Script for memory profiling astropy.modeling
from astropy.modeling.models import Mapping, Tabular1D
from astropy.modeling.functional_models import Shift, Scale
from astropy.modeling.polynomial import Polynomial2D
import gc
from memory_profiler import profile
import numpy as np
@profile
def test_models(x):
tab = Tabular1D(lookup_table = np.arange(62500000), bounds_error=False)
offx = Shift(1)
scl = Scale(2)
p2 = Polynomial2D(5)
p2.parameters = np.random.rand(21)
p22 = Polynomial2D(5)
p22.parameters = np.random.rand(21)
mapping = Mapping((0, 1, 0, 1))
m = tab & scl | mapping | p2 & p22
del offx, scl, p2, p22, mapping, tab
gc.collect()
#print(gc.garbage)
return m(x, x)
Results for python 3.5 and modeling/master
Line # Mem usage Increment Line Contents
================================================
29 11536.2 MiB 0.0 MiB @profile
30 def test_models(x):
31 12489.9 MiB 953.6 MiB tab = Tabular1D(lookup_table = np.arange(62500000), bounds_error=False)
32 12489.9 MiB 0.0 MiB offx = Shift(1)
33 12489.9 MiB 0.0 MiB scl = Scale(2)
34 12489.9 MiB 0.0 MiB p2 = Polynomial2D(5)
35 12489.9 MiB 0.0 MiB p2.parameters = np.random.rand(21)
36 12489.9 MiB 0.0 MiB p22 = Polynomial2D(5)
37 12489.9 MiB 0.0 MiB p22.parameters = np.random.rand(21)
38 12489.9 MiB 0.0 MiB mapping = Mapping((0, 1, 0, 1))
39 15351.1 MiB 2861.3 MiB m = tab & scl | mapping | p2 & p22
40 14397.5 MiB -953.7 MiB del offx, scl, p2, p22, mapping, tab
41 14397.5 MiB 0.0 MiB gc.collect()
42 #print(gc.garbage)
43 14397.5 MiB 0.0 MiB return m(x, x)
If Parameter.model is a weakref reference to model then the memory after deleting a model is cleaned
Line # Mem usage Increment Line Contents
================================================
29 1045.0 MiB 0.0 MiB @profile
30 def test_models(x):
31 1998.6 MiB 953.6 MiB tab = Tabular1D(lookup_table = np.arange(62500000), bounds_error=False)
32 1998.6 MiB 0.0 MiB offx = Shift(1)
33 1998.6 MiB 0.0 MiB scl = Scale(2)
34 1998.6 MiB 0.0 MiB p2 = Polynomial2D(5)
35 1998.6 MiB 0.0 MiB p2.parameters = np.random.rand(21)
36 1998.6 MiB 0.0 MiB p22 = Polynomial2D(5)
37 1998.6 MiB 0.0 MiB p22.parameters = np.random.rand(21)
38 1998.6 MiB 0.0 MiB mapping = Mapping((0, 1, 0, 1))
39 3906.0 MiB 1907.5 MiB m = tab & scl | mapping | p2 & p22
40 2952.4 MiB -953.6 MiB del offx, scl, p2, p22, mapping, tab
41 1045.0 MiB -1907.4 MiB gc.collect()
42 #print(gc.garbage)
43 1045.0 MiB 0.0 MiB return m(x, x)
Python 3.6 seems to collect cyclic references although there's nothing in the release notes about garbage collection improvements.
Results with python 3.6 and modeling/master
Line # Mem usage Increment Line Contents
================================================
29 1031.7 MiB 1031.7 MiB @profile
30 def test_models(x):
31 1985.2 MiB 953.5 MiB tab = Tabular1D(lookup_table = np.arange(62500000), bounds_error=False)
32 1985.2 MiB 0.0 MiB offx = Shift(1)
33 1985.2 MiB 0.0 MiB scl = Scale(2)
34 1985.2 MiB 0.0 MiB p2 = Polynomial2D(5)
35 1985.2 MiB 0.0 MiB p2.parameters = np.random.rand(21)
36 1985.2 MiB 0.0 MiB p22 = Polynomial2D(5)
37 1985.2 MiB 0.0 MiB p22.parameters = np.random.rand(21)
38 1985.2 MiB 0.0 MiB mapping = Mapping((0, 1, 0, 1))
39 3892.8 MiB 1907.5 MiB m = tab & scl | mapping | p2 & p22
40 2939.1 MiB -953.6 MiB del offx, scl, p2, p22, mapping, tab
41 1031.7 MiB -1907.4 MiB gc.collect()
42 #print(gc.garbage)
43 1031.7 MiB 0.0 MiB return m(x, x)