SpectraKit

Python toolkit for spectral data processing: smoothing, baseline correction, normalization, scatter correction, derivatives, peak analysis, and more.

SpectraKit is a lightweight, pip-installable library for preprocessing and analyzing spectral data from IR, Raman, and NIR spectroscopy. It follows a functional design with NumPy arrays as the primary data type and requires only NumPy + SciPy as core dependencies.

Documentation | API Reference | Project Page | Examples

Installation

pip install pyspectrakit

Note: The PyPI distribution name is pyspectrakit (due to a naming conflict). The import name is simply import spectrakit.

Optional extras for additional functionality:

pip install pyspectrakit[io]         # HDF5 file support
pip install pyspectrakit[cli]        # Command-line interface
pip install pyspectrakit[baselines]  # pybaselines backend (200+ methods)
pip install pyspectrakit[fitting]    # lmfit peak fitting
pip install pyspectrakit[sklearn]    # scikit-learn integration
pip install pyspectrakit[plot]       # Plotting utilities
pip install pyspectrakit[widgets]    # Jupyter interactive viewer
pip install pyspectrakit[all]        # Everything above

Quick Start

import numpy as np
from spectrakit import smooth_savgol, baseline_als, normalize_snv

# Load your spectral data (N spectra, W wavelengths)
spectra = np.loadtxt("data.csv", delimiter=",")

# Process with individual functions
smoothed = smooth_savgol(spectra, window_length=11)
corrected = baseline_als(smoothed, lam=1e6, p=0.01)
normalized = normalize_snv(corrected)

All functions accept both single spectra (W,) and batches (N, W).

Pipeline

Chain steps for reproducibility:

from spectrakit.pipeline import Pipeline

pipe = Pipeline()
pipe.add("smooth", smooth_savgol, window_length=11)
pipe.add("baseline", baseline_als, lam=1e6)
pipe.add("normalize", normalize_snv)

processed = pipe.transform(spectra)

scikit-learn Integration

Use any SpectraKit function in an sklearn pipeline:

from sklearn.pipeline import Pipeline as SkPipeline
from sklearn.decomposition import PCA
from sklearn.svm import SVC
from spectrakit.sklearn import SpectralTransformer

pipe = SkPipeline([
    ("smooth", SpectralTransformer(smooth_savgol, window_length=11)),
    ("baseline", SpectralTransformer(baseline_als, lam=1e6)),
    ("normalize", SpectralTransformer(normalize_snv)),
    ("pca", PCA(n_components=10)),
    ("svm", SVC()),
])

pipe.fit(X_train, y_train)
predictions = pipe.predict(X_test)

Features

Smoothing

Method	Function	Description
Savitzky-Golay	smooth_savgol(y)	Polynomial least-squares smoothing
Whittaker	smooth_whittaker(y)	Penalized least-squares smoother

Baseline Correction

Method	Function	Description
ALS	baseline_als(y)	Asymmetric least squares
SNIP	baseline_snip(y)	Statistics-sensitive peak clipping
Polynomial	baseline_polynomial(y)	Iterative polynomial fit
Rubberband	baseline_rubberband(y)	Convex hull envelope

Normalization

Method	Function	Description
SNV	normalize_snv(y)	Zero mean, unit variance
Min-Max	normalize_minmax(y)	Scale to [0, 1]
Area	normalize_area(y)	Unit area under curve
Vector	normalize_vector(y)	L2 norm = 1

Derivatives

Method	Function	Description
Savitzky-Golay	derivative_savgol(y)	SG polynomial derivative
Gap-Segment	derivative_gap_segment(y)	Norris-Williams derivative

Scatter Correction

Method	Function	Description
MSC	scatter_msc(y)	Multiplicative scatter correction
EMSC	scatter_emsc(y)	Extended MSC with polynomial terms

Spectral Transforms

Method	Function	Description
Kubelka-Munk	transform_kubelka_munk(y)	Reflectance to K-M units
ATR Correction	transform_atr_correction(y, wn)	ATR depth-of-penetration

Operations

Function	Description
spectral_subtract(a, b)	Spectral subtraction
spectral_average(y)	Mean spectrum from batch
spectral_interpolate(y, wn, new_wn)	Resample to new axis

Peak Analysis

Function	Description
peaks_find(y)	Find peaks with scipy.signal
peaks_integrate(y)	Integrate peak regions

Similarity Metrics

Metric	Function	Range
Cosine	similarity_cosine(a, b)	[-1, 1]
Pearson	similarity_pearson(a, b)	[-1, 1]
Spectral Angle	similarity_spectral_angle(a, b)	[0, pi]
Euclidean	similarity_euclidean(a, b)	[0, inf)

I/O Formats

Format	Function	Dependencies
JCAMP-DX	read_jcamp(path)	None
SPC	read_spc(path)	spc-spectra
CSV/TSV	read_csv(path)	None
HDF5	read_hdf5(path) / write_hdf5(spec, path)	h5py
Bruker OPUS	read_opus(path)	None

Optional Backends

Backend	Extra	Description
pybaselines	[baselines]	200+ baseline methods via pybaselines_method()
lmfit	[fitting]	Peak fitting with Gaussian, Lorentzian, Voigt models

Visualization

from spectrakit.plot import plot_spectrum, plot_comparison, plot_baseline

Requires pip install pyspectrakit[plot].

Jupyter Widget

Interactive spectrum viewer for Jupyter notebooks, powered by SpectraView:

from spectrakit.widgets import SpectrumViewer

viewer = SpectrumViewer()
viewer.set_spectrum(wavenumbers, intensities, label="ethanol")
viewer

Requires pip install pyspectrakit[widgets].

Spectrum Container

from spectrakit import Spectrum

spec = Spectrum(
    intensities=np.array([...]),       # (W,) or (N, W)
    wavenumbers=np.array([...]),       # (W,), optional
    metadata={"instrument": "Bruker"},
    source_format="jcamp",
    label="ethanol_ir",
)

CLI

pip install pyspectrakit[cli]

spectrakit info ethanol.dx
spectrakit convert ethanol.dx ethanol.h5

Examples

See the examples/ directory for Jupyter notebooks:

1. Quick Start — basic preprocessing workflow
2. Baseline Methods — comparing correction algorithms
3. Derivatives & Peaks — derivative analysis and peak finding
4. Scatter Correction — MSC vs EMSC vs SNV
5. sklearn Pipeline — classification with preprocessing

Development

git clone https://github.com/ktubhyam/spectrakit.git
cd spectrakit
pip install -e ".[all,dev]"
pytest

See CONTRIBUTING.md for guidelines.

Citation

If you use SpectraKit in your research, please cite:

@software{spectrakit,
  author = {Karthikeyan, Tubhyam},
  title = {SpectraKit: Python toolkit for spectral data processing},
  url = {https://github.com/ktubhyam/spectrakit},
  license = {MIT}
}

License

MIT