Should be clearer if you sort by cumtime.
Python: Interpreting cProfile logs
-
18-03-2022 - |
質問
I have a method to determine the most compatible person for every other person. Basically there are two nested loops over the items of a dict that maps from a person to a list (where similar lists determine compatibility), there compat is computed and saved if it is bigger then the previously maximal one for the person of the outer loop.
So I decided to optimize performance by updating the compatibility for the other person (the on in the inner loop) as well, because the compatibility is the same, and I won't have to do the same computation when the outer loop reaches person 2 and the inner one person 1 [use symmetrie of compatibility relation].
Well, I ended up 20 times slower. The c-profile logs are strange, because all operations of the improved version have a better (or similar) totaltime than the ones in the unimproved code. So I'm absolutely stuck finding the bottleneck. :(
Can anybody give me advice on how to interpret this logs? Where is the evil instruction gone?
log of normal code:
$ python -m cProfile -s time ./jukebox.py sample.txt
92661414 function calls (92661412 primitive calls) in 124.355 CPU seconds
Ordered by: internal time
ncalls tottime percall cumtime percall filename:lineno(function)
10000 93.324 0.009 124.168 0.012 jukebox.py:88(solve_problem_4)
42900000 16.616 0.000 16.616 0.000 {method 'intersection' of 'set' objects}
42900000 10.831 0.000 10.831 0.000 {len}
6180396 2.212 0.000 2.212 0.000 {method 'append' of 'list' objects}
670000 1.185 0.000 1.185 0.000 {method 'items' of 'dict' objects}
1 0.170 0.170 124.353 124.353 jukebox.py:1(<module>)
1 0.009 0.009 0.013 0.013 heapq.py:31(<module>)
1 0.004 0.004 0.004 0.004 bisect.py:1(<module>)
1 0.002 0.002 124.355 124.355 {execfile}
66 0.001 0.000 0.001 0.000 jukebox.py:18(update_bands)
67 0.001 0.000 0.001 0.000 fileinput.py:166(isfirstline)
1 0.000 0.000 0.002 0.002 jukebox.py:9(__init__)
1 0.000 0.000 0.000 0.000 {open}
198 0.000 0.000 0.000 0.000 {method 'strip' of 'str' objects}
132 0.000 0.000 0.000 0.000 {method 'split' of 'str' objects}
1 0.000 0.000 124.355 124.355 <string>:1(<module>)
1 0.000 0.000 0.000 0.000 fileinput.py:240(__iter__)
68 0.000 0.000 0.000 0.000 fileinput.py:243(next)
1 0.000 0.000 0.000 0.000 {range}
1 0.000 0.000 0.000 0.000 {method 'close' of 'file' objects}
1 0.000 0.000 0.000 0.000 {isinstance}
1 0.000 0.000 0.000 0.000 fileinput.py:80(<module>)
1 0.000 0.000 0.000 0.000 fileinput.py:184(FileInput)
1 0.000 0.000 0.000 0.000 fileinput.py:197(__init__)
2 0.000 0.000 0.000 0.000 {method 'readlines' of 'file' objects}
1 0.000 0.000 0.000 0.000 {method 'disable' of '_lsprof.Profiler' objects}
4/2 0.000 0.000 0.000 0.000 fileinput.py:292(readline)
396 0.000 0.000 0.000 0.000 {method 'setdefault' of 'dict' objects}
1 0.000 0.000 0.000 0.000 fileinput.py:91(input)
1 0.000 0.000 0.000 0.000 fileinput.py:266(nextfile)
1 0.000 0.000 0.000 0.000 jukebox.py:4(Reader)
67 0.000 0.000 0.000 0.000 fileinput.py:371(isfirstline)
log of "optimized" one:
$ python -m cProfile -s time ./jukebox-imp.py sample.txt
49761414 function calls (49761412 primitive calls) in 2166.613 CPU seconds
Ordered by: internal time
ncalls tottime percall cumtime percall filename:lineno(function)
10000 2147.248 0.215 2165.759 0.217 jukebox-imp.py:88(solve_problem_4)
21450000 8.952 0.000 8.952 0.000 {method 'intersection' of 'set' objects}
21450000 5.951 0.000 5.951 0.000 {len}
6180396 2.152 0.000 2.152 0.000 {method 'append' of 'list' objects}
660000 1.441 0.000 1.441 0.000 {method 'items' of 'dict' objects}
1 0.837 0.837 2166.611 2166.611 jukebox-imp.py:1(<module>)
10000 0.015 0.000 0.015 0.000 {method 'keys' of 'dict' objects}
1 0.010 0.010 0.013 0.013 heapq.py:31(<module>)
1 0.003 0.003 0.003 0.003 bisect.py:1(<module>)
1 0.002 0.002 2166.613 2166.613 {execfile}
66 0.002 0.000 0.002 0.000 jukebox-imp.py:18(update_bands)
1 0.000 0.000 0.000 0.000 {open}
198 0.000 0.000 0.000 0.000 {method 'strip' of 'str' objects}
132 0.000 0.000 0.000 0.000 {method 'split' of 'str' objects}
1 0.000 0.000 2166.613 2166.613 <string>:1(<module>)
1 0.000 0.000 0.000 0.000 fileinput.py:240(__iter__)
1 0.000 0.000 0.002 0.002 jukebox-imp.py:9(__init__)
68 0.000 0.000 0.000 0.000 fileinput.py:243(next)
1 0.000 0.000 0.000 0.000 {range}
1 0.000 0.000 0.000 0.000 {method 'close' of 'file' objects}
1 0.000 0.000 0.000 0.000 {isinstance}
1 0.000 0.000 0.000 0.000 fileinput.py:80(<module>)
1 0.000 0.000 0.000 0.000 fileinput.py:184(FileInput)
1 0.000 0.000 0.000 0.000 fileinput.py:197(__init__)
2 0.000 0.000 0.000 0.000 {method 'readlines' of 'file' objects}
1 0.000 0.000 0.000 0.000 {method 'disable' of '_lsprof.Profiler' objects}
4/2 0.000 0.000 0.000 0.000 fileinput.py:292(readline)
396 0.000 0.000 0.000 0.000 {method 'setdefault' of 'dict' objects}
67 0.000 0.000 0.000 0.000 fileinput.py:166(isfirstline)
1 0.000 0.000 0.000 0.000 fileinput.py:91(input)
1 0.000 0.000 0.000 0.000 fileinput.py:266(nextfile)
67 0.000 0.000 0.000 0.000 fileinput.py:371(isfirstline)
1 0.000 0.000 0.000 0.000 jukebox-imp.py:4(Reader)
//EDIT:
Just in case I can provide the code, too. To my humble understanding there is absolutely no reason for the latter to be 20x slower then the former.
The normal one:
def solve_problem_4(colleagues):
MIN_COMPAT = 1
compat_dict = dict()
for colleague_1, bands_1 in colleagues.items():
compat_dict[colleague_1] = (0,[])
for colleague_2, bands_2 in colleagues.items():
if colleague_1 == colleague_2:
continue
compat = len(set(bands_1).intersection(set(bands_2)))
if compat > MIN_COMPAT:
old_compat,top_colleagues = compat_dict[colleague_1]
if compat > old_compat:
compat_dict[colleague_1] = (compat,[colleague_2])
elif compat == old_compat:
top_colleagues.append(colleague_2)
return compat_dict
And the "optmized":
def solve_problem_4(colleagues):
MIN_COMPAT = 1
compat_dict = defaultdict(lambda: (0,[])) #change here
checked_pairs = []
for colleague_1, bands_1 in colleagues.items()[:-1]:
for colleague_2, bands_2 in colleagues.items():
if colleague_1 == colleague_2 or (colleague_2,colleague_1) in checked_pairs: # change here, exclude used pairs
continue
checked_pairs += [(colleague_1,colleague_2)] # change here, note down checked pair
compat = len(set(bands_1).intersection(set(bands_2)))
if compat > MIN_COMPAT:
old_compat, top_colleagues = compat_dict[colleague_1]
if compat > old_compat:
compat_dict[colleague_1] = compat,[colleague_2]
elif compat == old_compat:
top_colleagues.append(colleague_2)
old_compat, top_colleagues = compat_dict[colleague_2] # change here, update symmetric pair
if compat > old_compat: # imagine extract method refactoring here ;)
compat_dict[colleague_2] = compat,[colleague_1]
elif compat == old_compat:
top_colleagues.append(colleague_1)
return compat_dict
解決 2
他のヒント
Alternatively, running runsnakerun on cProfile dump provides an easy to understand graphical view.
python -m cProfile -o dump.cprofile script.py
runsnakerun dump.cprofile