Multiple outputs from model(input)

1

Hello,

My apologies upfront if this is a naive question, new to pytorch, coming from keras/lasagne, but I haven’t had much luck searching for answers online so thought I would post here.

There is an implementation of Pytorch here. The model that we are using is cifar_shakeshake26 in architectures.py here.

We have tried to adapt their implementation to our dataset by adding our own model that would accept a 70x70 image. However, when we run, we get this error during validation:

INFO:main:--- training epoch in 13.531290054321289 seconds ---
INFO:main:Evaluating the primary model:
Traceback (most recent call last):
  File "main.py", line 421, in <module>
    main(RunContext(__file__, 0))
  File "main.py", line 103, in main
    prec1 = validate(eval_loader, model, validation_log, global_step, epoch + 1)
  File "main.py", line 331, in validate
    output1, output2 = model(input_var)
ValueError: too many values to unpack (expected 2)

What I don’t understand, is how model(input_var) is returning two variables.

If more information is needed to answer this question, please let me know and i’ll provide it.

Thank you in advance!

-Tommy

2

Where is the code for model?

From the error it sounds like model(input_var) is returning more than 2 values (or maybe 1 value?). This is possible if the forward function of model returns a tuple.

3

thanks for the reply. see my edited post.

4

I’m not too sure what’s going on because the model is intended to return two outputs: https://github.com/CuriousAI/mean-teacher/blob/99cd11f646ec2e4cc6f19836be96a72c55923e00/pytorch/mean_teacher/architectures.py#L161 .

One thing you can do is set a pdb breakpoint there and see if the model output is actually two outputs.

5

Thank you very much, that pretty much answers my question. Greatly appreciate you taking the time to look at the code : )