dag-ml¶
dag-ml is the Rust execution coordinator for leakage-safe, reproducible ML
pipelines. It owns graph compilation, phases, folds, OOF joins, controller
dispatch, lineage, artifact/cache contracts and deterministic scheduling. Data
storage and feature-buffer ownership live in the companion dag-ml-data repo.
This site is the contributor and integration entry point for the producer-side contracts used by future nirs4all and nirs4all-core integrations. It does not wire nirs4all directly.
Start Here¶
Need |
Page |
|---|---|
Build and validate locally |
|
Understand runtime boundaries |
|
Integrate over C ABI |
|
Check the supported release surface |
|
Map aggregation across |
|
Run release performance probes |
|
Map nirs4all parity capabilities |
|
Inspect shared contracts |
|
Review accepted decisions |
|
Pick an example by audience |
|
API References¶
Rust core API: https://docs.rs/dag-ml-core/latest/
Rust facade API: https://docs.rs/dag-ml/latest/
C ABI source:
crates/dag-ml-capi/include/dag_ml.hPython binding source:
crates/dag-ml-pyWASM binding source:
crates/dag-ml-wasm
The nirs4all ecosystem¶
Typed sample-aligned multi-source data contracts — the data layer dag-ml consumes (shared, JSON-identical contracts).
Main Python modelling library — pipelines, SpectroDataset, predictions.
Portable C-ABI PLS/NIRS engine (libn4m) + bindings.
Rust readers for ~58 NIRS/spectroscopy file formats.
Dataset-assembly bridge → SpectroDataset.
Curated DOI-pinned NIRS dataset catalog (n4a-datasets).
Canonical portable aggregate distribution (Rust, Python, R, WASM, MATLAB/Octave).
