Re: generate De Bruijn sequence memory and string vs lists

From	Peter Otten <__peter__@web.de>
Subject	Re: generate De Bruijn sequence memory and string vs lists
Date	2014-01-24 09:23 +0100
Organization	None
References	(4 earlier) <lbrst6$vel$1@ger.gmane.org> <lbrugd$k3a$1@ger.gmane.org> <CALyJZZU2J3n67LwF1a2-6eeMZ=J18zbj8h+QV-w7Ujz=kHkgOw@mail.gmail.com> <lbs0nm$f4t$1@ger.gmane.org> <CALyJZZU8R8pn9sCNgTqn9ieajayjFjkoG6sQCS6xWWLhxhWm-w@mail.gmail.com>
Newsgroups	comp.lang.python
Message-ID	<mailman.5932.1390551780.18130.python-list@python.org> (permalink)

Show all headers | View raw

Vincent Davis wrote:

> Excellent Peter!
> I have a question, the times reported don't make sense to me, for example
> $ python3 -m timeit -s 'from debruijn_compat import debruijn_bytes as d'
> 'd(4, 8)'
> 100 loops, best of 3: 10.2 msec per loop
> This took ~4 secs (stop watch) which is much more that 10*.0102 Why is
> this?

Look at the output, it's "100 loops"

100 * 10 msec == 1 sec

together with "best of 3" you are already at 3 sec, minimum as it is the 
"best" run.

Then, how do you think Python /knows/ that it has to repeat the code 10 
times on my "slow" and 100 times on your "fast" machine? It runs the bench 
once, then 10, then 100, then 1000 times -- until there's a run that takes 
0.2 secs or more. The total expected minimum time without startup overhead 
is then

+----calibration------+   +-measurement--+
(1 + 10 + 100) * 10msec + 3 * 100 * 10msec

or about 4 secs.

> $ python3 -m timeit -s 'from debruijn_compat import debruijn_bytes as d'
> 'd(4, 11)'
> 10 loops, best of 3: 480 msec per loop
> This took ~20 secs vs .480*10

>>> .480*(1 + 10 + 3*10)
19.68

> d(4, 14) takes about 24 seconds (one run)

This is left as an exercise ;)

Back to comp.lang.python | Previous | Next | Find similar | Unroll thread

Thread

Re: generate De Bruijn sequence memory and string vs lists Peter Otten <__peter__@web.de> - 2014-01-24 09:23 +0100

csiph-web