37 mb - prime1

[francky]

I read in another thread here :

you can see that all fast Python AC submissions for that problem use about 37 MB, which is an indicator for the use of psyco.

I use python3 to code, and thought psyco was not possible.
But I would like to know if it was used for the best solutions in python3 for PRIME1 ?
I have a good solution with 5.7MB of memory print, and under 1s.
Better solutions use about 37MB. Is it a sign of psyco ?

I'm new here (not a real newbie, but these kind of challenges are new for me).

1) Please tell me if better solutions are obtained with psyco (or an equivalent way) or simply a different algo.

2) Where can I learn to read/write faster on I/O. I already use : sys.stdin and sys.stdout, (basically ?).

Thanks. :bowdown:

(I use Francky too on Project Euler, in case you crossed my posts)

[leppyr64]

Hey francky. I have seen you on Project Euler before. I have seen psyco used in python solutions here but don't use python myself.

numerix and hendrik are the resident python gurus around here. I'm sure they will chime in eventually.

In the meantime I would suggest checking out numerix' posts.

[numerix]

I have a good solution with 5.7MB of memory print, and under 1s.
Better solutions use about 37MB. Is it a sign of psyco ?

That is definitely the fastest Python 3.x solution so far. All faster solutions made use of psyco.
But how is that possible as psyco isn't available for Python 3.x?
Explanation: During a time from appr. Sep. 2009 until May/June 2010 Python 2.6 was added - before Sep. 2009 only Python 2.5 was available.
After a starting period without psyco they made psyco available also for Python 2.6. In May/June 2010 Python 2.6 was replaced by Python 3.1.2, and unfortunately all former Python 2.6 solutions were labeled as Python 3.1.2.

As you can see from my PRIME1 submission list, I have a Python 2.5 solution without psyco running in 0.31 s. I didn't try a really fast Python 3.1 solution until now, but your algo with runtime < 1 s seems to be a rather good implementation. I don't think that runtimes below 0.5 s will be possible. But we'll see ...

2) Where can I learn to read/write faster on I/O. I already use : sys.stdin and sys.stdout, (basically ?).

That's not the worst way. In fact the fastest way of I/O depends on the Python version used - Python 2.5 has other bottlenecks than Python 3.1 and you need other solutions to make them wider - and if you use psyco, which opens new ways of doing it even faster. But that's what you can find out yourself with some experimenting on your local machine.

[francky]

Many thanks numerix,
I was a little in trouble ( :crazyeyes: ) with that, now I understand well the problem.

Those 'python2.6w/o/wt psyco' should be rejudged, imho.
Edit : but it seems quite impossible...