[Topic-models] Regarding LDA-C

Veena T veenat2005 at gmail.com
Mon Nov 24 04:10:26 EST 2008


Dear all,
I am using the LDA-C code. The code has an option for estimating the alpha
parameters. When I use this part for my purpose, the Newtons method is not
converging at all. What happens is that, in the 8th iteration itself the
alpha value is getting reduced to zero.  So the algorithm ends up in NANs.
Whether the initial value of alpha, has any role to play?
I tried to understand the code there and I have a small doubt about the
expression log_a = log_a - df/(d2f * a + df); I can understand that, this is
in log scale. however if we directly compute this, the formula is
a=a-df/d2f. I am not able to obtain the log scale equivalent of this. If
anyone has it, please give me.
regards
veena

-- 
Veena Srinivas,
PhD scholar,
Speech and Vision Lab
IIT Madras, Chennai-36
Tamil Nadu
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/topic-models/attachments/20081124/716148f6/attachment.html>


More information about the Topic-models mailing list