-
Notifications
You must be signed in to change notification settings - Fork 26
Parameter Importance
arturluis edited this page Oct 9, 2020
·
3 revisions
It is possible to see the importance given to each input parameter during optimization. This can be enabled in the json scenario file with the following line:
"print_parameter_importance": true,
When this flag is enabled, HyperMapper will print the importance of each input parameter, according to the surrogate model. You can find this line in the hypermapper_logfile.log file. For example, for the Branin function, the line printed is:
Regression model on Value. Features names: ['x1', 'x2'], feature importances: [0.49627725 0.50372275]
HyperMapper will print this line every time the model is trained, i.e., at the start of each optimization iteration. So, it is also possible to see how the importances shift as the space is explored and the model is refined.
Note: this feature only works with the default Random Forests model.