I have a 3d array of dimension (1000, 300, 3), my_3d_array.
I want to modify each sample if this array then replace the old version with the modified one. In another word, I want to take the 1st example (300, 3) (lets called it old sample) apply some modification to it by calling a function called my_function, then the obtained new one will replace the old sample in my_3d_array, and so on. How can we do it please?
I'm not exactly sure this is what you want, but you can use the normal indexing notation:
# just an example array
my_3d_array = np.random.random((1000,300,3))
# extract the old sample
old_sample = my_3d_array[0,:,:]
# transform the sample
new_sample = my_function(old_sample)
# write the sample back to the original array
my_3d_array[0,:,:] = new_sample
If you want to do the same transformation for all samples it might be beneficial to vectorize my_function to process all samples at once instead of using a python loop.
for loop over the first dimension of the array not an option?apply_along_axis and vectorize numpy.org/doc/stable/reference/routines.functional.html. These could be used, but i doubt that they will give a significant performance boost because they also have to call your python function as many times as there are samples in your array. If my_function only uses numpy functions you can try to vectorize the function directly, this would probably give the best performance.