Concepts

This section explains the core ideas behind nirs4all. Each page covers one concept at a high level – what it is, why it matters, and how the pieces fit together. If you prefer to learn by reading code, jump to Interactive Example Browser. If you need exact parameter lists, see Reference.

Suggested Reading Order

Start with Pipelines and Datasets – they are the two objects you pass to every nirs4all.run() call. Then read Cross-Validation to understand how performance is measured. The remaining pages build on those foundations.

Pipelines --> Datasets --> Cross-Validation
    |                           |
    +---> Branching & Merging   |
    +---> Generators            |
    +---> Augmentation          |
                                |
              Predictions & Deployment

Pages

Pipelines

What a pipeline is, the four step types, and how steps are executed.

Pipelines

Datasets

SpectroDataset, partitions, multi-source data, signal types, and repetitions.

Datasets

Cross-Validation

Fold-based evaluation, out-of-fold predictions, scoring, and refit.

Cross-Validation

Branching and Merging

Parallel sub-pipelines, separation branches, merge strategies, and stacking.

Branching and Merging

Generators

Express a search space in one pipeline and let nirs4all expand it into many variants.

Generators

Augmentation

Expand small datasets with synthetic training samples and feature views.

Augmentation

Predictions and Deployment

RunResult, export, predict, explain, retrain, and the Session API.

Predictions and Deployment