New submission from Alex Gaynor <[email protected]>:
Not sure what causes this yet, playing with the number of iterations PyPy is
consistenly slower than CPython, I think it's because of the super high number
of
bridges compiled:
Alexanders-MacBook-Pro:tmp alex_gaynor$ PYPYLOG=jit-summary:- pypy
decision_tree.py
[ac4ea2fa4a5] {jit-summary
Tracing: 3553 19.518618
Backend: 3552 2.996290
TOTAL: 51.207841
ops: 5654103
recorded ops: 2932390
calls: 95188
guards: 663823
opt ops: 599629
opt guards: 176790
forcings: 3622
abort: trace too long: 1
abort: compiling: 0
abort: vable escape: 0
abort: bad loop: 0
abort: force quasi-immut: 0
nvirtuals: 3885619
nvholes: 518637
nvreused: 2282863
Total # of loops: 21
Total # of bridges: 3533
Freed # of loops: 7
Freed # of bridges: 3
[ac4ea6187bd] jit-summary}
Alexanders-MacBook-Pro:tmp alex_gaynor$ cat decision_tree.py
import random
class Node(object):
def __init__(self, idx, thresh, left, right):
self.idx = idx
self.thresh = thresh
self.left = left
self.right = right
def predict(self, x):
if x[self.idx] < self.thresh:
return self.left.predict(x)
else:
return self.right.predict(x)
def __eq__(self, other):
return other.__class__ is Node and \
self.idx == other.idx and \
self.thresh == other.thresh and \
self.left == other.left and \
self.right == other.right
class Leaf(object):
def __init__(self, value):
self.value = value
def predict(self, x):
return self.value
def __eq__(self, other):
return other.__class__ is Leaf and \
other.value == self.value
def gen_random_tree(n_features, depth):
#print "n_features = %d, depth = %d" % (n_features, depth)
if depth == 0:
return Leaf(random.random())
else:
feature_idx = random.randint(0, n_features-1)
threshold = random.random()
left = gen_random_tree(n_features, depth - 1)
right = gen_random_tree(n_features, depth - 1)
return Node(feature_idx, threshold, left, right)
def predict_labels(feature_list, tree):
return [tree.predict(v) for v in feature_list]
def gen_random_tuples(n_items, n_features):
return [tuple([random.random() for _ in range(n_features)])
for _ in range(n_items)]
if __name__ == '__main__':
n_samples = 25000
n_features = 10
for i in xrange(100):
tree = gen_random_tree(n_features, depth = 12)
features = gen_random_tuples(n_samples, n_features)
predict_labels(features, tree)
----------
messages: 6070
nosy: agaynor, pypy-issue
priority: performance bug
status: unread
title: Pathological case in bridges
________________________________________
PyPy bug tracker <[email protected]>
<https://bugs.pypy.org/issue1584>
________________________________________
_______________________________________________
pypy-issue mailing list
[email protected]
http://mail.python.org/mailman/listinfo/pypy-issue
