Pytest unit tests, new losses support, and normalization enhancement

[franckma31]

The Torchlip library has undergone several updates to align more closely with the functionalities provided by deel-lip (TensorFlow). Here are the key changes:

• Transition to pytest for unit tests, sharing a common test suite with deel-lip. The utils_framework.py file has been introduced to facilitate adaptation of names and parameters.

• Enhancements in the available loss functions:

  • Binary and Multiclass KR and HKR: now support any target style (0,1), (-1,1), ... [target>0 for true value],
  • Multi-GPU Support: binary and multiclass KR and HKR supports multi-gpu mode with preprocessed target
  • Modified HKR Loss: the $\alpha$ parameter (required in [0,1]) with (1- $\alpha$ )HKRLoss+ $\alpha$ HingeLoss (for backward compatibility a warning is raised in case of $\alpha>1$ replacing the given value by $\frac{\alpha}{1+\alpha}$)
  • Tau Cross-Entropy Losses: Support has been added for cross-entropy (binary or categorical) losses with temperature scaling.
  • Reduction Parameters: Loss functions now accept the 'reduction' argument with options: 'mean', 'none', or 'sum'.

• Normalizers (Spectral and Bjorck): Updated with a stopping criterion based on an epsilon value (eps) rather than a fixed number of iterations

• Bjorck parenthesis trick added (reduce computational complexity)

[cofri]

So many valuable changes! Nice and big work