Normalize a 2D numpy array so that each "column" is on the same scale (Linear stretch from lowest value = 0 to highest value = 100)

normalize_numpy.py

import numpy as np

rawpoints = np.array(
   [[2500, 0.15, 12],
    [1200, 0.65, 20],
    [6200, 0.35, 19]]
) 

# Scale the rawpoints array so that each "column" is
# normalized to the same scale
# Linear stretch from lowest value = 0 to highest value = 100
high = 100.0
low = 0.0

mins = np.min(rawpoints, axis=0)
maxs = np.max(rawpoints, axis=0)
rng = maxs - mins

scaled_points = high - (((high - low) * (maxs - rawpoints)) / rng)

"""
scaled points ->
   [[  26.,     0.,     0., ],
    [   0.,   100.,   100., ],
    [ 100.,    40.,    87.5,]]
"""

Thank you for the function!

Thank You! Is there a faster method using skicit-learn or any numpy normalize function to directly normalize 2D array to given range? If no, is this the fastest way to normalize?

Thanks, works perfectly 👍 👌

Hi! i'm copying this code and its working perfectly at example, but in my real code, sometimes it outputting NaN, because my UB (or high) and LB (or low) goes 1.0 and 0.0. Does this code work if i set for all array (not by column like yours) ?

Unfortunately, the formula is wrong. Try putting [100, 50, 0] into a column. You should get it back exactly the same, i.e., [100, 50, 0]. You will find the formula you want at Feature Scaling.