如何规范numpy数组中的子数组

Question

I have a numpy array data of shape: [128, 64, 64, 64], and I wonder what's the best way to normalized each of the 128 slices into range [0.0, 1.0]. I understand i could use np.max(data[0,...]), np.max(data[1,...])... np.max(data[127,...]) to compute max values in each slice, but wonder if i could do this more efficiently.

Essentially something like this:

data_min = np.min(data[:,...])
data_max = np.max(data[:,...])
norm_data = (data[:,...] - data_min)/(data_max - data_min)

The result should still have shape [128, 64, 64, 64] But i haven't figured out which particular min/max functions and options to use to obtain the results.

Please advise. Thanks!

Answer 1

Get the min and max values, while keeping dimensions to help us with broadcasting later on when we use those to normalize input data using the normalization formula, like so -

mins = data.min(axis=(1,2,3), keepdims=True)
maxs = data.max(axis=(1,2,3), keepdims=True)
norm_data = (data-mins)/(maxs-mins)

Answer 2

You can use np.vectorize to apply a function over all elements of an array:

def norm(element):
    return (element - data_min) / (data_max - data_min)

ndnorm = np.vectorize(norm)

data_min = data.min()
data_max = data.max()

norm_data = ndnorm(data)