How to determine if printing is being a bottleneck in my code?

[SherjilOzair]

Hello list,

When it comes to printing things while some computation is being done, there are 2 extremes.

1. printing speed is slower than data-to-print generation speed. In this case, printing is a bottleneck. Examples: "for i in xrange(2**30): print i". Without the print, this code would be much faster.

2. data-to-print generation speed is slower than printing speed. So, this case, printing does now slow you down much. Example: for m in matrices: print m.inverse() # inverse is a time-taking function

These two cases are pretty easy. But, my question is, how to draw the line? How do I know that print is slowing me down, and I should probably remove some of them? Is there a scientific way to do this, rather than just intuition and guesswork?

I can clarify, if needed.
Thanks,
Sherjil Ozair

 

[Roy Smith]

Hello list,

When it comes to printing things while some computation is being done, there
are 2 extremes.

1. printing speed is slower than data-to-print generation speed. In this
case, printing is a bottleneck. Examples: "for i in xrange(2**30): print i".
Without the print, this code would be much faster.

2. data-to-print generation speed is slower than printing speed. So, this
case, printing does now slow you down much. Example: for m in matrices: print
m.inverse() # inverse is a time-taking function

These two cases are pretty easy. But, my question is, how to draw the line?
How do I know that print is slowing me down, and I should probably remove
some of them? Is there a scientific way to do this, rather than just
intuition and guesswork?

I can clarify, if needed.
Thanks,
Sherjil Ozair

The profiler (http://docs.python.org/2/library/profile.html) is your
friend.

 

[rusi]

The profiler (http://docs.python.org/2/library/profile.html) is your
friend.

One added caveat: When 'printing' is the bottleneck it can be
- in the actual disk/console/network I/O
- in the str/repr builtin/customized that precedes it

[Or both of course]

So after you identify printing as the bottleneck, it may be worthwhile
to print to a dummy device.
On Unix one usually uses /dev/null for this but whether that will help
clarify or muddy the picture I am not sure (Whats the overhead to
writing to null?)
A more internal-to-python method may be to replace the print with a
str/repr assigned to say a global variable