Re: DSP - pattern search

[Peter Norlindh]

Thank you Freddy!

 

I shall look into SVM.  I've never heard of it before, but I believe it might be a very interesting thread for me.

 

Thanks again / Peter 

 

 

On Wed, May 11, 2011 at 5:19 PM, Freddy López <address@hidden> wrote:

Hello Peter,

If your need to fit a model to predict or adjust a sequence in
relation to a group of variables (conditions, predictors, etc),
perhaps can be usefull you check LIBSVM library [1] and train several
SVM models; it's a Octave's version.

In relation to your problem, I think that if your sequence present a
clear cycles, you can explore to include into your predictor dataset
variables related with time, such as sin, cos, etc. Additionally, if
it presents an autoregressive pattern you must try to include it into
the model (lagged sequence?).

Cheers.

[1]: http://www.csie.ntu.edu.tw/~cjlin/libsvm/

2011/5/11 Peter Norlindh <address@hidden>
>
> Thank you for your most valuable replies!
>
> I am not familiar with wavelets, but I shall look in to it now that I know to look for it.
>
> The Fourier Transform might be an excellent foundation for the search.  However, both the amplitudes and angular frequencies can vary significantly and non-uniformly.  Comparing two characteristic sequencies, one might be longer in total and at the same time contain a peak segment that is shorter than the corresponding peak section in the other sequence.
>
> I know very little about it but I believe the FFT approach requires a clever search algorithm that can handle the fact that the occurances of the characteristic sequence vary in length among other things.
>
> Are there more potentially interesting approaches?
>
> Thanks!
>
> /Peter
>
> 2011/5/11 Jordi Gutiérrez Hermoso <address@hidden>
>>
>> On 11 May 2011 01:14, Marius Roets <address@hidden> wrote:
>>
>> > I couldn't find one, but I managed to get Wavelab (for Matlab) working in
>> > Octave without too much trouble.
>>
>> As I recall, the authors of Wavelab are friendly to Octave. After all,
>> righton the front page they talk about reproducible research and
>> revealing source, which is not a goal you can fully accomplish with
>> Matlab.
>>
>> Anyways, if you had problems with Wavelab, do report them and they
>> might fix them for you, or submit patches to them and they might apply
>> them.
>>
>> - Jordi G. H.
>
>

> _______________________________________________
> Help-octave mailing list
> address@hidden
> https://mailman.cae.wisc.edu/listinfo/help-octave
>



--
«But Gwindor answered: 'The doom lies in yourself, not in your name.'»

JRR Tolkien