I'm currently working on signal processing, where i've to compare signals for different lengths. Please note that, i shouldn't do zero padding or interpolation.
For eg., 1000s signal will produce 10 windows, 10 features (for each window)
2000s signal will produce 20 windows, 20 features (for each window)
I've to transform this different length feature vectors to a feature space where i can do classification? How to do this for comparing different length signals? Please note that, i shouldn't use Nan values too.
Hi Arunraj,
You could scale the values of the attributes using the Euclidean length (or L2 norm).
HTH.
Samer, PhD.
Hi,
you have a time series problem. In time series problem, each observation or sample has different length, but your classifier must be designed for this problem type such as HMM, GMM.
regards
If You need to find a similarity between two vectors of different lenght, consider using DTW (Dynamic Time Warping)
- Badges
- Science topic
- Similar topics
- Data Mining
- Classification