Ridge and Logistic regression with a fast implementation of statistical significance test.
Prerequisite:
1, python >= 3.6 developer version. We suggest install anaconda (https://www.anaconda.com) to include all required packages.
2, numpy >= 1.19;
3, pandas >= 1.1.4;
4, gcc >= 4.2;
5, gsl-2.6: https://ftp.gnu.org/gnu/gsl (please don't use other versions, such as 2.7)
Install:
python3 setup.py install
Test:
python3 -m unittest tests.regression
Usage:
Call the regression function in python code as follows:
beta, se, zscore, pvalue = data_significance.ridge(X, Y, alpha, alternative, nrand, verbose).
Input:
X: explanatory matrix, numpy matrix in C-contiguous order (last-index varies the fastest).
Y: response variable, numpy matrix in C-contiguous order (last-index varies the fastest).
The row dimension of X and Y should be the same. Y input allows multiple columns, with each column as one input response variable. The function will return the same number of output variables in output matrices (beta, se, zscore, pvalue).
alpha: penalty factor in ridge regression (>= 0). If alpha is 0, we will use regular ordinary least square.
alternative: one-tailed or two-tailed statistical test, with three options: 1, two-sided; 2, greater; 3, less.
nrand: number of randomizations (>=0). If nrand = 0, we will use student t-test for the statistical test. Otherwise, we will use permutation test.
verbose: 1 or 0. Report intermediate results.
Output:
beta: regression coefficient.
se: standard error.
zscore: beta/se.
pvalue: statistical significance from permutation test (nrand>0) or student t-test (nrand=0).