Permalink
Fetching contributors…
Cannot retrieve contributors at this time
21 lines (13 sloc) 508 Bytes
Method Improvement
-- run the same samples on a series of models before them,
and include the tuples in overall ranking, since we allow
for distinct date in (person,date,perps)
-- try ngrams of different orders
-- try different smoothing methods
-- predict
for a prefix, generate possible suffixes, and measure the end cloud
-- change
for two LMs, compare predicted clouds
or measure KL divergence
-- sample.ml
add an option to match an existing sample set, instead of generating a new one