0

I am pretty new to Python and totally new to multiprocessing, I found a few tutorials online to help me understand the multiprocessing package.

My code has a set of differential equations using RungeKutta4 method and I need to run tons of calculations with different starting conditions.

The code itself works but without multiprocessing it takes long to finish and I thought it might be ideal to use parallelization because the calculations are independent....

I am using Anaconda as IDE btw..

So I used

import multiprocessing as mp iterations = np.arange(start,end,step) pool = mp.Pool(mp.cpu_count()) # Step 1: Init multiprocessing.Pool() results = pool.map(rungeKutta4, [k for k in iterations]) # Step 2: apply pool map pool.close() # Step 3: close

When I run it in Anaconda I dont get an error, it starts calculating but it never stops.... Where did I go wrong?

Thanks in advance for help...

Best

EDIT: I added the whole code here...

# Python program to implement Runge Kutta method # Markus Schmid # 2020 Appalachian State University # jupyter nbconvert --to python FILENAME.ipynb # y" + 2*beta*y' + w0*sin(y) = A + B*cos(w*t) # Import used libraries import numpy as np import math import matplotlib.pyplot as plt import time from matplotlib import rc rc('font',**{'family':'sans-serif','sans-serif':['Helvetica']}) rc('text', usetex=True) import multiprocessing as mp print("Number of processors: ", mp.cpu_count()) # list for time (x axis) and result (y axis) result = [] w0 = 10 # undamped angular frequency of the oscillator beta = 1 B = 1 A = 0 B = 10 w = 4 theta_init = 0 theta_end = 20 theta_step = 0.1 # initial conditions t0 = 0 y0 = 0 z0 = 0 t_final = 20 # final time h = 0.01 # fixed step size n = (int)((t_final - t0)/h) # datapoints per RK4 iteration # define functions def funcf(t, y, z): return (z) def funcg(t, y, z): return (A+B*math.cos(w*t) - 2*beta*z - w0*math.sin(y)) # Finds value of y for a given x using step size h # and initial value y0 at x0. def rungeKutta4(y): # Count number of iterations using step size or # step height h t = t0 z = z0 n = (int)((t_final - t)/h) for i in range(1, n + 1): # Apply Runge Kutta to find next value of y k1 = h * funcf(t, y, z) l1 = h * funcg(t, y, z) k2 = h * funcf(t + 0.5 * h, y + 0.5 * k1, z + 0.5 * l1) l2 = h * funcg(t + 0.5 * h, y + 0.5 * k1, z + 0.5 * l1) k3 = h * funcf(t + 0.5 * h, y + 0.5 * k2, z + 0.5 * l2) l3 = h * funcg(t + 0.5 * h, y + 0.5 * k2, z + 0.5 * l2) k4 = h * funcf(t + h, y + k3, z + l3) l4 = h * funcg(t + h, y + k3, z + l3) # Update next value of y y = y + (1.0 / 6.0)*(k1 + 2 * k2 + 2 * k3 + k4) z = z + (1.0 / 6.0)*(l1 + 2 * l2 + 2 * l3 + l4) #result.append(y) t = t + h # Update next value of t return y iterations = np.arange(theta_init,theta_end+theta_step,theta_step) # number iterations for omega sweep start_time = time.time() #for k in iterations: # for serial calculation # rungeKutta4(k) pool = mp.Pool(mp.cpu_count()) # Step 1: Init multiprocessing.Pool() results = pool.map(rungeKutta4, [k for k in iterations]) # Step 2: apply pool map end_time = time.time() pool.close() # Step 3: close print ("The program took", end_time - start_time, "s to run") #table = np.array(result).reshape(len(iterations),n) # rearrange array, 1 row is const. theta0 timer = np.arange(t0,t_final,h) # time array
Improve this question
4
  • You should post the rangeKutta4 function as well as the values of start, end, step. Commented Apr 18, 2020 at 14:46
  • IDE's can interact with Python in unexpected ways. There are several mentions of multiprocessing misbehaving in IDE's on stackoverflow. Therefore it is probably best to test your script by directly calling Python from a shell (cmd.exe on ms-windows). Commented Apr 18, 2020 at 15:01
  • I tried running it without Anaconda .. RuntimeError: An attempt has been made to start a new process before the current process has finished its bootstrapping phase. This probably means that you are not using fork to start your child processes and you have forgotten to use the proper idiom in the main module: if name == 'main': freeze_support() ... The "freeze_support()" line can be omitted if the program is not going to be frozen to produce an executable. Commented Apr 18, 2020 at 16:08
  • If you want to speed up your code further, use a method with adaptive step size, like RKF or DoPri. The scipy solvers are, per hearsay, not very suitable for parallel computing, so you would have to implement your own routine. Does multiprocessing do JIT? Commented Apr 19, 2020 at 13:06

3 Answers 3

Reset to default
0

My initial comment:

IDE's can interact with Python in unexpected ways. There are several mentions of multiprocessing misbehaving in IDE's on stackoverflow. Therefore it is probably best to test your script by directly calling Python from a shell (cmd.exe on ms-windows).

You replied:

RuntimeError: An attempt has been made to start a new process before the current process has finished its bootstrapping phase.

For multiprocessing to work well on ms-windows, you have to be able to import the code in your script without side effects such as starting a new process.

This means that you have to put your everything except imports and function/class definitions inside a main block, like in this exemple inspired by the documentation:

from multiprocessing import Process def foo(): print('hello') if __name__ == '__main__': p = Process(target=foo) p.start()

Basically, ms-windows lacks the pretty nifty fork() system call that exists on UNIX-like systems. So the Python developers had to come up with a clever hack to make multiprocessing work on ms-windows. See e.g. this answer for the gory details.

Edit On ms-windows it seems that it is also necessary to put the worker function in its own file and import it. See e.g. here.

Improve this answer
Sign up to request clarification or add additional context in comments.

2 Comments

thank you very much that helped a lot!! I have it working on the terminal but not on Anaconda but it's a start.. I tried it with the needed resolution and it cut the time to 1/4 Next step is to use Anaconda and rework my loop (I need all calculations not only the last one and in the right order but 1 problem at a time.. ) Thank you so much
Do you mean Jupyter notebooks? If so, you should be aware that this is also known to not work well with multiprocessing on ms-windows without some extra steps. See (1), (2)
0

I found the solution to make it work in Anaconda.

apart from the 

if __name__ == '__main__':

the function that is called (the worker) has to be in its own .py file that has to be imported at the beginning..

I put the RK4 in a RK4.py file that i imported at the beginning and not it works but that opens new problems in my case with passing the values back and forth...

Improve this answer

Comments

0

Your code seems fine, and it works for me (Ipython with python3.6.9), I've attached the results and configuration below.

Did you try to run it in a different python virtual environment (outside of Anaconda)? As it has been stated before (in Ronald's comment) other IDE's have bugs with python's multiprocessing, try run it using the command line/Ipython. I have previously encountered one even in vscode

Configuration:

python configuration

Results:

script results

Improve this answer

1 Comment

thank you very much, I got it to work in win terminal... actually its windows thats messing everything up (once again...) unfortunately our lab PCs all run windows - I would prefer linux

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.