calculate_correlation_matrix about ml-from-scratch HOT 1 CLOSED

Yes, the diagonal values in the correlation matrix should be 1. I have compared my methods to the corresponding numpy implementations, and the scaling factor in calculating the correlation matrix should have been 1/N instead of 1/(N-1) (used in calculating the covariance matrix).

The latest commit fixed this.

> import numpy as np
> from data_operation import *
> x = np.array(np.random.random((10, 4)))
> calculate_correlation_matrix(x)
array([[ 1. , 0.18390596, 0.34851226, 0.83499889],
[ 0.18390596, 1. , 0.27027579, 0.44432891],
[ 0.34851226, 0.27027579, 1. , 0.52136316],
[ 0.83499889, 0.44432891, 0.52136316, 1. ]])
> np.corrcoef(x, rowvar=0)
array([[ 1. , 0.18390596, 0.34851226, 0.83499889],
[ 0.18390596, 1. , 0.27027579, 0.44432891],
[ 0.34851226, 0.27027579, 1. , 0.52136316],
[ 0.83499889, 0.44432891, 0.52136316, 1. ]])
> calculate_covariance_matrix(x)
array([[ 0.11678785, 0.019298 , 0.0229638 , 0.07239497],
[ 0.019298 , 0.09428342, 0.01600117, 0.03461353],
[ 0.0229638 , 0.01600117, 0.03717523, 0.02550297],
[ 0.07239497, 0.03461353, 0.02550297, 0.06436463]])
> np.cov(x, rowvar=0)
array([[ 0.11678785, 0.019298 , 0.0229638 , 0.07239497],
[ 0.019298 , 0.09428342, 0.01600117, 0.03461353],
[ 0.0229638 , 0.01600117, 0.03717523, 0.02550297],
[ 0.07239497, 0.03461353, 0.02550297, 0.06436463]])

from ml-from-scratch.