Releases: basf/mamba-tabular
New Version Release 1.0.0
v1.0.0: Major Update for Mambular
This release marks a significant upgrade to Mambular, introducing new models, enhanced functionality, and improved efficiency. Below is an overview of the key changes:
🚀 New Models
- TabM: A cutting-edge tabular deep learning model optimized for performance and flexibility.
- NODE: Incorporates Neural Oblivious Decision Ensembles for more robust handling of tabular data.
- NDTF: Neural Decision Tree Forest, combining decision tree logic with deep learning capabilities.
- TabulaRNN: A recurrent neural network tailored for tabular tasks, with configurable options to use GRU or LSTM cells for sequence modeling.
🎛️ Hyperparameter Optimization
- Integrated support for hyperparameter optimization:
- Built-in Bayesian optimization for more advanced tuning.
- Fully compatible with Scikit-learn's optimization framework, enabling seamless integration for all models.
⚡ Efficiency Improvements
- Leveraging the new mamba-ssm package for a more efficient implementation of the Mamba framework, ensuring faster runtime and reduced memory usage.
🛠️ Enhanced Preprocessing
- Expanded preprocessing options for greater control over feature transformations.
- Improved feature information handling to better accommodate various dataset types and structures.
🧬 Improved Embedding Layers
- New embedding layers, including PLR.
- Customizable activation functions for enhanced flexibility in embedding generation.
This release sets the foundation for continued innovation in tabular deep learning. Feedback and contributions are welcome!
Release v0.2.4
What's Changed
- sklearn base_modules: Modified conditional checks to use if X_val is not None instead of if X_val in the build_model and fit methods. by @AnFreTh in #142
- mambular/data_utils/datamodule.py: Ensured that keys are converted to strings when constructing cat_key, binned_key, and num_key in the setup and preprocess_test_data methods. by @AnFreTh in #142
Full Changelog: v0.2.3...v0.2.4
Contributors
Assets 2
Release v0.2.3
- Including Quantile Regression
- Fixing param count bug in sklearnbaselss
v0.2.2
Release v0.2.1
What's Changed
- Included new models (MambaTab, TabulaRNN)
- Added utility functionality to model build
- Improved Embedding layers
- Added AB layernorm and weight decay to Mamba
- Added score function to sklearn base classes
Full Changelog: v0.1.7...v0.2.1
Hotfix Release v0.1.7
New Release v0.1.6
New Version Release Highlights:
-
Addition of New Models: We've expanded our model suite to include the following architectures:
- FT-Transformer: Leverages transformer encoders for improved performance on tabular data.
- MLP (Multi-Layer Perceptron): A classical deep learning model for handling a wide range of tabular data tasks.
- ResNet: Adapted from the classical ResNet architecture and proven to be a good baseline for tabular tasks.
- TabTransformer: Utilizes transformer-based models for categorical features.
-
Bidirectional and Feature Interaction Capabilities: Mambular now includes bidirectional capabilities and enhanced feature interaction mechanisms, enabling more complex and dynamic data representations and improving model accuracy.
-
Architectural Restructuring: The internal architecture has been restructured to facilitate the easy integration of new models. This modular approach simplifies the process of extending Mambular with custom models.
-
New Preprocessing Methods: We have introduced new preprocessing techniques to better prepare your data for modeling:
- Quantile Preprocessing: Transforms numerical features to follow a uniform or normal distribution, improving robustness to outliers.
- Polynomial Features: Generates polynomial and interaction features to capture more complex relationships within the data.
- Spline Transformation: Applies piecewise polynomial functions to numerical features, effectively capturing nonlinear relationships.
Beta release: v0.1.4
Mambular: Tabular Deep Learning with Mamba Architectures
Introduction
Mambular is a Python package that brings the power of Mamba architectures to tabular data, offering a suite of deep learning models for regression, classification, and distributional regression tasks.
Features
Comprehensive Model Suite: Includes modules for regression (MambularRegressor), classification (MambularClassifier), and distributional regression (MambularLSS), catering to a wide range of tabular data tasks.
- State-of-the-Art Architectures: Leverages the Mamba architecture, known for its effectiveness in handling sequential and time-series data within a state-space modeling framework, adapted here for tabular data.
- Seamless Integration: Designed to work effortlessly with scikit-learn, allowing for easy inclusion in existing machine learning pipelines, cross-validation, and hyperparameter tuning workflows.
- Extensive Preprocessing: Comes with a powerful preprocessing module that supports a broad array of data transformation techniques, ensuring that your data is optimally prepared for model training.
- Sklearn-like API: The familiar scikit-learn fit, predict, and predict_proba methods mean minimal learning curve for those already accustomed to scikit-learn.
- PyTorch Lightning Under the Hood: Built on top of PyTorch Lightning, Mambular models benefit from streamlined training processes, easy customization, and advanced features like distributed training and 16-bit precision.