Failure in trying to reproduce the example notebook provided by the repository.

[cae-ins]

Description

When training the model, an IndexError: Target [X] is out of bounds occurs due to a mismatch between the model’s output dimension (num_classes) and the actual number of classes in the dataset.

Steps to Reproduce

Initialize the model with an incorrect num_classes (e.g., smaller than the maximum label index in the dataset).

Train the model on data where labels exceed the declared num_classes - 1.

Error occurs during loss calculation (e.g., cross-entropy).

Error Message

IndexError: Target [X] is out of bounds.  
# (Full traceback pointing to the loss function)  

Expected Behavior

The model should either:

Automatically infer num_classes from the training data.

Throw a clear validation error during initialization if num_classes is misconfigured.

[cae-ins]

Hey. It seems you did not add the correction in the stable version of the package downloadable by pip install torchFastText. By the way, getting a strange error when trying to install the development version :

  Downloading https://github.com/InseeFrLab/torch-fastText.git
     - 0 bytes ? 0:00:00
     | 32.8 kB ? 0:00:00
     - 65.5 kB 161.4 kB/s 0:00:00
     / 196.6 kB 403.5 kB/s 0:00:00
     \ 294.5 kB 429.8 kB/s 0:00:00
  ERROR: Cannot unpack file [C:\Users\USER\AppData\Local\Temp\pip-unpack-yzclc_h2\torch-fastText.git](file:///C:/Users/USER/AppData/Local/Temp/pip-unpack-yzclc_h2/torch-fastText.git) (downloaded from [C:\Users\USER\AppData\Local\Temp\pip-req-build-da89aaqe](file:///C:/Users/USER/AppData/Local/Temp/pip-req-build-da89aaqe), content-type: text/html; charset=utf-8); cannot detect archive format
ERROR: Cannot determine archive format of [C:\Users\USER\AppData\Local\Temp\pip-req-build-da89aaq](file:///C:/Users/USER/AppData/Local/Temp/pip-req-build-da89aaqe)

[cae-ins]

Still same error unfortunately. With the development package on git.

[meilame-tayebjee]

Hi ! Thanks for raising the issue. The commit b2ef31f is raising an error when num_classes is smaller than the max of the values in y_train, provided that you call the .build() method, as in the notebook.

Could you please provide us with a reproducible code that gives you the error you want to solve ?

[micedre]

@cae-ins Did you manage to resolve your issue ?