Deployment

This section covers exporting trained models and using them in production.

Overview

Once you’ve trained a successful pipeline, NIRS4ALL makes it easy to deploy to production or share with collaborators.

📦 Export Bundles

Package trained pipelines as standalone .n4a bundles.

Production

Export and Deployment

🔄 Model Reuse

Load and use exported models for new predictions.

Inference

Prediction and Model Reuse

🔧 Transfer Learning

Adapt models to new instruments or sample types.

Adaptation

Retrain and Transfer Learning

Deployment Workflow

        graph LR
    A[Train Pipeline] --> B[Evaluate]
    B --> C{Good?}
    C -->|Yes| D[Export .n4a]
    C -->|No| A
    D --> E[Production Use]
    D --> F[Share/Collaborate]

Quick Example

Export a Model

import nirs4all

# Train and get results
result = nirs4all.run(pipeline, dataset="data/")

# Export the best model
result.export("exports/best_model.n4a")

Use an Exported Model

import nirs4all

# Load and predict
predictions = nirs4all.predict(
    bundle="exports/best_model.n4a",
    data="new_samples/"
)

Bundle Contents

A .n4a bundle contains:

Trained model weights
Preprocessing transformers (fitted)
Pipeline configuration
Metadata (training date, metrics, etc.)