Hello..

To add Jordi's comment, modern data mining methods (SVM, RF, GBT, ...) have 
been designed to adapt small amount of sample sets. So, as Jordi has said, 
small polygons would be enough. It is however advisable to use remaining 
data (polygon) to validate the model (ComputeConfusionMatrix) to see if the 
model can adapt to a different data. I used to swap training and testing 
data to see if the model behaves differently. Fortunately, it is quite easy 
to do that in OTB. My biggest salute to OTB developers..

Saluti,
Bambang

On Sunday, May 29, 2016 at 10:21:19 PM UTC+10, Jordi Inglada wrote:
>
> Hristo Nikolov <[email protected] <javascript:>> wrote: 
> > 
> > hi, 
> > when using TrainImagesClassifier in the process of training i'm getting 
> this 
> > message: 
> > 
> > WARNING: using -h 0 may be faster 
> > 
> > how could i set this parameter for the gui or cli version? 
> > i read the info for this module, but no information for this option. 
> > since the learning takes long time any speeding is highly desirable. 
>
> Hi Hristo, 
>
> This warning message comes from the libsvm, but OTB's interface does no 
> expose this option (shrinking). 
>
> One thing that makes SVM very slow is using too many samples. I don't know 
> if this is your case, but most of the time, we tend to use huge polygons 
> for training a classifier and these polygons contain many redundant (nearly 
> identical) pixels. 
>
> You may want to use smaller training areas if this is possible for your 
> application. 
>
> For an interesting discussion on the shrinking topic, see [1]. 
>
> Good luck. 
>
> Jordi 
>
> [1] 
> http://stackoverflow.com/questions/12500261/libsvm-shrinking-heuristics 
>
> > 
> > thank you. 
> > hristo 
> > 
> > -- 
>

-- 
-- 
Check the OTB FAQ at
http://www.orfeo-toolbox.org/FAQ.html

You received this message because you are subscribed to the Google
Groups "otb-users" group.
To post to this group, send email to [email protected]
To unsubscribe from this group, send email to
[email protected]
For more options, visit this group at
http://groups.google.com/group/otb-users?hl=en
--- 
You received this message because you are subscribed to the Google Groups 
"otb-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
For more options, visit https://groups.google.com/d/optout.

Reply via email to