rwalk / gsdmm-rust Goto Github PK

@rwalk Dear Ryan, Thanks so much for sharing this code!

For my runs there seems to be a little deviation from the behavior described in the docs: The file examples/grades/grades_out_label_probabilities.csv is not produced. (The others mentioned are and everything seems to work fine.)

(Also, I suspect examples/grades_out_labels.csv is a typo and should be examples/grades/grades_out_labels.csv?)

Cheers,
Leo

The Yin and Wang 2014 paper also specifies an implementation with non-binary term counts. The formulas are a bit more difficult and the benefit is uncertain so I've not implemented it in the first pass.

The constructor for GSDMM with signature:

pub fn new(alpha:f64, beta:f64, K: usize, maxit:isize, vocab:HashSet<String>, docs:Vec<Vec<String>>) -> GSDMM

consumes the docs and vocab objects. This means that user will need to make copies of the data if they wish to use either of these vectors in subsequent processing.

Should be simple to refactor the interface to borrow these instead.