A hyperopt integration would be nice. I would gladly see a cross_validate.py file, just like the train.py but with cross validation. No duplication of code however: we FACTORIZE !!
- Do we keep the
raw rewards function as a metric ?
- IMPORTANT: The smaller the network (n_hidden layer), the faster the computations, and since we care about performance, we should select the network as small as possible - by cross-validating
A hyperopt integration would be nice. I would gladly see a cross_validate.py file, just like the train.py but with cross validation. No duplication of code however: we FACTORIZE !!
raw rewardsfunction as a metric ?