Options
"Display of eigenvectors in Generalized Hebbian Algorithm. Is this a bug?"
Contributor II
Hello,
I'm applying a Generalized Hebian Algorithm (GHA) to compute a PCA on text data. I compute only the few first dimensions (3 to 10, thats the particularity of the algorithm that it can calculate only a certain number of dimensions). To get the eigenvectors, I print the "pre" output of the operator which gives an insight into the eigenvectors and eigenvalues computed by the operator. The problem is that when I ask for the eigenvectors, the model gives me only the first components (exactly the same number of components as the number of dimensions computed) in the term space in alphabetical order (so its not even chose among the importants terms)!
Is this a bug or is it normal (but then I don't really get the meaning of the eigenvectors in this algorithm)
Thanks for your help,
Barthélémy
I'm applying a Generalized Hebian Algorithm (GHA) to compute a PCA on text data. I compute only the few first dimensions (3 to 10, thats the particularity of the algorithm that it can calculate only a certain number of dimensions). To get the eigenvectors, I print the "pre" output of the operator which gives an insight into the eigenvectors and eigenvalues computed by the operator. The problem is that when I ask for the eigenvectors, the model gives me only the first components (exactly the same number of components as the number of dimensions computed) in the term space in alphabetical order (so its not even chose among the importants terms)!
Is this a bug or is it normal (but then I don't really get the meaning of the eigenvectors in this algorithm)
Thanks for your help,
Barthélémy
Tagged:
0
Answers
http://bugs.rapid-i.com
I assume it is only a pretty simple error in calculating the maximum number of rows of the displayed table which accidentally just returns the number of columns as well for the rows...
Cheers,
Ingo
Its a little contribution to the great tool you are developping,
Thanks for all,
Barthélémy
thanks for your kind words
If you want you could do us a favor and support RapidMiner at the currently running poll at KDNuggets. It's only a single click but really helps us a lot. More information can be found here:
http://rapid-i.com/rapidforum/index.php/topic,3677.msg13710.html
Thanks :-*
Cheers,
Ingo
Best wisches,
Barthélémy
just as a side note: I have fixed this bug.
Greetings,
Sebastian
And how can I get the fixed version ?
Thanks,
Barthélémy
there are two ways: Buy an enterprise edition. The Hotfixing guarantee ensures that we would send you a fixed version immediately after the fix.
the second way: Checkout the source code and build RapidMiner yourself. A tutorial how to to this is on our website. If you understandably don't want to go this tedious way and don't want to pay, you simply will have to wait for the next release. Sorry.
With kind regards,
Sebastian
Barthélémy