AUSWAHL

AUSWAHL (AUtomatic Selection of WAvelengtH Library) is a python module that provides a collection of Feature Selection Methods for Near-Infrared Spectroscopy.

Scope

This library provides a collection of supervised feature selection methods for near-infrared spectra. (Note, that features are also called variables of simply wavelengths in the field of chemometrics.) It does not provide any preprocessing methods or calibration models. All features selection methods have been implemented in compliance with the sklearn library. Thus, the methods can be used in a sklearn pipeline before an estimator.

Methods

The libraray provides methods for selecting wavelengths without spatial constraints and methods for selecting continuous bands of wavelengths. The following methods are implemented:

• Wavelength Point Selection: VIP, MC-UVE, Random Frog, CARS, SPA
• Wavelength Interval Selection: iPLS, FiPLS, BiPLS, iRF

Installation

AUSWAHL is currently not available through PyPi. We are working on releasing it in the near future. Right now, you can install it from source:

git clone https://github.com/LSX-UniWue/auswahl.git
cd auswahl
pip install .

Usage

The provided feature selection methods implement the SelectorMixin base class from sklearn and can by used in the same way as the sklearn feature selection methods. Below you can find a simple example for the VIP selection method:

import numpy as np
from auswahl import VIP

rs = np.random.RandomState(1337) # Sample seed for reproducibility
x = rs.randn(100, 10)            # 100 samples and 10 features
y = 5 * x[:,0] - 2 * x[:,5]      # y depends only on two features

selector = VIP(n_features_to_select=2)
selector.fit(x,y)

selector.get_support()
>>> array([True, False, False, False, False, True, False, False, False, False])

If you want to use another feature selection method, you can simply replace VIP(...) with any other method.

The VIP method also allows to select the number of features after the model has been fitted. An example is given below:

vip = VIP(n_features_to_select=x.shape[1]-1)
vip.fit(x,y)
vip.get_support_for_threshold(threshold=1)
>>> array([True, False, False, False, False, True, False, False, False, False])