Study what is the role of docConcentration parameter, whether it is completely equivalent to alpha in mallet or not. A second issue. Does it pay off to allow the user to set the a priori distribution for topics (eta)