v0.0.2

Changelog

All notable changes to C-ML will be documented in this file.

The format is based on Keep a Changelog,
and this project adheres to Semantic Versioning.

[0.0.2] - Pre-Alpha - 2025-11-08

Added

• Complete Automatic Differentiation System: Implemented dynamic computation graphs that are built on-the-fly during forward pass, with automatic gradient computation during backward pass. Features include topological sorting for correct gradient flow, chain rule application, and gradient accumulation across multiple backward passes.

• Forward and Backward Operations: All tensor operations now support autograd with proper forward/backward implementations for binary operations (add, sub, mul, div, pow), unary operations (exp, log, sqrt, trigonometric functions), activation functions (ReLU, Sigmoid, LeakyReLU), and reduction operations (sum, mean).

• Loss Functions with Autograd: Implemented comprehensive loss functions including MSE, MAE, Binary Cross Entropy, Cross Entropy, Huber, and KL Divergence - all with full autograd support for seamless integration into training loops.

• Gradient Checkpointing: Added memory-efficient backward pass implementation that recomputes activations instead of storing them, trading computation for memory. Configurable via autograd_set_checkpointing() API.

• Real-Time Visualization UI: Built comprehensive React-based dashboard with Server-Sent Events (SSE) for live updates during training. Features include:

  • Real-time training metrics visualization (loss curves, accuracy, learning rate, gradient health)
  • Interactive computational graph visualization using Cytoscape
  • Model architecture view with interactive exploration
  • Automatic launch capability via VIZ=1 environment variable
  • FastAPI backend with SSE streaming for real-time data updates

• Documentation and Examples: Added comprehensive user guide (docs/AUTOGRAD.md), detailed implementation notes (docs/AUTOGRAD_IMPLEMENTATION.md), example code (examples/autograd_example.c), and complete test suite covering all operations.

Changed

• Updated build system to include autograd sources in Makefile and CMakeLists.txt
• Enhanced tensor operations to integrate with autograd system

Technical Details

This release adds a complete autograd system with real-time visualization to C-ML, enabling gradient-based training of neural networks with automatic differentiation and live monitoring capabilities.

Status: Pre-Alpha
Branch: autograd-integration

---

[0.0.1] - Initial Release

Added

• Basic tensor operations
• Neural network layers (Linear, Conv2d, BatchNorm2d, Pooling, Activations, Dropout)
• Optimizers (SGD, Adam)
• Basic training utilities
• Memory management system

v0.0.1

C-ML v0.0.1-pre Release

We're excited to announce the first pre-release of C-ML, version 0.0.1-pre! This release marks a significant milestone in the development of our lightweight machine learning library in C. While still in its early stages, this version provides a foundation for building and experimenting with neural network components.

What's Included

This pre-release includes the following core modules:

• Layers:
  • Dense
  • Dropout
• Activations:
  • ReLU
  • Sigmoid
  • Tanh
  • Softmax
  • ELU
  • Leaky ReLU
  • Linear
• Loss Functions:
  • Mean Squared Error
  • Binary Cross-Entropy
  • Focal Loss
  • Mean Absolute Error
  • Mean Absolute Percentage Error
  • Root Mean Squared Error
  • Reduce Mean
• Optimizers:
  • SGD
  • Adam
  • RMSprop
• Preprocessing:
  • Label Encoding
  • One-Hot Encoding
  • Standard Scaler
  • Min-Max Scaler
• Regularizers:
  • L1
  • L2
  • Combined L1-L2

Important Notes

• This is a pre-release, so expect potential bugs and incomplete features.
• Your feedback is highly appreciated to guide further development!

Changes Since Initial Commit

• Workflow Added: CI workflow added for automated testing and deployment.
• MkDocs Support: Integrated MkDocs for documentation generation and deployment.
• Documentation: Added comprehensive documentation for all modules.
• Testing: Added unit tests for all modules to ensure code correctness and robustness.
• Makefile: Streamlined the build process with automated source directory inclusion.
• Licensing: Added the "Don't Be a Jerk" Non-Commercial Care-Free License (DBaJ-NC-CFL).
• Example Usage: Provided a basic example in main.c to demonstrate library usage.
• Code Refactoring: Improved code structure and readability.
• Bug Fixes: Patched minor bugs and memory allocation issues.
• Spelling and Comment Updates: Fixed spelling mistakes and removed unnecessary comments.
• Logo Added: Added a logo to the project.
• Header File Migration: Migrated from a single header file to separate header files for better organization.