DonanML

DonanML is a cross-platform open-source machine learning framework designed to provide a compact, beginner-friendly environment for building and deploying models. It aims to balance ease of use with performance, offering a Python interface suitable for education, prototyping, and lightweight production tasks.

The library provides core tensor operations, automatic differentiation, and a modular neural network API. It supports common layers (dense, convolutional, recurrent), training utilities (loss functions, optimizers, learning rate scheduling), and data loading utilities for common formats. Model serialization enables saving and loading trained models in a portable format.

DonanML's architecture centers on a minimal core engine, an autograd subsystem, a set of neural network modules, and a data pipeline. The design emphasizes a clean API, with optional backends for acceleration on CPUs and GPUs. It favors deterministic behavior for educational purposes and provides debugging aids.

Development is community-driven, hosted on a public repository with permissive licensing. Contributions include documentation, backends, and example projects across computer vision, natural language processing, and tabular data tasks. The project is commonly used in classrooms and for quick experimentation, rather than as a replacement for industrial-scale frameworks.

Related projects include major libraries such as TensorFlow, PyTorch, and JAX. DonanML remains a niche option for learners and researchers seeking a lightweight alternative that highlights core ML concepts without heavy infrastructure.