Projects

  • Type4Py:  Deep similarity learning-based type inference for Python

    • The neural model is trained on the ManyTypes4Py dataset and achieves 79.2\% accuracy considering the prediction of type annotations for parameters, return values, and variables.
    • It has a Flask app that exposes an endpoint to query the pre-trained model and return predicted type information for a given Python source file.
    • A VS Code extension that provides ML-based type auto-completion for Python and assists developers to gradually add type annotations to their codebases (has more than 1800 installs)

  • LIBTwinSVM: A Library for Twin Support Vector Machines

    • A simple and user-friendly Graphical User Interface
    • An improved and fast optimizer (clipDCD) is implemented in C++.
    • Supports LinearRBF, and Rectangular kernel.
    • The OVA and OVO estimators are compatible with scikit-learn tools such as GridSearchCV, cross_val_score, etc.
    • The classification results can be logged during the grid search process to not lose results in case of power failure.
    • feature-rich visualization tool to show decision boundaries and geometrical interpretation of TwinSVMs.
    • The best-fitted classifier can be saved on the disk after the grid search process.
    • The pre-trained models can be loaded and evaluated on the test samples.

  • LightTwinSVM: Simple and Fast Implementation of Standard TwinSVM Classifier

    • simple console program for running TwinSVM classifier.
    • Fast optimization algorithm. The clipDCD algorithm is implemented in C++.
    • Binary and Multi-class classification (One-vs-All & One-vs-One) are supported.
    • LinearRBF kernel and Rectangular are supported.
    • The classifier can be evaluated using either K-fold cross-validation or Training/Test split.
    • It supports grid search over C and gamma parameters.
    • A detailed classification result will be saved in a spreadsheet file.