Re: [otb-users] Re: ANN produce not stable results

[Hristo Nikolov]

Hi,
I can't see such messages when the model is being built by the GUI 
application.
Could share the version of the OTB you use?

Hristo

On Wednesday, 10 February 2016 16:31:48 UTC+2, Guillaume Pasero wrote:
>
> Hi, 
>
> I think it is possible to get outputs from libSVM during processing.
> When I use this library, I get messages like :
>
> optimization finished, #iter = 113
> nu = 0.038199
> obj = -22.175591, rho = 0.331874
> nSV = 38, nBSV = 28
>
>  Do you get these messages when running the command line application ?
>
> Guillaume
>
> On 02/03/2016 11:20 AM, Hristo Nikolov wrote:
>
> Dear Guillaume, 
> Thank you for your suggestion. As I pointed above I use the GUI version of 
> OTB.
> What I shall do is to follow your advice and move to CLI.
>
> Just a small addition - the libSVM runs for more than 7 days and still not 
> finished.
> Is it possible to integrate some parameter that shows the convergence when
> running the learning phase of the modules in this library? 
> By now the only way for me is to check the amount of memory used by this
> process which is not a good estimate for this purpose.
>
> Hristo
>
> On Friday, 29 January 2016 10:59:44 UTC+2, Guillaume Pasero wrote: 
>>
>> Hi,
>>
>> In order to have stable results you should use the seed parameter, so 
>> that the Random Number Generator used during sample selection always gives 
>> the same samples : in the otbcli , it is the parameter key '-rand'.
>>
>> For the ANN, with the parameter "-classifier.ann.sizes", you can give 
>> multiple sizes to have several layers.
>>
>> HTH,
>>
>> Guillaume
>>
>> On 01/18/2016 05:41 PM, Hristo Nikolov wrote:
>>
>> One more observation - when using polynomial as kernel function I can not 
>> set the degree of the polynomial.
>>
>> On Wednesday, 13 January 2016 15:40:40 UTC+2, Hristo Nikolov wrote: 
>>>
>>> Dear all, 
>>> I have been experimenting with ANN provided from OTB for some months 
>>> now. 
>>> What I experienced is that as module in QGIS the ANN from OTB has 
>>> limitation
>>> of creating only single layer networks, but the results produced remain 
>>> the same
>>> under same initial parameters. Lastly I tried to the otbgui version 
>>> provided by 
>>> OSGeo4W, but it results that even the parameters in the dialog box are 
>>> the 
>>> same the output produced and more specifically the error matrix is 
>>> different 
>>> every time the training takes place.
>>> Could you advise?
>>>
>>> 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 otb-...@googlegroups.com
>> To unsubscribe from this group, send email to
>> otb-users+...@googlegroups.com
>> 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 otb-users+...@googlegroups.com.
>> For more options, visit https://groups.google.com/d/optout.
>>
>>
>> -- 
>>  *Guillaume PASERO*
>> Responsable technique
>> *Business Unit E-SPACE & Geo Information - Département Image & 
>> Applications*
>>
>> *CS Systèmes d'Information*
>> Parc de la Grande Plaine - 5, Rue Brindejonc des Moulinais - BP 15872
>> 31506 Toulouse Cedex 05 - FRANCE
>> +33 561 17 64 21 - guillaum...@c-s.fr 
>>
> -- 
> -- 
> 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 otb-...@googlegroups.com 
> 
> To unsubscribe from this group, send email to
> otb-users+...@googlegroups.com 
> 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 otb-users+...@googlegroups.com .
> For more options, visit https://groups.google.com/d/optout.
>
>
> -- 
>  *Guillaume PASERO*
> Responsable technique
> *Business Unit E-SPACE & Geo Information - Département Image & 
> Applications*
>
> *CS Systèmes d'Information*
> Parc de la Grande Plaine - 5, Rue Brindejonc des Moulinais - BP 15872
> 31506 Toulouse Cedex 05 - FRANCE
> +33 561 17 64 21 - guillaum...@c-s.fr  
>

-- 
-- 
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 otb-users@googlegroups.com
To unsubscribe from this group, send email to
otb-users+unsubscr...@googlegroups.com
For more options, visit this group at
http://groups.google.com/group/otb-users?hl=en
--- 
You received this 

Re: [otb-users] Re: ANN produce not stable results

[Guillaume Pasero]

  
  
Hi, 

I think it is possible to get outputs from libSVM during processing.
When I use this library, I get messages like :
optimization finished, #iter = 113
  nu = 0.038199
  obj = -22.175591, rho = 0.331874
  nSV = 38, nBSV = 28

 Do you get these messages when running the command line application
?

Guillaume

On 02/03/2016 11:20 AM, Hristo Nikolov
  wrote:


  Dear Guillaume,
Thank you for your suggestion. As I pointed above I use the
  GUI version of OTB.
What I shall do is to follow your advice and move to CLI.


Just a small addition - the libSVM runs for more than 7
  days and still not finished.
Is it possible to integrate some parameter that shows the
  convergence when
running the learning phase of the modules in this library? 
By now the only way for me is to check the amount of memory
  used by this
process which is not a good estimate for this purpose.


Hristo

  On Friday, 29 January 2016 10:59:44 UTC+2, Guillaume Pasero
  wrote:
  
 Hi,
  
  In order to have stable results you should use the seed
  parameter, so that the Random Number Generator used during
  sample selection always gives the same samples : in the
  otbcli , it is the parameter key '-rand'.
  
  For the ANN, with the parameter "-classifier.ann.sizes",
  you can give multiple sizes to have several layers.
  
  HTH,
  
  Guillaume
  
  On 01/18/2016 05:41 PM, Hristo Nikolov wrote:
  
  
One more observation - when using
  polynomial as kernel function I can not set the degree
  of the polynomial.
  
  On Wednesday, 13 January 2016 15:40:40 UTC+2, Hristo
  Nikolov wrote:
  
Dear all,
  I have been experimenting with ANN provided
from OTB for some months now. 
  What I experienced is that as module in QGIS
the ANN from OTB has limitation
  of creating only single layer networks, but
the results produced remain the same
  under same initial parameters. Lastly I tried
to the otbgui version provided by 
  OSGeo4W, but it results that even the
parameters in the dialog box are the 
  same the output produced and more
specifically the error matrix is different 
  every time the training takes place.
  Could you advise?
  
  
  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 otb-...@googlegroups.com
To unsubscribe from this group, send email to
otb-users+...@googlegroups.com
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 otb-users+...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
  
  
  -- 

  


Guillaume
PASERO
  Responsable technique
  Business Unit E-SPACE & Geo Information
- Département Image & Applications
  
  CS Systèmes d'Information
  Parc de la Grande Plaine - 5, Rue Brindejonc
  des Moulinais - BP 15872
  31506 Toulouse Cedex 05 - FRANCE
  +33 561 17 64 21 - guillaum...@c-s.fr 

  

  

  

  
  -- 
  -- 
  Check the OTB FAQ at

Re: [otb-users] Re: ANN produce not stable results

[Hristo Nikolov]

Dear Guillaume,
Thank you for your suggestion. As I pointed above I use the GUI version of 
OTB.
What I shall do is to follow your advice and move to CLI.

Just a small addition - the libSVM runs for more than 7 days and still not 
finished.
Is it possible to integrate some parameter that shows the convergence when
running the learning phase of the modules in this library? 
By now the only way for me is to check the amount of memory used by this
process which is not a good estimate for this purpose.

Hristo

On Friday, 29 January 2016 10:59:44 UTC+2, Guillaume Pasero wrote:
>
> Hi,
>
> In order to have stable results you should use the seed parameter, so that 
> the Random Number Generator used during sample selection always gives the 
> same samples : in the otbcli , it is the parameter key '-rand'.
>
> For the ANN, with the parameter "-classifier.ann.sizes", you can give 
> multiple sizes to have several layers.
>
> HTH,
>
> Guillaume
>
> On 01/18/2016 05:41 PM, Hristo Nikolov wrote:
>
> One more observation - when using polynomial as kernel function I can not 
> set the degree of the polynomial.
>
> On Wednesday, 13 January 2016 15:40:40 UTC+2, Hristo Nikolov wrote: 
>>
>> Dear all, 
>> I have been experimenting with ANN provided from OTB for some months now. 
>> What I experienced is that as module in QGIS the ANN from OTB has 
>> limitation
>> of creating only single layer networks, but the results produced remain 
>> the same
>> under same initial parameters. Lastly I tried to the otbgui version 
>> provided by 
>> OSGeo4W, but it results that even the parameters in the dialog box are 
>> the 
>> same the output produced and more specifically the error matrix is 
>> different 
>> every time the training takes place.
>> Could you advise?
>>
>> 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 otb-...@googlegroups.com 
> 
> To unsubscribe from this group, send email to
> otb-users+...@googlegroups.com 
> 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 otb-users+...@googlegroups.com .
> For more options, visit https://groups.google.com/d/optout.
>
>
> -- 
>  *Guillaume PASERO*
> Responsable technique
> *Business Unit E-SPACE & Geo Information - Département Image & 
> Applications*
>
> *CS Systèmes d'Information*
> Parc de la Grande Plaine - 5, Rue Brindejonc des Moulinais - BP 15872
> 31506 Toulouse Cedex 05 - FRANCE
> +33 561 17 64 21 - guillaum...@c-s.fr  
>

-- 
-- 
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 otb-users@googlegroups.com
To unsubscribe from this group, send email to
otb-users+unsubscr...@googlegroups.com
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 otb-users+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Re: [otb-users] Re: ANN produce not stable results

[Guillaume Pasero]

  
  
Hi,

In order to have stable results you should use the seed parameter,
so that the Random Number Generator used during sample selection
always gives the same samples : in the otbcli , it is the parameter
key '-rand'.

For the ANN, with the parameter "-classifier.ann.sizes", you can
give multiple sizes to have several layers.

HTH,

Guillaume

On 01/18/2016 05:41 PM, Hristo Nikolov
  wrote:


  One more observation - when using polynomial as
kernel function I can not set the degree of the polynomial.

On Wednesday, 13 January 2016 15:40:40 UTC+2, Hristo Nikolov
wrote:

  Dear all,
I have been experimenting with ANN provided from OTB
  for some months now. 
What I experienced is that as module in QGIS the ANN
  from OTB has limitation
of creating only single layer networks, but the results
  produced remain the same
under same initial parameters. Lastly I tried to the
  otbgui version provided by 
OSGeo4W, but it results that even the parameters in the
  dialog box are the 
same the output produced and more specifically the
  error matrix is different 
every time the training takes place.
Could you advise?


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 otb-users@googlegroups.com
  To unsubscribe from this group, send email to
  otb-users+unsubscr...@googlegroups.com
  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 otb-users+unsubscr...@googlegroups.com.
  For more options, visit https://groups.google.com/d/optout.


-- 
  

  
  
  Guillaume PASERO
Responsable technique
Business Unit E-SPACE & Geo Information -
  Département Image & Applications

CS Systèmes d'Information
Parc de la Grande Plaine - 5, Rue Brindejonc des
Moulinais - BP 15872
31506 Toulouse Cedex 05 - FRANCE
+33 561 17 64 21 - guillaume.pas...@c-s.fr
  
  

  

  




-- 
-- 
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 otb-users@googlegroups.com
To unsubscribe from this group, send email to
otb-users+unsubscr...@googlegroups.com
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 otb-users+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[otb-users] Re: ANN produce not stable results

[Hristo Nikolov]

One more observation - when using polynomial as kernel function I can not 
set the degree of the polynomial.

On Wednesday, 13 January 2016 15:40:40 UTC+2, Hristo Nikolov wrote:
>
> Dear all,
> I have been experimenting with ANN provided from OTB for some months now. 
> What I experienced is that as module in QGIS the ANN from OTB has 
> limitation
> of creating only single layer networks, but the results produced remain 
> the same
> under same initial parameters. Lastly I tried to the otbgui version 
> provided by 
> OSGeo4W, but it results that even the parameters in the dialog box are the 
> same the output produced and more specifically the error matrix is 
> different 
> every time the training takes place.
> Could you advise?
>
> 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 otb-users@googlegroups.com
To unsubscribe from this group, send email to
otb-users+unsubscr...@googlegroups.com
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 otb-users+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[otb-users] Re: ANN produce not stable results

[Hristo Nikolov]

I'm wondering if I'm the only one to experiment with ANN' provided in 
OTB?
I still keep receiving different results when using those methods.
It is interesting to notice that I'v provided max values for training and
validation samples, but in the experiments both are different every 
run. Perhaps this is why the results does not coincide.

Any opinion or suggestion will be appreciated.

Hristo

On Wednesday, 13 January 2016 15:40:40 UTC+2, Hristo Nikolov wrote:
>
> Dear all,
> I have been experimenting with ANN provided from OTB for some months now. 
> What I experienced is that as module in QGIS the ANN from OTB has 
> limitation
> of creating only single layer networks, but the results produced remain 
> the same
> under same initial parameters. Lastly I tried to the otbgui version 
> provided by 
> OSGeo4W, but it results that even the parameters in the dialog box are the 
> same the output produced and more specifically the error matrix is 
> different 
> every time the training takes place.
> Could you advise?
>
> 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 otb-users@googlegroups.com
To unsubscribe from this group, send email to
otb-users+unsubscr...@googlegroups.com
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 otb-users+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[otb-users] Re: ANN produce not stable results

[Hristo Nikolov]

An addition from today - the same goes for libSVM method. 
No changes in the data & params 3 different results.



On Wednesday, 13 January 2016 15:40:40 UTC+2, Hristo Nikolov wrote:
>
> Dear all,
> I have been experimenting with ANN provided from OTB for some months now. 
> What I experienced is that as module in QGIS the ANN from OTB has 
> limitation
> of creating only single layer networks, but the results produced remain 
> the same
> under same initial parameters. Lastly I tried to the otbgui version 
> provided by 
> OSGeo4W, but it results that even the parameters in the dialog box are the 
> same the output produced and more specifically the error matrix is 
> different 
> every time the training takes place.
> Could you advise?
>
> 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 otb-users@googlegroups.com
To unsubscribe from this group, send email to
otb-users+unsubscr...@googlegroups.com
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 otb-users+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[otb-users] Re: ANN produce not stable results

[Hristo Nikolov]

Dear all,
I have been experimenting with ANN provided from OTB for some months now. 
What I experienced is that as module in QGIS the ANN from OTB has limitation
of creating only single layer networks, but the results produced remain the 
same
under same initial parameters. Lastly I tried to the otbgui version 
provided by 
OSGeo4W, but it results that even the parameters in the dialog box are the 
same the output produced and more specifically the error matrix is 
different 
every time the training takes place.
Could you advise?

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 otb-users@googlegroups.com
To unsubscribe from this group, send email to
otb-users+unsubscr...@googlegroups.com
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 otb-users+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.