from joblib import Parallel, delayed  
import multiprocessing
import time
import os
import numpy


# define some data
inputs = range(100)  

# function to execute in parallel
def myFun(i): 
	# Perform some computations 	  
	for x in range (0, 10000):
		y = numpy.mean( numpy.random.normal(0, 1, 10000))

	return i * i



# detect how many cores machine has
num_cores = multiprocessing.cpu_count()
print ("This machine has " +  str(num_cores) + " cores")

# get the number of cores that was requested by the submission script
num_cores_requested = int( os.getenv('NSLOTS') )
print ("The number of slots requested by the batch script is " + str(num_cores_requested) )


# use only as many cores as you have requested 
results = Parallel( n_jobs = num_cores_requested )( delayed( myFun )(i) for i in inputs)
print ("The result array is " +  ', '.join( str(x) for x in results))