[SciPy-dev] huge speed regression in loadmat from 0.6.0 to 0.7.0

Matthew Brett matthew.brett@gmail....
Wed Feb 18 01:30:01 CST 2009


Hi,

> So now the GzipInputStream overhead is only about 20-25%. Still seems
> a bit higher than it should be, but certainly usable, and worth it for
> the memory win.
>
> BTW, the name GzipInputStream is very confusing for something that
> reads raw deflate format and not, say, gzip format :-).

Thanks to all in this thread.  I've committed a version of this patch
with the final suggestions here, and renamed the class to
ZlibInputStream.

I've also put in a two shot version of the class that's a bit better
matched to the matlab io situation, where we want to be able to do one
small read for the variable name, then a long read for the rest of the
data.

Chris Burns kindly put some benchmarks in place.

Could you test current SVN and let me know how speed is for you?

Matthew


More information about the Scipy-dev mailing list