Embedded von Mises-Fisher Allocation
Author: Honglei Zhuang (hzhuang3@illinois.edu)

This code implements an Embedded von Mises-Fisher Allocation model based on word embedding.  

=======USAGE EXAMPLE=======

java -jar EmbeddedVMFAllocation.jar -corpus nyt.txt -wordembs vectors_sampled.txt [-options]

Other options:
-numvmf [int]       : Number of von Mises-Fisher topics (default=10)
-vmfiter [int]      : Iteration over the entire corpus in Gibbs sampling for model inference  (default=50)


===========OUTPUT==========

The output is stored in models.vmfs

The first line consists of three integers: 
K, representing the number of vMF distributions; 
L, the length of embedded vectors; 
N, the number of documents respectively.

Starting from the line 2 to line (2*K+1), every two lines represent a vMF distribution.  
The first line is a real number, indicating the concentration parameter kappa of the vMF distribution.  
The second line consists of L real number, as a (normalized) vector of the mean direction of the vMF distribution.  

The following N lines represent the allocation probability of each document.  
Each line consists of K real values, which parameterize a categorical (multinomial) distribution of the document.  
Each number represents the probability that a word of this document is selected from the corresponding vMF distribution.