This repository contains simple functions to interactively analyze results from the DLP NIRScan Nano EVM by Texas Instruments. It was initially created for a student course to hide any boilerplate code and significantly reduce the programming experience requirements.
Installation via a package manager is not yet supported. Instead, copy the library folder NIRScanNano and add the directory to your PYTHONPATH environment variable.
The depencies can be installed via pip or conda (Python >= 3.7):
pip install -r requirements.txt
conda create -n nirscan --file requirements.txt
A full example of a complete student course (20 groups) can be found in the examples folder.
Reading a single spectrum returns a NIRSpetrum object containing the data and all header information.
from NIRScanNano.spectrum import read_spectrum
spectrum = read_spectrum("caffeine.csv")It is also supported to read a batch of spectra using a data file. This is especially useful for performing a principal component analysis of all measured spectra in the course.
from NIRScanNano.course import DataReader
data = DataReader("datafile.csv")One can specify the column names and delimiters. The default is: "Group";"Name";"File" with Group as the key value to access the data later, Name as the compound name, and File is the path to the file relative to the data file directory.
# returns a list of spectra for the key value (e.g. group id)
spectra = data.spectra_by_group(key_value) The DataReader also provides a method to select a random sample.
spectrum = data.random_sample()The compound name will be saved for each spectrum as an entry in the header dictionary.
spectrum.header["Name"]The list can be filtered using list comprehensions to obtain all spectra for the desired substance.
caffeine_spectra = [s for s in spectra if s.header["Name"] == "Caffeine"]We currently support the most common methods for preprocessing NIR spectra.
from NIRScanNano.analysis import snv, savgol, mscnorm_spectrum = snv(spectrum, norm=True)snv_spectrum = snv(spectrum)If no reference spectrum is available, it is also possible to average multiple spectra of a compound. Please be aware that no checks are performed.
avg_spectrum = average_spectra(spectra)
msc_spectrum = msc(spectrum, avg_spectrum)The savgol function is a simple wrapper around the SciPy function savgol_filter to directly handle a NIRSpectrum object.
# spectrum, window, polynom order, derivative
savgol_spectrum = savgol(spectrum, 11, 2, 0)The library provides a plot_spectrum function to visualize a single spectrum or multiple spectra (by passing a list) using Matplotlib.
import matplotlib.pyplot as plt
from NIRScanNano.visualization import plot_spectrum
fig, ax = plt.subplots(1)
plot_spectrum(example_spectra, ax=ax)We have implemented basic functionalities to perform Principal Component Analysis, export everything into a pandas DataFrame and visualize it in a seaborn pairplot.
from NIRScanNano.course import pca_to_pandas
from NIRScanNano.visualization import pca_pairplot
from NIRScanNano.analysis import PCAnalysis
pca = PCAnalysis(spectra)
pca.run()
pca_df = pca_to_pandas(pca, label="Name")
pca_pairplot(pca_df)The distances to all hypersphere centers of individual substances in transformed space are calculated to identify an unknown compound. The radius is defined as the maximum distance of each corresponding data point plus a quarter of the standard deviation.
from NIRScanNano.course import pca_centroids, nearest_centroids, eval_distances
# column names
columns = [col for col in pca_df.columns if "PC" in col]
# centroids of all substances and maximum distances
centroids, max_dist = pca_centroids(pca_df, columns)
# distances of the unknown substance to all centroids
distances = nearest_centroids(test_spectrum, pca, centroids, 10)
# multiple compounds are returned if there is an overlap
# of the hyperspheres and none if the spectrum is not
# within any hypersphere
possible_compounds = eval_distances(distances, max_dist)We are currently preparing the manuscript.