Re: [Databionic-ESOM-User] Weight errors while running ESOM!

[<fa...@my...>]

Hi Preeti,

it is possible that your problem is related to missing values or
calculation errors from division by zero or something
like that. this is hard to tell without seeing the data. can you send me
a sample to my personal email where the
problem occurs?

for best result you should be using the online algorithm, see
http://www.mathematik.uni-marburg.de/~databionics/de//downloads/papers/noecker06fast.pdf

the parameter selection is in general not all that crucial. the default
settings will most likely work well for many problems.

increasing the number of training epochs mainly means smoothing the
training process, since the radius and learning
rate are always cooled from start to end.

if you have a large dataset you might want to make the map bigger to
increase the resolution of you views into the high
dimensional data space.

pca weight initialization will give you more reproducible map layouts
but is very expensive. random works too and is faster.

more important than the learning rate is the radius and more important
than start values are end values. a larger end
radius will give you a smoother mapping, while a smaller end radius will
give you more details.

bye
fabian

Preeti Singh wrote:

> Hello! everyone,
>  
> I am new to ESOM and have been using it to classify a huge database.
> First of all, I have noticed that many times when I train the dataset,
> I save the weights (*.wts) and the best matching (*.bm) along the way.
> After the training is complete, I get a weird visualization, where
> most of the map is blue and white dots (best matching units) in
> between here and there. When I go into the weight file, I can see that
> it has some weird characters instead of numbers. I dont know what I am
> doing wrong? I have tried several times and I keep getting the same
> visualization and weight files most of the time. One out of 10
> trainings gives me the correct visualization. Can anyone help me in
> finding out where the mistake lies? second of all, how do you know
> what parameters are the optimal ones for running your dataset? For
> example, how many epochs I should run it for? Which training algorithm
> should I use? which weight initilization method is the best? start
> value for learning rate? etc.
>  
> Looking forward to hearing from some of you!
>  
> Best Regards,
> Preeti
>
>------------------------------------------------------------------------
>
>-------------------------------------------------------------------------
>Using Tomcat but need to do more? Need to support web services, security?
>Get stuff done quickly with pre-integrated technology to make your job easier
>Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
>http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642
>
>------------------------------------------------------------------------
>
>_______________________________________________
>Databionic-ESOM-User mailing list
>Dat...@li...
>https://lists.sourceforge.net/lists/listinfo/databionic-esom-user
>  
>