LDA Coefficients meaning

Concerning processing components: filters, file load/save, visualizations, communication ...
Post Reply
benry
Posts: 12
Joined: Thu Sep 22, 2011 3:59 pm
Location: Milan, Italy
Contact:

LDA Coefficients meaning

Post by benry »

Hi all,
I am using the classifier processor box to classify simple bidimensional features... since they are in 2D I would like, given the LDA coefficients which are in the configuration file generated by the trainer, to plot the line used to divide the two classes (off-line, using an external application).
I noticed that the coefficients are always one more than the features dimensions... I am not a machine learning expert at all, thus my question may be very stupid, but how can I plot the line dividing the classes given the coefficients?
I read that the "dimension+1"-th coefficient, may be the bias factor... but in the configuration file it is the first or the last one? And furthermore, is it actually the bias factor?

Any information on the interpretation of the coefficients saved by the LDA trainer would be kindly appreciated.


Thanks in advance,

Enrico

jtlindgren
Posts: 775
Joined: Tue Dec 04, 2012 3:53 pm
Location: INRIA Rennes, FRANCE

Re: LDA Coefficients meaning

Post by jtlindgren »

Hello Enrico,

the extra coefficient is the bias term, as you suggest. Moreover, it should be the first coefficient of the sequence, based on casual inspection of ovpCAlgorithmClassifierLDA.cpp (see functions train() and classify()).

Happy hacking,
Jussi

Post Reply