Undergraduate Project, Autumn 2016
Vikas Jain
Under the guidance of Prof Piyush Rai
Department of Computer Science and Engineering, IIT Kanpur, India

##Project Idea To develop a probabilistic model using mixture of factor analyzers for extreme multi-label learning. The MFA model used is adapted from the work by Montrari et al. "Dimensionally reduced mixtures of regression models" (paper link).
The purpose of the project is to scale the learning of the model and incorporating 0-1 labelling instead of continuous values. To achieve this, online EM algorithm and EM algorithm for binary logistic regression are used based on the work by James G. Scott et al. "Expectation-maximization for logistic regression" (paper link).

##Code Source From https://cran.r-project.org/web/packages/FactMixtAnalysis/

##Modifications ####Online EM Added a new function onlinefma() for online EM algorithm.
How to use:

fit = onlinefma(y,k,r,x.z=NULL,x.w=NULL,it=1,eps=0.0001,seed=4,scaling=FALSE,init=NULL,no_iter=100,batch_size=1000)

Refer to the document here for details of the variables.

TODO

##Experiments TODO