TFxKit

TensorFlow Utilities for NeuralNet Modeling Workflows

TFxKit is a modular Python package that provides a collection of tools, components, and utilities for building, training, and evaluating machine learning models.

It is designed with:

• Hydra/OmegaConf for flexible configuration management
• Keras/TensorFlow for model building
• Scikit-learn & Pandas for data preprocessing and tabular input support
• Hyperparameter tuning with Keras Tuner
• Visualization with Matplotlib and Seaborn, plothist, shapely, etc.
• Clear modular separation of components for extensibility and reuse
• Comprehensive logging for easy debugging and monitoring

---

Key Features

• ModelFactory: Orchestrates end-to-end ML workflows: data loading, model building, training, evaluation, tuning.

  from tfxkit import ModelFactory
  mf = ModelFactory("/path/to/config.yaml")
  mf.fit()
  mf.attach_predictions()
  mf.make_plots()
  mf.hyper_tune()

• DataManager: Handles file I/O, feature-label extraction, and sample weights.

  mf.test.X # to access the test features
  mf.train.sample_weight # to access training sample weights

• ModelBuilder: Dynamically creates MLP (default) and other user defined models with configurable layers, dropout, batch norm, and more.

  model = mf.builder.define_model()
  mf.summary()

• Trainer: Manages model training with support for early stopping, learning rate scheduling, and custom callbacks.

  history = mf.fit()
  mf.plot_history()

• Evaluator: Computes evaluation metrics and attaches model predictions to test/train sets.

• HyperTuner: Wraps KerasTuner with domain-specific enhancements.

  tuner = mf.hyper_tuner()
  tuner.search(mf.train.X, mf.train.y, ...)

• Hydra CLI: Enables command-line execution, dynamically overriding config parameters.

  tfxkit --config-path /path/to/config.yaml trainer.epochs=50 model.parameters.dropout=0.3

---