Re: advanced Weka segmentation

[Anda Cornea]

Hello!

I am a very happy end user of the weka segmentation with very limited understanding of the algorithm.  Can anyone recommend any easy reading of general concepts and practical implications of the large choice of training features?

Thank you in advance,

Anda



Anda Cornea, PhD
Director of the Imaging and Morphology Support Core
Oregon National Primate Research Center
Oregon Health & Science University
503-690-5293

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[Ignacio Arganda-Carreras]

Dear Anda,

I can only think on this old survey from 1997:
http://machine-learning.martinsewell.com/feature-selection/DashLiu1997.pdf

In general, feature selection helps reducing the amount of memory to use
during the training and the posterior classification. In AWS, the default
classifier is a random forest, which works very well even in the presence
of too many non-informative features. If you have enough RAM, my advice is
that you start playing around with as many features as possible and then
keep reducing them to select the best of them.

The Weka explorer has an option to perform feature selection as well.

I hope this helps!

ignacio




On Tue, Feb 5, 2013 at 4:52 PM, Anda Cornea <[hidden email]> wrote:

> Hello!
>
> I am a very happy end user of the weka segmentation with very limited
> understanding of the algorithm.  Can anyone recommend any easy reading of
> general concepts and practical implications of the large choice of training
> features?
>
> Thank you in advance,
>
> Anda
>
>
>
> Anda Cornea, PhD
> Director of the Imaging and Morphology Support Core
> Oregon National Primate Research Center
> Oregon Health & Science University
> 503-690-5293
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>

... [show rest of quote]

--
Ignacio Arganda-Carreras, Ph.D.
Seung's lab, 46-5065
Department of Brain and Cognitive Sciences
Massachusetts Institute of Technology
43 Vassar St.
Cambridge, MA 02139
USA

Phone: [phone]
Website: http://bioweb.cnb.csic.es/~iarganda/index_EN.html

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[Anda Cornea]

Dear Ignacio, thank you so much!  Both the pdf and your info are quite helpful.

Thanks again,

Anda



-----Original Message-----
From: ImageJ Interest Group [mailto:[hidden email]] On Behalf Of Ignacio Arganda-Carreras
Sent: Tuesday, February 05, 2013 7:10 PM
To: [hidden email]
Subject: Re: advanced Weka segmentation

Dear Anda,

I can only think on this old survey from 1997:
http://machine-learning.martinsewell.com/feature-selection/DashLiu1997.pdf

In general, feature selection helps reducing the amount of memory to use during the training and the posterior classification. In AWS, the default classifier is a random forest, which works very well even in the presence of too many non-informative features. If you have enough RAM, my advice is that you start playing around with as many features as possible and then keep reducing them to select the best of them.

The Weka explorer has an option to perform feature selection as well.

I hope this helps!

ignacio




On Tue, Feb 5, 2013 at 4:52 PM, Anda Cornea <[hidden email]> wrote:

> Hello!
>
> I am a very happy end user of the weka segmentation with very limited
> understanding of the algorithm.  Can anyone recommend any easy reading
> of general concepts and practical implications of the large choice of
> training features?
>
> Thank you in advance,
>
> Anda
>
>
>
> Anda Cornea, PhD
> Director of the Imaging and Morphology Support Core Oregon National
> Primate Research Center Oregon Health & Science University
> 503-690-5293
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>

... [show rest of quote]

--
Ignacio Arganda-Carreras, Ph.D.
Seung's lab, 46-5065
Department of Brain and Cognitive Sciences Massachusetts Institute of Technology
43 Vassar St.
Cambridge, MA 02139
USA

Phone: [phone]
Website: http://bioweb.cnb.csic.es/~iarganda/index_EN.html

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html