v6.0.0 Machine Learning

With version 6 tuo gets first machine learning support. The machine learned model tries to extend the entries from database.yml to decks that have not been actually simulated. Currently you can find a ml/train.py file in the tuo sourcecode to train your own xgboost-tree. Check out the Readme in the ml/Readme.md it is easy! (Feel free to change the code in there to use other models that are pmml compatible).

TUO expects the files data/{win,loss,stall,points}.pmml as inputs to use a trained model. You can find some general non-optimized once in the source code here https://github.com/APN-Pucky/tyrant_optimize/tree/master/data.

Options include:

• ml TUO will use the machine learned prediction only if it seems reliable, otherwise a normal simulation is performed.
• boost-ml Runs the asked operation once in the fast ml-mode then in normal mode without ml. This is equivalent to using a parameter file with -p ml.params, where ml.params is

noop
only-ml
deck @1@ no-ml

• ml-precision D sets the approximate required precision of the machine learned result (default 0.01 meaning 1%).
• only-ml TUO will only use the machine learned predictions. This is more of a debug/curiosity option since it is almost impossible to setup machine learning to find a global optimum. It is rather a fit, covering a broad range of decks.
• no-ml TUO won't use machine learned predictions

Enabling either of the ml options will disable saving to the database, since the results are not to be trusted 100%.

Note:

• noop was introduced to do exactly nothing, so it is a space holder useless parameter