Производительность multiprocessing.Pool

def proc(n):
    pool = multiprocessing.Pool(n)
    
    #bar = progressbar.ProgressBar(max=len(train))
    distances = np.zeros((len(train), len(train)))
    #print("len=",len(train))
    for i, img in enumerate(train): #enumerate(bar(train)):
        #print("i=", i)
        all_imgs = [(img, f) for f in train]
        dists = pool.map(compare, all_imgs)
        distances[i, :] = dists
 
import time
for i in range(1,20):
    start = time.clock()
    print("Число потоков=",i)
    #print("")
    proc(i)
 
    print((time.clock() - start), "время")

import pandas as pd
import numpy as np
import glob
from sklearn import cluster
from scipy.misc import imread
import cv2
import skimage.measure as sm
# import progressbar
import multiprocessing
import random
import matplotlib.pyplot as plt
import seaborn as sns
#%matplotlib inline
new_style = {'grid': False}
plt.rc('axes', **new_style)
 
# Function to show 4 images
def show_four(imgs, title):
    #select_imgs = [np.random.choice(imgs) for _ in range(4)]
    select_imgs = [imgs[np.random.choice(len(imgs))] for _ in range(4)]
    _, ax = plt.subplots(1, 4, sharex='col', sharey='row', figsize=(20, 3))
    plt.suptitle(title, size=20)
    for i, img in enumerate(select_imgs):
        ax[i].imshow(img)
 
# Function to show 8 images
def show_eight(imgs, title):
    select_imgs = [imgs[np.random.choice(len(imgs))] for _ in range(8)]
    _, ax = plt.subplots(2, 4, sharex='col', sharey='row', figsize=(20, 6))
    plt.suptitle(title, size=20)
    for i, img in enumerate(select_imgs):
        ax[i // 4, i % 4].imshow(img)
 
 
#In2
select = 50 # Only load 500 images for speed
# Data loading
train_files = sorted(glob.glob('/home/user-name/Data/Save_Fish/train/*/*.jpg'), key=lambda x: random.random())[:select]
train = np.array([imread(img) for img in train_files])
print('Length of train {}'.format(len(train)))
 
 
#In3
 
print('Sizes in train:')
shapes = np.array([str(img.shape) for img in train])
pd.Series(shapes).value_counts()
 
 
#In4
 
#for uniq in pd.Series(shapes).unique():
#    show_four(train[shapes == uniq], 'Images with shape: {}'.format(uniq))
#    plt.show()
 
 
 
#In5
 
# Function for computing distance between images
def compare(args):
    img, img2 = args
    img = (img - img.mean()) / img.std()
    img2 = (img2 - img2.mean()) / img2.std()
    return np.mean(np.abs(img - img2))
 
# Resize the images to speed it up.
train = [cv2.resize(img, (224, 224), cv2.INTER_LINEAR) for img in train]
 
# Create the distance matrix in a multithreaded fashion
 
#from multiprocessing import set_start_method, Pool
#set_start_method('forkserver')
 
# You can then start your Pool without each process
# cloning your entire memory
#pool = Pool()
#func = partial(parallelUpdateJSON, paramMatch, predictionmatrix)
#pool.map(func, data)
 
def proc(n):
    pool = multiprocessing.Pool(n)
 
    #bar = progressbar.ProgressBar(max=len(train))
    distances = np.zeros((len(train), len(train)))
    #print("len=",len(train))
    for i, img in enumerate(train): #enumerate(bar(train)):
        #print("i=", i)
        all_imgs = [(img, f) for f in train]
        dists = pool.map(compare, all_imgs)
        distances[i, :] = dists
 
import time
for i in range(1,20):
    start = time.clock()
    print("Число потоков=",i)
    #print("")
    proc(i)
 
    print((time.clock() - start), "время")