Models

This section covers model training, comparison, and optimization in NIRS4ALL.

• Model Training
  • Overview
  • Basic Model Training
  • Cross-Validation
  • Accessing Results
  • Regression Models
  • Classification Models
  • Target Processing
  • Model Persistence
  • Tips and Best Practices
  • Complete Example
  • See Also
• Hyperparameter Tuning
  • Overview
  • Basic Usage
  • Configuration Options
  • Combining with Preprocessing Search
  • Pipeline Generators vs Optuna Tuning
  • Viewing Optimization Results
  • Examples
  • Best Practices
  • See Also
• Deep Learning Models
  • Overview
  • Built-in Architectures
  • Quick Start
  • Model Configuration
  • Available Models
  • Architecture Details
  • GPU Configuration
  • Comparing Deep Learning with Traditional Models
  • Framework Comparison
  • Hyperparameter Tuning for Neural Networks
  • Best Practices
  • Troubleshooting
  • See Also

Overview

NIRS4ALL supports a wide range of machine learning models from different frameworks. This section covers model training workflows.

🎯 Model Training

Train models with cross-validation and evaluate performance.

Essential

Model Training

⚙️ Hyperparameter Tuning

Optimize models with Optuna integration and grid search.

Optimization

Hyperparameter Tuning

🧠 Deep Learning

Use TensorFlow, PyTorch, and JAX models in pipelines.

Neural Networks

Deep Learning Models

Supported Model Frameworks

Framework	Import	Examples
scikit-learn	Direct	PLSRegression, RandomForest, SVM
TensorFlow/Keras	nirs4all[tensorflow]	Sequential, Custom architectures
PyTorch	nirs4all[torch]	nn.Module subclasses
JAX	nirs4all[jax]	Flax/Haiku models

Quick Example

from sklearn.cross_decomposition import PLSRegression
from sklearn.model_selection import KFold

pipeline = [
    MinMaxScaler(),
    KFold(n_splits=5),
    {"model": PLSRegression(n_components=10)}
]

result = nirs4all.run(pipeline, dataset="data/")
print(f"RMSE: {result.best_rmse:.4f}")

Built-in NIRS Models

NIRS4ALL includes specialized models optimized for spectroscopy:

• nicon - 1D CNN for NIR classification

• decon - 1D CNN for NIR regression

from nirs4all.operators.models import nicon, decon

pipeline = [
    MinMaxScaler(),
    KFold(n_splits=5),
    {"model": decon(n_outputs=1)}
]

See Also

• Operator Catalog - All available operators

• Meta-Model Stacking User Guide - Model stacking and ensembles

• Developer Guide - Custom model integration