v0.70

New Feature

• Introduced TabularBinning

v0.60

New Features

• Support modular designs for embedding and backbone module.
  The currently supported modules are:
  • Embedding modules:
    identity
feature_tokenizer (from Revisiting Deep Learning Models for Tabular Data)
  • Backbone modules:
    mlp
transformer

Enhancements

• Split reconstruction head for DAE and VIME: Implement separate reconstruction heads for categorical and continuous features, enhancing the handling of heterogeneous data.

v0.50

1. Module Optimization
   Enhanced performance by optimizing several modules, resulting in more efficient and faster execution.

2. Refactored Forward and Loss Calculation Logic
   Moved the forward and loss calculation logic from Lightning modules to functional modules.
   This change improves modularity and maintainability of the codebase.

3. Code Refactoring
   Refactored the overall codebase for better readability, and maintainability.

4. New Freeze Encoder Flag
   Added a freeze_encoder flag to the set_second_phase method. This flag allows users to easily freeze or unfreeze the encoder during the second phase of training, providing greater flexibility in model training.

v0.41

Bug Fix

Removed unnecessary print statements

v0.40

1. Release SwitchTab
2. Add the weight initialization when initializing the model

v0.30

Release Denoising AutoEncoder (DAE).

v0.21

Incorporate data_hparams into Config

v0.20

Release Lightning Modules for VIME, SubTab, and SCARF.

v0.10

Release of nn.Module Implementations for VIME, SubTab, and SCARF, Including Their Corresponding Datasets