About Vectorphile
One import.
Every layer.
Vectorphile is a research-first Python library project building a unified ML/DL computational stack. Tensors, autograd, neural networks, classical machine learning, DataFrames, linear algebra, signal processing — one coherent system, zero external framework dependencies.
Why we're building this
Python's ML ecosystem is fragmented. Tensors live in one library, classical ML in another, DataFrames in a third, signal processing somewhere else. Bridging code is boilerplate — it wastes time and creates type-system friction.
Vectorphile is built on a single conviction: every numeric primitive in a data pipeline should share the same underlying representation and operator set. No type fragmentation. No hidden conversions. No bridging code.
We are building Phase 1 entirely in pure Python and NumPy — readable, inspectable, and deployable anywhere Python works. Phase 2 brings hand-written C extensions and CUDA kernels, same API, real hardware performance.
At a glance
What we stand for
Open by Default
Every line of Vectorphile is public, Apache-2.0 licensed, and reproducible. No paywalls, no closed checkpoints, no gated APIs.
No Hidden Magic
No opaque C extensions, no global dispatch tables, no framework-level implicit state in Phase 1. What you read is what runs — readable and traceable end-to-end.
Research Portability
From a quick prototype to a full training loop — zero idiom changes required. High-level APIs and raw tensor ops share the same surface.
Radical Transparency
We document architecture decisions, known limitations, and what each module actually does. No marketing claims without reproducible code.
Progressive Depth
Phase 1 is pure Python and NumPy. Phase 2 replaces the backend with C + CUDA kernels — same API, real GPU performance. No breaking changes across phases.
Community Over Moat
Vectorphile is built in public, for the community. Contributions, forks, and derivative work are welcome and encouraged.
Get involved
Vectorphile is open source and community-driven. Open an issue, contribute a module, or just star the repo — every bit helps.