How to parallelize loops

In image processing, we frequently apply the same algorithm on a large batch of images. In this paragraph, we propose to use joblib to parallelize loops. Here is an example of such repetitive tasks:

from skimage import data, color, util
from skimage.restoration import denoise_tv_chambolle
from skimage.feature import hog
 
def task(image):
    """
    Apply some functions and return an image.
    """
    image = denoise_tv_chambolle(image[0][0], weight=0.1, multichannel=True)
    fd, hog_image = hog(color.rgb2gray(image), orientations=8,
                        pixels_per_cell=(16, 16), cells_per_block=(1, 1),
                        visualise=True)
    return hog_image
 
 
# Prepare images
hubble = data.hubble_deep_field()
width = 10
pics = util.view_as_windows(hubble, (width, hubble.shape[1], hubble.shape[2]), step=width)

To call the function task on each element of the list pics, it is usual to write a for loop. To measure the execution time of this loop, you can use ipython and measure the execution time with %timeit.

def classic_loop():
    for image in pics:
        task(image)
 
 
%timeit classic_loop()

Another equivalent way to code this loop is to use a comprehension list which has the same efficiency.

def comprehension_loop():
    [task(image) for image in pics]
 
%timeit comprehension_loop()

joblib is a library providing an easy way to parallelize for loops once we have a comprehension list. The number of jobs can be specified.

from joblib import Parallel, delayed
def joblib_loop():
    Parallel(n_jobs=4)(delayed(task)(i) for i in pics)
 
%timeit joblib_loop()

Links:

http://scikit-image.org/docs/0.12.x/user_guide/tutorial_parallelization.html

doc_scikit_image

2025-01-10 15:47:30

Comments