PyDeepFlow v1.0.0 🥳

Pydeepflow Release Notes

Version: 1.0.0
Release Date: 23rd October 2024

Overview

This is the initial release of Pydeepflow, a Python library designed for building and training artificial neural networks (ANNs). The library provides a flexible framework for various classification tasks, including support for batch normalization, dropout, and L2 regularization.

Features

• Multi-Layer Artificial Neural Network (ANN):

  • Implementation of a versatile ANN class that supports both binary and multi-class classification tasks.
  • Configurable architecture with customizable hidden layers, activation functions, and loss functions.

• Forward and Backward Propagation:

  • Efficient forward propagation through hidden and output layers.
  • Backpropagation implementation with gradient clipping to enhance training stability.

• Regularization Techniques:

  • L2 regularization to reduce overfitting.
  • Dropout layers for improved generalization during training.

• Batch Normalization:

  • Included batch normalization to stabilize and accelerate training by normalizing layer inputs.

• Checkpointing:

  • ModelCheckpoint class for saving and loading model weights during training to facilitate easy resumption.

• Cross-Validation Support:

  • CrossValidator class to perform K-fold cross-validation for robust model evaluation.

• Training Utilities:

  • Support for early stopping and learning rate scheduling to optimize training performance.

Improvements

• Weight Initialization:

  • Enhanced weight initialization using He initialization for better convergence of deep networks.

• Training History Tracking:

  • Mechanism to track and visualize training and validation loss and accuracy over epochs.

• Verbose Logging:

  • Training progress updates to provide insights into model performance during training.

Bug Fixes

• Gradient Clipping:

  • Fixed issues with gradient clipping implementation to prevent exploding gradients during training.

• Output Layer Handling:

  • Improved handling of output layers for both binary and multi-class classification to ensure proper activation function application.

Usage Instructions

To get started with Pydeepflow, please refer to the GitHub Wiki for detailed instructions on installation, configuration, and usage examples.

Installation

pip install pydeepflow

Acknowledgments

Thank you to all the contributors for your support!