor. Rows only exist in 2D tensors (matrices), at least to me. Can you clarify?