ProjSe-VaSP: Variable selection by projection operators

Requirements

The application of the ProjSe assumes the Python interpreter, version at least 3.7, and the numpy package, version at least 1.20.

To run the examples also requires the matplotlib. All these packages can be freely downloaded and installed from pypi.org.

Installation

The ProjSe package might be installed by the following procedures.

Directly from the github

pip3 install git+https://github.com/aalto-ics-kepaco/ProjSe.git#egg=ProjSe

Downloading from github

mkdir projective_selection

cd projective_selection

git clone https://github.com/aalto-ics-kepaco/ProjSe

After downloading the ProjSe package it can be installed by the following command:

pip3 install projective_selection/ProjSe

Before installing the ProjSe package the latest version of the Python packages pip and build need to be installed.

pip3 install --upgrade pip

pip3 install --upgrade build

The ProjSe can be imported as

import ProjSe

Running the projective selection algorithm

There is a demonstration in the "examples" directory. It requires the installation of the matplotlib.

Interface:

The basic class definition:

cprojector = cls_projector_kern(func_kernel = None, **kernel_params)

class cls_projector_kern:

def init(self, func_kernel = None, **kernel_params):

"""

Input:

func_kern

reference to a kernel function, it assumes two 2d array X1,X2 inputs, a 2d array of kernel matrix. The inner product are computed between the rows of X1 and X2. The number of rows in X1 and X2 can be different. See the default example: self.lin_kern in this class.
If func_kern == None then the linear kernel is used.

kernel_params

dictionary of parameters transferred to the function given in the func_kern.

"""

Running the variable selection:

cprojector.full_cycle(Y, X, nitem, ilocal = 1, iscale = 1)

def full_cycle(self,Y, X, nitem, ilocal = 1, iscale = 1):

"""

Task: to enumerate the x variables best correlating with Y but conditionally uncorrelating with the previous selection

Input:

       Y   2d array reference set, variables in the columns

       X   2d array of variables to be selected variables in the columns

       nitem   maximum number of x variables selected

       ilocal  =1 X,Y centralized =0 not

       iscale  =1 X,Y normalized row wise to have length 1 = 0 not

Output lorder list of x variables arranged by selection order

"""

The selection score can be read out of cprojector.xstat which contains the scores in the order of selection.