Suppose that we have N signals for subjects. When we decompose and extract features from them, the length of feature vector for each of them is not identical. This is due to decomposition algorithm.
Can anyone suggest algorithms for classification of data with non-identical feature vectors?
hello friend
for such kind of problems, you can use statistical techniques like SVm and similar methods.
You can achieve good results.
I hope it will help you.
Best regards,
Dr.Indrajit Mandal
A simple method to deal with vectors of different length would be to pad the vectors so that they are all some maximum length. Here, though, one has to be careful to make sure that like dimensions (factors) appear in the same place in each vector. Another method would be to compare only shared dimensions between any pair of vector using some sort of norm (Euclidean norm for example) and then use all such pairwise norms from a given data set to derive whatever comparative metrics you are interested in.
Thank you friends,
I think , svm algorithm needs features with equal dimensions. Is there any extension to svm with variable length features?
One way to do that is to assign the dimension according to the vector that contains the most features. It is possible to be assigned zero values for some elements of the vectors.
If the size of the features is fixed, you might want to consider cross-domain matching. This might be helpful: http://cbl.uh.edu/~pmoutafis/research.html
Thanks all,
All comments was very well. Also, I think Olasimbo's description was more useful for me.
H
Did you find a solution for working with variable length signals?
Thanks
The following paper may also be relevant:
Classification when the covariate vectors have unequal dimensions; Journal of Statistical Planning and Inference 141 (2011) 1944–1957 .
- Badges
- Science topic