a CMA-ES based hyperparameter optimization tool.

Prerequisite

pip install cma

basic usage:

./run.sh 0

evo_single.py : load initial hyperparameter or previous checkpoint, generate genes for current generation.

gene2conf.py: load genes for current generation, convert them to model description file.

reporter.py: load the evalution result file (e.g. BLEU score) output by each model. report them in a file.

toy_nmt.py: Script to generate random score as a fake evaluation score for testing. (format from sockeye-recipe)

Copyright (c) 2018 Shinozaki-lab, Tokyo Institute of Technology.