[Numpy-discussion] Convolve returning zero array

[Peter Verveer]

Hi Remco,

> OK, here is the example:

[code deleted]

> When I run this code, the first convolution gives an array of zeros, the
> second convolution is good.

The problem is that the output of the convolve functions (both the convolve 
and the nd_image versions) return the result with the same type as the input. 
In your first case the input was of Bool type and thus the result could only 
be 0 or 1. Conversion to Float64 before calculation seems to fix the problem. 
(I inserted 'data1 = inputarray(data1, Float64)' before the function call.) 
Alternatively in nd_image.convolve1d you can use the output_type flag to save 
you the conversion.

> I also did a quick try with convolve from nd_image, but this function was
> slower than the convolve from numarray with a large Gaussian distribution.

Yes, nd_image convolve is designed for the general multi-dimensional case and 
may therefore be slower in some cases. Also I think the convolve version uses 
an internal conversion of the input to double type which appears to faster. 
The nd_image version does not do that because for large multi-dimensional 
arrays that may cost a lot of memory. I think if you do the conversion to 
Float64 before the speed should be closer. For one-dimensional cases, you 
should probably just stick to the convolve version.

I hope this solved your problem.

Cheers, Peter

-- 
Dr. Peter J. Verveer
Cell Biology and Cell Biophysics Programme
European Molecular Biology Laboratory
Meyerhofstrasse 1
D-69117 Heidelberg
Germany
Tel. : +49 6221 387245
Fax  : +49 6221 387306