Square2rect

[WeatherGod]

This should make the hungarian algorithm accept rectangular cost matrices. Also enabled the tests.
NOTE: Only tested on rectangular matrices of shape n-by-m such that m > n. Tests need to be expanded to test m < n.

[GaelVaroquaux]

Looks good thanks.

Out of curiosity, are you using the Hungarian for a machine learning related task? I have been a bit slow at merging it, because the machine-learning usage that I had went away, and I hate merging code in the scikit without a clear cut usecase that fits in the big picture.

[WeatherGod]

On Mon, Jun 6, 2011 at 4:29 PM, GaelVaroquaux <
reply@reply.github.com>wrote:

Looks good thanks.

Out of curiosity, are you using the Hungarian for a machine learning
related task? I have been a bit slow at merging it, because the
machine-learning usage that I had went away, and I hate merging code in the
scikit without a clear cut usecase that fits in the big picture.

Reply to this email directly or view it on GitHub:
#2 (comment)

No, I am using it for storm cell tracking. No machine-learning here. Also,
note that I have not rigorously verified the results from these changes.
The results when used in a tracker seems to make sense.

If it doesn't seem to belong in scikits, I am fine with it going somewhere
else, but I have been wanting a hungarian solver within some part of scipy
packages for a while now.

Ben Root

[GaelVaroquaux]

It is actually used a lot in the machine learning community; it's just that in the scikit, we don't currently have a usage for it. I use it in machine-learning related code that is not yet in the scikit.

The code is fairly stand alone and under a BSD license, so it could fit in scipy too, but it is probably too specific for that.

I still think that it should be integrated in the scikit-learn, I just need to find a good usecase.