qianyizhang

def scaled_l2(X, C, S):
    """
    scaled_l2 distance
    Args:
        X (b*n*d):  original feature input
        C (k*d):    code words, with k codes, each with d dimension
        S (k):      scale cofficient
    Return:
        D (b*n*k):  relative distance to each code
    Note:

qianyizhang

import numpy as np
def one_hot(nparray, depth = 0, on_value = 1, off_value = 0):
    if depth == 0:
        depth = np.max(nparray) + 1
    assert np.max(nparray) < depth, "the max index of nparray: {} is larger than depth: {}".format(np.max(nparray), depth)
    shape = nparray.shape
    out = np.ones((*shape, depth)) * off_value
    indices = []
    for i in range(nparray.ndim):
        tiles = [1] * nparray.ndim