-
Notifications
You must be signed in to change notification settings - Fork 103
Results
For genetic algorithm -g, results are visible at the end of last epoch, as dict and Gekko UI-friendly format (TOML).
Results can also be visible in middle epoches.. check generations.evaluateSettingsPeriodically @ Settings.py to set that interval.
Logs are tab separated values files with several columns.
The evaluation column corresponds to the best score attained in the evaluation dataset, which is the one used to evolve the model.
The secondary column is the best score attained by the strategy in the secondary dataset, the one which is not used to evolve the model and works as an independent benchmark.
japonicus should output many different sets of viable parameters, probably a few per generations.evaluateSettingsPeriodically.
So, which one to use? No simple answer, but you'll probably want the ones from later epoches. Also, check the number of trades and sharpe ratio (on the log file) and select you preferred market 'stance' from that.
Maybe the same parameters will be shown on the web UI or in the logs more than once at different epoches. This can be interpreted as a signal of late game fitness and maybe you should abort the GA run and try those parameters live. With more complex strategies (more parameters) this will take longer to happen than with let's say a strat of 4 parameters.