Distances

[kremnik]

What has been done

With this PR, some little optimizations of distance calculations have been done:

1. Using np.linalg.norm for norm and Euclidean distance calculations
2. Adding zero denominator checks in the l2_normalize function

How to test

make lint && make test

[serengil]

old find_euclidean_distance function returns 0.9086272014052611 for dataset/img1.jpg and dataset/img2.jpg whereas new function returns 0.908627201405261

thresholds are tuned for the old function. i am not confident to merge this to cause any trouble.

[kremnik]

@serengil The values are extremely close, with the difference appearing in the 15th decimal digit. Since the thresholds precision is set to 2 decimal digits, so small difference should not affect the overall results.

[serengil]

yeah you are totally right, it is my bad.

[serengil]

TLDR: np.linalg.norm(a - b) is equivalent to np.sqrt(np.sum(np.multiply(a - b, a - b))). Both methods compute the Euclidean distance between two vectors a and b.

[serengil]

@kremnik LGTM and merged. Thank you for your contribution.