0

I want to compute this for loop:

import numpy as np import cmath import multiprocessing as mp from functools import partial def main(): nodes = 53 f_nom = 50 U = [1.000005327453146, 1.0000057162412561, 1.0000070448192537, 1.0000069227537496, 1.0000069227537496, 1.0000041605235332, 1.0000035222393044, 1.000004263922246, 1.0000047513736152, 1.0000047550593771, 1.0000049581713173, 1.000004263922246, 1.0000039401443763, 1.0000026446387185, 1.0000014323088564, 1.0000026446387185, 1.0000036287590721, 1.0000026446387185, 1.0000039401443768, 1.0000014323088589, 1.0000039401443768, 1.0000014323088553, 1.0000014323088553, 1.0000030231390993, 1.0000014323088615, 1.0000014323088615, 1.0000014323088606, 1.0000014323088604, 1.0000014323088597, 1.0000014323088606, 1.0000014323088604, 1.0000014323088597, 1.0000014323088593, 1.000001432308859, 1.000001432308858, 1.000001590333777, 1.0000020451347542, 1.000001432308861, 1.0000014323088602, 1.000001432308863, 1.00000143230886, 1.000001432308859, 1.000001432308863, 1.000001432308863, 1.000001432308859, 1.000001432308863, 1.00000143230886, 1.0000014323088615, 1.000001432308859, 1.0000014323088615, 1.0000057853629756, 1.0000059478472048, 1.0000000056539258, 1.0] theta = [-7.355656043185493e-06, -3.464594992534604e-06, -1.0583594144863342e-05, -9.626585410799617e-06, -9.626585410799618e-06, -5.6945086783273145e-06, -2.674752704682853e-06, -5.841699103904964e-06, -6.535596440474819e-06, -3.118569281977461e-06, -6.829976913326792e-06, -5.841699103904964e-06, -2.825199109428268e-06, -2.3588148383740883e-06, -1.1617097845584577e-06, -2.3588148383740883e-06, -4.937528901600708e-06, -2.3588148383740883e-06, -2.8251991094452968e-06, -1.1617097837371578e-06, -2.8251991094452976e-06, -1.1617097849706262e-06, -1.1617097849706262e-06, -4.075412032614385e-06, -1.1617097820641042e-06, -1.1617097820641045e-06, -1.16170978206629e-06, -1.161709782064495e-06, -1.1617097822154165e-06, -1.16170978206629e-06, -1.161709782064495e-06, -1.1617097822154165e-06, -1.1617097834417375e-06, -1.1617097836565141e-06, -1.1617097839483619e-06, -2.0357655204310186e-06, -1.766840164692082e-06, -1.1617097820892918e-06, -1.1617097821105786e-06, -1.1617097821630116e-06, -1.1617097822743808e-06, -1.1617097836888828e-06, -1.161709782163012e-06, -1.1617097821630119e-06, -1.1617097836179758e-06, -1.1617097821630119e-06, -1.1617097822743808e-06, -1.1617097820641042e-06, -1.161709783617976e-06, -1.1617097820641047e-06, -8.007497618793153e-06, -8.238796094878206e-06, -1.972012101284986e-08, 0.0] U_start = [396.98922, 397.590444, 396.6735, 396.663384, 396.625768, 397.134352, 397.85536, 397.132184, 397.123008, 397.680784, 397.074752, 397.131104, 397.78544, 397.975768, 397.642204, 398.124984, 397.186264, 397.87854, 397.766736, 397.505908, 397.706156, 397.640524, 397.643368, 397.294984, 397.816152, 397.855528, 397.792792, 397.785664, 397.736036, 397.802636, 397.783008, 397.746864, 397.641188, 397.56948, 397.635652, 397.676168, 397.972024, 397.835444, 397.76222, 397.440376, 397.740988, 397.405996, 397.40994, 397.429992, 397.336748, 397.466372, 397.744532, 397.846052, 397.311572, 397.728728, 396.841804, 396.767284, 397.679944, 400.] t = np.linspace(0, 0.1, int(0.1 * 20000), dtype=np.float64) wave_line = np.zeros((3, nodes, len(t))) wave_line_mp = np.zeros((3, nodes, len(t))) looprange = 5 for i in range(looprange): harm_order = i + 2 for n in range(nodes): U_line = cmath.polar(cmath.rect(1.0, 0.0) - cmath.rect(U[n], theta[n])) wave_line[0, n] += np.sqrt(2) / np.sqrt(3) * U_line[0] * U_start[n] * np.sin((t * 2 * np.pi * f_nom + U_line[1]) * harm_order) wave_line[1, n] += np.sqrt(2) / np.sqrt(3) * U_line[0] * U_start[n] * np.sin((t * 2 * np.pi * f_nom + U_line[1] - 2 / 3 * np.pi) * harm_order) wave_line[2, n] += np.sqrt(2) / np.sqrt(3) * U_line[0] * U_start[n] * np.sin((t * 2 * np.pi * f_nom + U_line[1] + 2 / 3 * np.pi) * harm_order) print(i, wave_line[0][0][0]) wave_line[:] = 0 if __name__ == '__main__': main()

Output 1:

0 -0.0027971102683270466 1 0.000897146546025977 2 0.0017446246602729246 3 -0.0029438493445009956 4 0.0017089460869939772

in a multiprocessing.Pool.map function:

import numpy as np import cmath import multiprocessing as mp from functools import partial def main(): nodes = 53 f_nom = 50 U = [1.000005327453146, 1.0000057162412561, 1.0000070448192537, 1.0000069227537496, 1.0000069227537496, 1.0000041605235332, 1.0000035222393044, 1.000004263922246, 1.0000047513736152, 1.0000047550593771, 1.0000049581713173, 1.000004263922246, 1.0000039401443763, 1.0000026446387185, 1.0000014323088564, 1.0000026446387185, 1.0000036287590721, 1.0000026446387185, 1.0000039401443768, 1.0000014323088589, 1.0000039401443768, 1.0000014323088553, 1.0000014323088553, 1.0000030231390993, 1.0000014323088615, 1.0000014323088615, 1.0000014323088606, 1.0000014323088604, 1.0000014323088597, 1.0000014323088606, 1.0000014323088604, 1.0000014323088597, 1.0000014323088593, 1.000001432308859, 1.000001432308858, 1.000001590333777, 1.0000020451347542, 1.000001432308861, 1.0000014323088602, 1.000001432308863, 1.00000143230886, 1.000001432308859, 1.000001432308863, 1.000001432308863, 1.000001432308859, 1.000001432308863, 1.00000143230886, 1.0000014323088615, 1.000001432308859, 1.0000014323088615, 1.0000057853629756, 1.0000059478472048, 1.0000000056539258, 1.0] theta = [-7.355656043185493e-06, -3.464594992534604e-06, -1.0583594144863342e-05, -9.626585410799617e-06, -9.626585410799618e-06, -5.6945086783273145e-06, -2.674752704682853e-06, -5.841699103904964e-06, -6.535596440474819e-06, -3.118569281977461e-06, -6.829976913326792e-06, -5.841699103904964e-06, -2.825199109428268e-06, -2.3588148383740883e-06, -1.1617097845584577e-06, -2.3588148383740883e-06, -4.937528901600708e-06, -2.3588148383740883e-06, -2.8251991094452968e-06, -1.1617097837371578e-06, -2.8251991094452976e-06, -1.1617097849706262e-06, -1.1617097849706262e-06, -4.075412032614385e-06, -1.1617097820641042e-06, -1.1617097820641045e-06, -1.16170978206629e-06, -1.161709782064495e-06, -1.1617097822154165e-06, -1.16170978206629e-06, -1.161709782064495e-06, -1.1617097822154165e-06, -1.1617097834417375e-06, -1.1617097836565141e-06, -1.1617097839483619e-06, -2.0357655204310186e-06, -1.766840164692082e-06, -1.1617097820892918e-06, -1.1617097821105786e-06, -1.1617097821630116e-06, -1.1617097822743808e-06, -1.1617097836888828e-06, -1.161709782163012e-06, -1.1617097821630119e-06, -1.1617097836179758e-06, -1.1617097821630119e-06, -1.1617097822743808e-06, -1.1617097820641042e-06, -1.161709783617976e-06, -1.1617097820641047e-06, -8.007497618793153e-06, -8.238796094878206e-06, -1.972012101284986e-08, 0.0] U_start = [396.98922, 397.590444, 396.6735, 396.663384, 396.625768, 397.134352, 397.85536, 397.132184, 397.123008, 397.680784, 397.074752, 397.131104, 397.78544, 397.975768, 397.642204, 398.124984, 397.186264, 397.87854, 397.766736, 397.505908, 397.706156, 397.640524, 397.643368, 397.294984, 397.816152, 397.855528, 397.792792, 397.785664, 397.736036, 397.802636, 397.783008, 397.746864, 397.641188, 397.56948, 397.635652, 397.676168, 397.972024, 397.835444, 397.76222, 397.440376, 397.740988, 397.405996, 397.40994, 397.429992, 397.336748, 397.466372, 397.744532, 397.846052, 397.311572, 397.728728, 396.841804, 396.767284, 397.679944, 400.] t = np.linspace(0, 0.1, int(0.1 * 20000), dtype=np.float64) wave_line = np.zeros((3, nodes, len(t))) wave_line_mp = np.zeros((3, nodes, len(t))) looprange = 5 func_part = partial(worker_function, nodes=nodes, f_nom=f_nom, U=U, theta=theta, U_start=U_start, t=t, wave_line_mp=wave_line_mp) wave_line_results = mp.Pool(1).map(func_part, range(looprange)) print('') for idx, wave_line_mp in wave_line_results: print(idx, wave_line_mp[0][0][0]) def worker_function(i=None, nodes=None, f_nom=None, U=None, theta=None, U_start=None, t=None, wave_line_mp=None): harm_order = i + 2 wave_line_mp[:] = 0 for n in range(nodes): U_line = cmath.polar(cmath.rect(1.0, 0.0) - cmath.rect(U[n], theta[n])) wave_line_mp[0, n] += np.sqrt(2) / np.sqrt(3) * U_line[0] * U_start[n] * np.sin((t * 2 * np.pi * f_nom + U_line[1]) * harm_order) wave_line_mp[1, n] += np.sqrt(2) / np.sqrt(3) * U_line[0] * U_start[n] * np.sin((t * 2 * np.pi * f_nom + U_line[1] - 2 / 3 * np.pi) * harm_order) wave_line_mp[2, n] += np.sqrt(2) / np.sqrt(3) * U_line[0] * U_start[n] * np.sin((t * 2 * np.pi * f_nom + U_line[1] + 2 / 3 * np.pi) * harm_order) print(i, wave_line_mp[0][0][0]) return i, wave_line_mp if __name__ == '__main__': main()

Output 2:

0 -0.0027971102683270466 1 0.000897146546025977 2 0.0017446246602729246 3 -0.0029438493445009956 4 0.0017089460869939772 0 0.000897146546025977 1 0.000897146546025977 2 -0.0029438493445009956 3 -0.0029438493445009956 4 0.0017089460869939772

As you can see output (2.1) inside the worker_function is correct (it is similar to the above output (1) of the for loop) while the returned wave_line_results and thus output (2.2) is not. There seems to be a value overlap, maybe a scheduling problem with the Pool function?

Improve this question
4
  • 2
    You seem to be passing in a mutable argument and then changing it. Maybe you want to make a copy? Commented Oct 8, 2020 at 6:35
  • 2
    Works for me if I put wave_line_mp = wave_line_mp.copy() at the start of worker_function - although of course it then overwrites the values with zeros, so maybe you could just create the array of zeros inside the worker function, based on the shape, and not pass in the array at all. Commented Oct 8, 2020 at 6:37
  • Welp, looks like you have resolved my problem in a few minutes. Do you want to post an answer :-) ? Commented Oct 8, 2020 at 6:41
  • 1
    Okay, have done. Commented Oct 8, 2020 at 6:50

1 Answer 1

Reset to default
1

It seemt that you are passing the (mutable) array wave_line_mp into the function and then changing it, and that you should be operating on separate copies of it in order to do so. A simple one-line fix is to insert wave_line_mp = wave_line_mp.copy() at the start of worker_function, but a better solution is probably to remove it from the argument list of the function entirely (and of course also from where it is called), and move the line:

 wave_line_mp = np.zeros((3, nodes, len(t)))

from inside main to the start of worker_function. (This works because the variables nodes and t are passed through to variables of the same names inside worker_function.) You would also remove the line wave_line_mp[:] = 0, which is then redundant.

Doing this, you get the desired output:

0 -0.00279711026833 1 0.000897146546026 2 0.00174462466027 3 -0.0029438493445 4 0.00170894608699 0 -0.00279711026833 1 0.000897146546026 2 0.00174462466027 3 -0.0029438493445 4 0.00170894608699
Improve this answer
Sign up to request clarification or add additional context in comments.

Comments

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.