"The advantage of xrange() over range() is minimal (since xrange()
still has to create the values when asked for them) except when a very
large range is used on a memory-starved machine or when all of the
range's elements are never used (such as when the loop is usually
terminated with break)." - from Python Library Reference.
I decided to measure the performance of range and xrange. I did it with
the following functions:
def rprint( n ):
a = time.time()
for i in range(n): print i
print time.time() - a
def xrprint( n ):
a = time.time()
for i in xrange(n): print i
print time.time() - a
def rpass( n ):
a = time.time()
for i in range(n): pass
print time.time() - a
def xrpass( n ):
a = time.time()
for i in xrange(n): pass
print time.time() - a
The results were as follows:
n rprint xrprint
10^4 0.37 s 0.34 s <- (1)
10^5 4.26 s 4.25 s
10^6 42.57 s 42.57 s
10^7 431.94 s 438.32 s <- (2)
n rpass xpass
10^4 0.0012 s 0.0011 s
10^5 0.0220 s 0.0139 s
10^6 0.1463 s 0.1298 s
10^7 1.4818 s 1.1807 s
The values are the average times printed by tested functions.
Conclusions:
1) According to (1) I could say that xrange might be somewhat faster
than range with lower numbers of iterations.
2) According to (2) I could say that xrange might be slower than range
with higher number of iterations.
The test with pass is not so important as the test with print (because
we usually do something inside of loops). So despite xpass has beaten
rpass, I would say that range is not slower than xrange (especially for
higher numbers of iterations). The final conclusion is : if you want
speed, you should use xrange privided that there aren't many
iterations. If you want less memory usage, you should use xrange.
I've also made more tests. The code was as follows:
-----------------------------
import array, random, time
def test1( n, size ):
a = time.time()
for i in xrange(n):
l = []
for i in xrange(size):
l.append( random.randint( 1,10 ) )
e = sum(l) / float(size) # just for taking some time
print time.time() - a
def test2( n, size ):
a = time.time()
l = []
for i in xrange(n):
del l[:]
for i in xrange(size):
l.append( random.randint( 1,10 ) )
e = sum(l) / float(size)
print time.time() - a
def test3( n, size ):
a = time.time()
l = range(size)
for i in xrange(n):
for i in xrange(size):
l[i] = random.randint( 1,10 )
e = sum(l) / float(size)
print time.time() - a
def test4( n, size ):
a = time.time()
l = array.array( 'L', xrange(size) )
for i in xrange(n):
for i in xrange(size):
l[i] = random.randint( 1,10 )
e = sum(l) / float(size)
print time.time() - a
def test5( n, size ):
a = time.time()
ind1 = range(size)
ind2 = range(size)
for i in xrange(n):
des1 = []
des2 = []
point = random.randint( 1, size-1 )
des1 = ind1[:point] + ind2[point:]
des2 = ind2[:point] + ind1[point:]
print time.time() - a
def test6( n, size ):
a = time.time()
ind1 = range(size)
ind2 = range(size)
des1 = []
des2 = []
for i in xrange(n):
del des1[:]
del des2[:]
point = random.randint( 1, size-1 )
des1 = ind1[:point] + ind2[point:]
des2 = ind2[:point] + ind1[point:]
print time.time() - a
def test7( n, size ):
a = time.time()
ind1 = range(size)
ind2 = range(size)
des1 = range(size)
des2 = range(size)
for i in xrange(n):
point = random.randint( 1, size-1 )
des1[:point] = ind1[:point]
des1[point:] = ind2[point:]
des2[:point] = ind2[:point]
des2[point:] = ind1[point:]
print time.time() - a
def test8( n, size ):
a = time.time()
ind1 = array.array( 'L', xrange(size) )
ind2 = array.array( 'L', xrange(size) )
des1 = array.array( 'L', xrange(size) )
des2 = array.array( 'L', xrange(size) )
for i in xrange(n):
point = random.randint( 1, size-1 )
des1[:point] = ind1[:point]
des1[point:] = ind2[point:]
des2[:point] = ind2[:point]
des2[point:] = ind1[point:]
print time.time() - a
-----------------------------
And this is my session with Python 2.4.1:
>>> for i in xrange(5): test.test1( 10000, 10 )
...
2.27345108986
2.51863479614
2.49968791008
2.68024802208
2.28194379807
>>> for i in xrange(5): test.test2( 10000, 10 )
...
2.54866194725
2.36415600777
2.71178197861
2.32558512688
2.71971893311
>>> for i in xrange(5): test.test3( 10000, 10 )
...
2.29083013535
2.5563249588
2.32064318657
1.90063691139
2.30613899231
>>> for i in xrange(5): test.test4( 10000, 10 )
...
2.55809211731
2.42571187019
2.59921813011
2.19631099701
2.16659498215
>>> for i in xrange(5): test.test5( 10000, 10 )
...
0.318142175674
0.442049980164
0.367480039597
0.327154874802
0.322648048401
>>> for i in xrange(5): test.test6( 10000, 10 )
...
0.356222867966
0.471677780151
0.332046031952
0.339803934097
0.48833990097
>>> for i in xrange(5): test.test7( 10000, 10 )
...
0.467595815659
0.317886829376
0.311239004135
0.312664031982
0.49030995369
>>> for i in xrange(5): test.test8( 10000, 10 )
...
0.499684095383
0.330184936523
0.332714080811
0.329524040222
0.50562787056
>>> for i in xrange(5): test.test5( 10000, 100 )
...
0.387717962265
0.45348906517
0.507198095322
0.402877807617
0.526827096939
>>> for i in xrange(5): test.test6( 10000, 100 )
...
0.525599002838
0.41659784317
0.443000078201
0.403271913528
0.591446876526
>>> for i in xrange(5): test.test7( 10000, 100 )
...
0.399652957916
0.416820049286
0.400202035904
0.404708862305
0.57714009285
>>> for i in xrange(5): test.test8( 10000, 100 )
...
0.357075929642
0.540817022324
0.378996133804
0.372053146362
0.554198980331
>>> for i in xrange(5): test.test5( 10000, 250 )
...
0.630347967148
0.497437000275
0.687075138092
0.497366905212
0.935706853867
>>> for i in xrange(5): test.test6( 10000, 250 )
...
0.493726015091
0.683156013489
0.512520074844
0.697488069534
0.746694803238
>>> for i in xrange(5): test.test7( 10000, 250 )
...
0.519948005676
0.583598136902
0.624222993851
0.528346061707
0.948079824448
>>> for i in xrange(5): test.test8( 10000, 250 )
...
0.553761005402
0.401547908783
0.389595985413
0.578064918518
0.394165039062
>>> for i in xrange(5): test.test1( 10000, 1000 )
...
233.676990032
229.95272994
228.739851952
228.541095018
226.404256105
>>> for i in xrange(5): test.test2( 10000, 1000 )
...
223.505224943
225.172422886
223.084803104
223.407966137
224.717788935
>>> for i in xrange(5): test.test3( 10000, 1000 )
...
210.81110096
211.956163168
212.362264156
211.730306149
209.519776106
>>> for i in xrange(5): test.test4( 10000, 1000 )
...
241.220864773
248.316150904
247.426213026
239.199230909
242.972666025
>>> for i in xrange(5): test.test5( 10000, 1000 )
...
1.32021999359
1.29506993294
1.14080190659
1.50338101387
1.30436086655
>>> for i in xrange(5): test.test6( 10000, 1000 )
...
1.28036403656
1.09035301208
1.07259607315
1.0751209259
1.07368779182
>>> for i in xrange(5): test.test7( 10000, 1000 )
...
1.38129281998
1.36377501488
1.40786099434
1.35044002533
1.37256002426
>>> for i in xrange(5): test.test8( 10000, 1000 )
...
0.524824142456
0.696274995804
0.544312000275
0.719218969345
0.77623295784
Tests with size equal to 10 were for testing a "small" size case. The
sizes 100 and 250 are more adequate in my case. The size 1000 is a
"big" size case. There are "random" tests (test1 ... test4) and "slice"
tests (test5 ... test8).
Conclusions:
1) Small size tests: there is no one winner of the random tests. The
differences are rather small and might be accidental. There is also no
winner of the slice tests.
2) Big size tests: as I expected, test3 is better than test2, and test
2 is better than test1. As I definitelly hadn't expected is the fact
that test4 was the worst (shouldn't arrays be more efficient than
lists?). So the winner of random tests is test3. And the winner in
slicing is test8.
I'm going to implement genetic algorothm, so I think that slicing tests
are more important than random tests. And the final conclusion is I
should use arrays instead of lists.
I'm going also to write tests that use Numeric.
--
http://mail.python.org/mailman/listinfo/python-list
