Automatic Model Selection in Archetype Analysis

摘要: Archetype analysis involves the identification of representative objects from amongst a set multivariate data such that can be expressed as convex combination these objects. Existing methods for archetype assume fixed number archetypes priori. Multiple runs different choices are required model selection. Not only is this computationally infeasible larger datasets, in heavy-noise settings selection becomes cumbersome. In paper, we present novel extension to existing with specific focus relaxing need provide beforehand. Our fast iterative optimization algorithm devised automatically select right using BIC scores and easily scaled noisy, large datasets. These benefits achieved by introducing Group-Lasso component popular sparse linear regression. The usefulness approach demonstrated through simulations on real world application document identifying topics.