PioNIR

PioNIR is a Python library designed for the processing and analysis of near-infrared (NIR) spectra. It offers a suite of chemometric tools for preprocessing, modeling, and evaluation, facilitating the development of spectroscopy workflows. The library tightly integrates with the scientific Python ecosystem, with NumPy, SciPy, and scikit-learn.

Installation

The latest release is available on PyPI and can be installed via pip.

pip install pionir

For development purposes, use uv to manage the installation.

uv sync --dev

The project uses ruff for linting and formatting, ty for type checking, and pytest for testing.

uvx ruff check src
uvx ty check src
uv run pytest

Features

Data Representation

PioNIR provides specialized Spectrum and SpectrumCollection classes to handle NIR spectra. These classes serve as the foundation for handling individual measurements and larger datasets within the library.

from pionir import Spectrum, SpectrumCollection, Metadata

# x, y: NumPy ndarrays
s = Spectrum(x=..., y=..., metadata=...)  # Single spectrum
c = SpectrumCollection([s])  # Collection of spectra

Unified Interface with Single Dispatch

PioNIR uses a single dispatch approach, allowing the same function interface to handle both individual Spectrum objects and SpectrumCollection instances as well as plain NumPy ndarray.

from pionir import Spectrum, SpectrumCollection
from pionir.scatter import snv

s = Spectrum(...)
c = SpectrumCollection(...)

# The same function handles both single spectra and collections
s_snv = snv(s)
c_snv = snv(c)

scikit-learn Integration

The library is designed for compatibility with scikit-learn, providing transformers that can be directly incorporated into pipelines.

from sklearn.decomposition import PCA
from sklearn.linear_model import LinearRegression
from sklearn.pipeline import Pipeline

from pionir import SpectrumCollection
from pionir.sklearn.transformer import SNVTransformer

c = SpectrumCollection([...])

pcr_pipeline = Pipeline([
    ("snv", SNVTransformer()),
    ("pca", PCA(n_components=5)),
    ("regression", LinearRegression())
])

pcr_pipeline.fit(c.y, ...)
pcr_pipeline.predict(...)