Additions #1
Additions #1
Conversation
There was a problem hiding this comment.
I've looked at the separate files now, and added quite some remarks. In general, I think your code needs some refactoring and cleaning up.
I am not sure whether you really need to create this fairseq release, or whether you could also add your own functionality in the worker repo. Most items you added are additions rather than modifications, except the fairseq/modules/transformer_layer, in which your modifications are so impactful that it seems like you are actually creating new/different objects.
I did not try to run your code yet. Please provide a description of what/how to run when you create a PR :)
| preds = list(logging_output)[0] # First value | ||
| probabilities = list(logging_output)[1] # Second value | ||
|
|
||
| return preds, probabilities |
There was a problem hiding this comment.
You are modifying the blueprint of the function here, which affects the compatibility. Is that really necessary, or could you also create a new function with your desired return type?
| loss, sample_size, logging_output = criterion(model, sample) | ||
| return loss, sample_size, logging_output | ||
| logging_output = criterion(model, sample) #calling the same criterion loss function #loss deleted MODIFIED | ||
| #print("Prediction", sample_size) #MODIFIED |
There was a problem hiding this comment.
Remove spurious print comments (old print statements)
| @@ -0,0 +1,130 @@ | |||
| # Copyright (c) Facebook, Inc. and its affiliates. | |||
There was a problem hiding this comment.
Is this Facebook code or your own?
|
|
||
| from fairseq import models | ||
|
|
||
| model = models.build_model(args, self) #/home/gsir059/Documents/PhD/MulFie/fairseq/fairseq/models/__init__.py' |
There was a problem hiding this comment.
It seems like you based your version on an existing example file. It's more clear if you first commit theexample file, and then commit your changes to it. That makes it easier to review your additions/modifications.
Could you point me to the file you used as a starting point? Or even better, try to insert it post hoc? See https://printf2linux.wordpress.com/2012/04/09/insert-a-commit-in-the-past-git/ for an instruction of inserting a commit in your history
| # self.emotion_dictionary = { #modei senti 2 class | ||
|
|
||
| # '0': 0, | ||
| # '1':1 | ||
| # } |
There was a problem hiding this comment.
Alternative encoding is commented out? Should this be configurable?
| #print("DEBUG dataset sources length") | ||
| #print(len(sources)) | ||
| #print(sources) |
There was a problem hiding this comment.
Here as well, please provide the basis that you extended explicitly.
Also, the functions are quite complex: refactor/break them down into meaningful, testable bits. Provide a description (ideally docstring) explaining what each part should be doing.
| # pred_real_i='pred_mos_real' | ||
| # truth_real_i='truth_mos_real' | ||
|
|
||
| # logging_output.update({truth_real_i : targets.view(-1).cpu().detach().numpy()}) | ||
| # logging_output.update({pred_real_i : logits.view(-1).cpu().detach().numpy()}) | ||
|
|
|
|
||
| exclude_zero=True | ||
|
|
||
| #This gives a problem when running with with batch size of one and that batch consist of a '0' as the truth |
There was a problem hiding this comment.
Not clear what you are trying to accomplish with exclude_zero
No description provided.