The Strategy Learner is unreliable at predicting future events

[TommyHolmberg]

The evaluation is run with the same values as the training set. The Strategy learner has only learned what happened during the time 2008-01-01 - 2009-12-31. Look in the expriment1.py at lines 79-87 to see what I mean.

I tried the Strategy Learner to see if could predict the future, so I trained it with data from 2005-2008 and let it evaluate over 2009-2010. Unfortunately I get very random results. Some times it performs better than the Manual strategy, but just as often it performs less well. It mostly performs better than the benchmark, but I wouldn't call it reliable. Maybe it can be configured better, but I haven't found any reliable settings for it.

Do you have any recommendations for amount of bags and leafs, or any other configuration setting that I may be missing?