Machine learning in ShapeLogic 1.5 plugin

[Sami Badawi-2]

ShapeLogic 1.5 plugin has been released.

The categorizer in the Color Particle Analyzer can now use either a
rule set or a multi-layer feed forward neural network.  This is the
first release to combine machine learning and declarative programming.
 The neural network needs to be trained externally, and you need to
override a class to import weights. This override is simple but does
require some knowledge of Java.

Override example: http://www.shapelogic.org/particle.html#override
Machine learning: http://www.shapelogic.org/machine-learning.html
Download page: http://code.google.com/p/shapelogic/downloads/list

-Sami Badawi

[Francesco De Comite]

Sami Badawi a écrit :

>ShapeLogic 1.5 plugin has been released.
>
>The categorizer in the Color Particle Analyzer can now use either a
>rule set or a multi-layer feed forward neural network.  This is the
>first release to combine machine learning and declarative programming.
> The neural network needs to be trained externally, and you need to
>override a class to import weights. This override is simple but does
>require some knowledge of Java.
>
>Override example: http://www.shapelogic.org/particle.html#override
>Machine learning: http://www.shapelogic.org/machine-learning.html
>Download page: http://code.google.com/p/shapelogic/downloads/list
>
>-Sami Badawi
>  
>

did you consider using weka as machine learning engine ?

www.cs.waikato.ac.nz/ml/weka/

should avoid the external processing part

[Sami Badawi-2]

Hi Francesco,

Thanks for the suggestion, I looked at different good open source
machine learning packages, and spend a lot of times trying to find the
most user friendly approach.

I am interested in applying supervised learning to the color particle
analyzer.  To train the neural network you need a list of images each
with one or more particles, here are some of the things you need to
do:

* You have to assign different categories to the different particles
* You need to select which geometric property data streams you want as input
* This should be a minimal list otherwise you can get over fitting problems
* Possibly create new geometric property data streams
* Find out how many neural network layers you need
* Find out how many nodes in each layer you need
* Store this information so it can be reused

ImageJ has the simple and powerful PlugInFilter concept and I do not
see how I can get all that functionality into a PlugInFilter without
creating something that is quite complex.

I found Joone - Java Object Oriented Neural Engine
http://www.jooneworld.com/ to be a great tool, but decided against
integrating it directly since it I was concerned about adding
complexity and too many extra external dependencies.

In ShapeLogic 1.6, I am going to add a simple reader that can load the
network weights, names of geometric property data streams and names of
categories, so you can use an externally trained neural network
without any programming, as long as it does not use any new geometric
property data streams.

-Sami Badawi
http://www.shapelogic.org



2009/4/14 Francesco De Comité <[hidden email]>:

> Sami Badawi a écrit :
>
>> ShapeLogic 1.5 plugin has been released.
>>
>> The categorizer in the Color Particle Analyzer can now use either a
>> rule set or a multi-layer feed forward neural network.  This is the
>> first release to combine machine learning and declarative programming.
>> The neural network needs to be trained externally, and you need to
>> override a class to import weights. This override is simple but does
>> require some knowledge of Java.
>>
>> Override example: http://www.shapelogic.org/particle.html#override
>> Machine learning: http://www.shapelogic.org/machine-learning.html
>> Download page: http://code.google.com/p/shapelogic/downloads/list
>>
>> -Sami Badawi
>>
>
> did you consider using weka as machine learning engine ?
>
> www.cs.waikato.ac.nz/ml/weka/
>
> should avoid the external processing part
>