import numpy as np

channels, width, height=200, 1700, 350
# and out goal to get
# channels = 10, width = 1700, height=350
NUMBER_OF_DIM_TO_REDUCE = 10

sample_matrix = np.random.randint(0, 100, size=(width, height, channels))

# declare only once from scikit learn
pca = None

def get_reduced_dimensions(input_matrix):
    # do dimension reduction on
    # in theory you want to return of shape
    # height, NUMBER_OF_DIM_TO_REDUCE
    # but i am just returning random one now
    return np.random.randint(0, 100, size=(height, NUMBER_OF_DIM_TO_REDUCE))

def get_channel_as_pca_input(matrix, column):
    return matrix[column,:,:]

# we need to reduce from 200 channels to day 10, so we need to get all channels for a given channel
all_channels_for_col_0 = get_channel_as_pca_input(sample_matrix, 0)
# all_channels_for_col_0.shape = 350, 200
# send all_channels_for_col_0 through get_reduced_dimensions function

new_matrix_with_reduced_dimensions = np.random.randint(0, 100, size=(width, height, NUMBER_OF_DIM_TO_REDUCE))

for i in range(width):
    all_channels = get_channel_as_pca_input(sample_matrix, i)
    reduced_dimension = get_reduced_dimensions(all_channels)
    new_matrix_with_reduced_dimensions[i,:,:] = reduced_dimension