I'm looking for a Sparse matrix dimension reduction. I already used some feature selection methods like PCA but it doesn't give me good results. I want to apply mixture models for clustering my data. I appreciate any comment or answer .
- Anybody can ask a question
- Anybody can answer
- The best answers are voted up and rise to the top
You said: "I'm looking for a Sparse matrix dimension reduction." My intuition is that you want to find a solution for e recommendation engine. If it is so, IMHO the best alternative to PCA is Singular Value Decomposition SVD has implementation in domain specific languages Matlab and R , and in many mainstream languages Java, Python
Moreover you said: "I want to apply mixture models for clustering my data." There are many methods data clustering in general, but is very important "what" and "when" (in which step) is clustered. For example you can cluster reduced scores for items or for products or for both. You will need to prototype to find best solution for your case. This is also a good resource.
You may also be interested in the Nyström Method. There are a vast number of papers demonstrating different uses: clustering, Kernel Methods like SVM and also Gaussian Processes. I personally am not familiar with any concrete implementation, but if you will find many of it for different purposes in the Internet