scispectrum 0.3.0

1D detector spectrum analysis — energy calibration, background estimation, peak detection and fitting

scispectrum

A Python package for 1D spectrum analysis, designed for physicists and scientists working with detector data such as gamma-ray, X-ray, or particle spectra.

scispectrum provides a clean, extensible framework for the full spectrum analysis pipeline — from raw detector output to calibrated, fitted, background-subtracted results — with proper uncertainty propagation throughout.

---

Features

• Spectrum object with axis calibration, resolution calibration, and error propagation using the uncertainties package
• Domain slicing by physical axis values (e.g. energy in keV) or channels
• List-mode parser for time-channel detector data, with chunked reading for large files
• Domain fitting and analysis — multiple fitting and analysis methods for a domain including sum of Gaussians fitting, find peaks, centers and fwhm
• Background estimation — multiple global background extimation methods including Asymmetric Least Squares (ALS)
• Extensible base classes — build your own fitting, background, or analysis procedures with a consistent interface
• xarray throughout — labeled, sliceable data with named coordinates
• Uncertainty propagation via the uncertainties library

---

Installation

scispectrum is not yet available on PyPI. Install directly from GitHub:

git clone https://github.com/achiyaAmrusi/pySpectrum
cd pySpectrum
pip install -e .

This installs the package in editable mode, so any changes you make to the source are reflected immediately.

---

Quick Start

Load a spectrum from a DataFrame

import pandas as pd
from scispectrum.core import Spectrum

df = pd.read_csv("my_spectrum.csv")
spectrum = Spectrum.from_dataframe(df, channel_col="channel", counts_col="counts")

Apply an energy calibration

from scispectrum.calibration import AxisCalibration

# Linear calibration: energy = 0.5 * channel + 1.2
calib = AxisCalibration(lambda ch: 0.5 * ch + 1.2, name="energy_keV")
spectrum.set_axis_calibration(calib)

Slice a domain by axis values

# Select the region between 1460 and 1480 keV
domain = spectrum.domain(1460, 1480)

Fit peaks

from scispectrum.domain_fitting import SumOfGaussians

result = SumOfGaussians.fit(domain)

print(result["center"].values)   # peak centers in keV
print(result["fwhm"].values)     # peak widths
print(result["amplitude"].values) # peak amplitudes

Estimate and subtract background

from scispectrum.background import ALSBackground

bg_estimator = ALSBackground(lam=1e5, p=0.001, max_iter=50)
als_bg = bg_estimator.estimate(spectrum.axis, spectrum.counts)
domain_subtracted = domains.subtract_background(als_bg[domain.indices])

Parse raw list-mode data

from scispectrum.parsers import TimeChannelParser

# From a large file — processed in chunks to save memory
spectrum = TimeChannelParser.from_file(
    "detector_run.csv",
    axis_calib=calib,
    num_of_channels=2**14
)

# From an in-memory DataFrame
spectrum = TimeChannelParser.from_dataframe(df, axis_calib=calib)

---

Uncertainty Propagation

scispectrum propagates measurement uncertainties through arithmetic operations using the uncertainties library. Poisson errors are assigned automatically when parsing list-mode data.

# Arithmetic preserves errors
subtraction = spectrum_a - spectrum_b
subtraction.counts_err  # propagated uncertainties

---

Extending scispectrum

scispectrum is designed to be extended. Each analysis category has an abstract base class that defines the interface:

Custom fitting method

from scispectrum.domain_fitting.abstract_fitting_class import PeakFit

class MyFitter(PeakFit):
    @classmethod
    def fit(cls, domain, **kwargs):
        # your fitting logic here
        ...

Custom background estimator

from scispectrum.background.base import BackgroundEstimator

class MyBackground(BackgroundEstimator):
    def estimate(self, x, y):
        # your background logic here
        ...

All custom classes integrate seamlessly with Domain, Spectrum, and the rest of the pipeline.

---

Examples

Full worked examples are available in the examples directory:

Core

• Loading a spectrum — reading and constructing a Spectrum from data
• Calibration — applying axis and resolution calibrations
• Domain slicing — creating and working with domains

Background Estimation

• Background subtraction — estimating and subtracting background from a domain

Domain Analysis and Fitting

• Domain fitting — fitting peaks in a single domain
• Full spectrum fitting — fitting peaks across an entire spectrum

SNR Identification

• Peak domain identification — identifying signal regions automatically
• Complex spectrum domains — handling overlapping and complex peak structures

Library Sample data files for running the examples are provided in the Library directory.

---

Requirements

• numpy>=2.0.0,<3.0
• pandas>=2.3,<4.0
• scipy>=1.14.0
• xarray>=2024.6.0
• uncertainties>=3.1

---

License

MIT License. See LICENSE for details.

---

Author

Achiya Yosef Amrusi — GitHub

Contributions and feedback are welcome.