likelihood bounds for heldout set in time mode are always 0.0

[muellermarkus]

For any given heldout set, the heldout_post.dat file shows a 0.0 likelihood bound for each document. The issue appears to stem from the code referenced below. Instead of saving the previously calculated likelihood variable to the table, the log_likelihood from the doc_t struct is saved, which is initialised to 0.0 in data.c.

dtm/dtm/main.c

Line 277 in 810d9ab

table[t][d] = post.doc->log_likelihood;

To solve this problem, I propose to simply use table[t][d] = likelihood;.