:) It does slow it down a little, but that's unavoidable. It's also
unavoidable in C, Fortran, Perl, etc. As long as the number of
operations in that loop is minimal, then it's the best you can do. If
it worries you, you could run a test where you call the target
function say 1e6 times, with and without the loop to see the timing
difference. Or simply running in Python 2:
for i in xrange(1000000):
x = 1
Then try:
for i in xrange(100000000):
x = 2
These two demonstrate the slowness of the Python loop. But the second
case is extreme and you won't encounter that much looping in these
functions. So while it is theoretically slower than C and Fortran
looping, you can probably see that no one would care :) Here is
another test, with Python 2 code:
"""
import cProfile as profile
def loop_1e6():
for i in xrange(int(1e6)):
x = 1
def loop_1e8():
for i in xrange(int(1e8)):
x = 1
def sum_conv():
for i in xrange(100000000):
x = 2 + 2.
def sum_normal():
for i in xrange(100000000):
x = 2. + 2.
def test():
loop_1e6()
loop_1e8()
sum_normal()
sum_conv()
profile.runctx('test()', globals(), locals())
"""
Running this on my system shows:
"""
7 function calls in 6.707 seconds
Ordered by: standard name
ncalls tottime percall cumtime percall filename:lineno(function)
1 0.000 0.000 6.707 6.707 <string>:1(<module>)
1 2.228 2.228 2.228 2.228 aaa.py:11(sum_conv)
1 2.228 2.228 2.228 2.228 aaa.py:15(sum_normal)
1 0.000 0.000 6.707 6.707 aaa.py:19(test)
1 0.022 0.022 0.022 0.022 aaa.py:3(loop_1e6)
1 2.228 2.228 2.228 2.228 aaa.py:7(loop_1e8)
1 0.000 0.000 0.000 0.000 {method 'disable' of
'_lsprof.Profiler' objects}
"""
That should be self explanatory. The better optimisation targets are
the repeated maths operations and the maths operations that can be
shifted into the target function or the target function
initialisation. Despite the numbers above which prove my int to float
speed argument as utter nonsense, it might still good to remove the
int to float conversions, if only to match the other functions.
Regards,
Edward
On 5 May 2014 18:45, Troels Emtekær Linnet <tlinnet@xxxxxxxxxxxxx> wrote:
The reason why I ask, is that I am afraid that this for loop slows
everything down.
What do you think?
2014-05-05 18:41 GMT+02:00 Edward d'Auvergne <edward@xxxxxxxxxxxxx>:
This is not Python specific though :) As far as I know, C uses
pass-by-value for arguments, unless they are arrays or other funky
objects/functions/etc.. This is the same behaviour as Python.
Pass-by-reference and pass-by-value is something that needs to be
mastered in all languages, whether or not you have pointers to play
with.
Regards,
Edward
On 5 May 2014 18:30, Troels Emtekær Linnet <tlinnet@xxxxxxxxxxxxx> wrote:
This reminds me:
http://combichem.blogspot.dk/2013/08/you-know-what-really-grinds-my-gears-in.html
2014-05-05 17:52 GMT+02:00 Edward d'Auvergne <edward@xxxxxxxxxxxxx>:
Hi,
This is an important difference. In the first case (back_calc[i] =
Minty[i]), what is happening is that your are copying the data into a
pre-existing structure. In the second case (back_calc = Minty), the
existing back_calc structure will be overwritten. Therefore the
back_calc structure in the calling code will be modified in the first
case but not the second. Here is some demo code:
def mod1(x):
x[0] = 1
def mod2(x):
x = [3, 2]
x = [0, 2]
print(x)
mod1(x)
print(x)
mod2(x)
print(x)
I don't know of a way around this.
Regards,
Edward
On 5 May 2014 17:42, Troels Emtekær Linnet <tlinnet@xxxxxxxxxxxxx> wrote:
Hi Edward.
In the library function of b14.py, i am looping over
the dispersion points to put in the data.
for i in range(num_points):
# The full formula.
back_calc[i] = Minty[i]
Why can I not just set:
back_calc = Minty
_______________________________________________
relax (http://www.nmr-relax.com)
This is the relax-devel mailing list
relax-devel@xxxxxxx
To unsubscribe from this list, get a password
reminder, or change your subscription options,
visit the list information page at
https://mail.gna.org/listinfo/relax-devel
Re: Why does one have to loop over the dispersion points?