如何使用列表推导加速 function？

Question

Is there a way to speed up or change the function to increase efficiency?

This function is needed to transform about 200.000 arrays which again consists of 2700 arrays each. Currently the process takes an eternity.

def normalization(descriptors, max_value, min_value):
    if len(descriptors) == 3:
        output = []
        for descriptor in descriptors:
            output.append([[ ((feature-min_value)/(max_value-min_value)) for feature in des ] for des in descriptor])
        return np.asarray(output)
    else:
        return np.asarray([[ ((feature-min_value)/(max_value-min_value)) for feature in des ] for des in descriptors])

Answer 1

If descriptor is a numpy array, then

(descriptor-min_val)/(max_val-min_val)

should do the job, regardless of whether it's 2D or 3D.

I think that's all your list comprehension is doing - scaling each element of the array.