[SPARK-11465] Support multiple eigenvectors in power iteration clustering

• Type: New Feature
• Status: Resolved
• Priority: Major
• Resolution: Incomplete
• Component/s: MLlib
• Labels: bulk-closed
• Assignee: Unassigned
• Reporter: Xiangrui Meng
• Created: 02/Nov/15 23:16
• Updated: 21/May/19 04:35
• Resolved: 21/May/19 04:35

Description

I created this JIRA to collect use cases of using more than 1 (pseudo)-eigenvectors in power iteration clustering, in order to decide whether we want to add this feature. Some related work:

• Deflation-based power iteration clustering: http://www.cs.ucsb.edu/~veronika/MAE/deflation-PIC.pdf
• Spectral Clustering via the Power Method – Provably: http://arxiv.org/abs/1311.2854

If you are interested in this feature, please describe your use cases in comments and why using a single eigenvector is not sufficient.

Issue Links

• relates to: SPARK-4259 Add Power Iteration Clustering Algorithm with Gaussian Similarity Function
• relates to: SPARK-6441 Add Deflation/Schur Complement to Power Iteration Clustering for improved resilience to inter-class collisions