Can I get more precise answers in the verbose option of fitlda.

Specifically, I want more digits for the training perplexity of each iteration.

Why I need it:

I am trying to calculate the batch mean standard errors of perplexity of fitlda. This error is much smaller than the training perplexity, so I don't get an accurate estimate of standard error.

For details on the equation see:

Jones, Galin L, Murali Haran, Brian S Caffo, and Ronald Neath, "Fixed-width output analysis for Markov chain Monte Carlo," Journal of the American Statistical Association, 2006, 101 (476), 1537-1547.

Answer 1

[Stephen Bruestle]

I discovered the answer myself.

The answer is yes. Instead of relying on the verbose (printed) output, you rely on the model fit statistics. After running fitlda, click "FitInfo", then "History", and then "Perplexity".

To get it as a matrix, use "mdl.FitInfo.History.Perplexity".