Model retraining using recipes¶
This page describes how to retrain a model using recipes.
To avoid any extra auto-generated individuals for refit or retrained experiments with a custom individual, set fixed_ensemble_level
and fixed_num_individuals
equal to the desired number of custom individuals in the final model. Also, ensure that enable_genetic_algorithm
is set to off
.
Additionally, you can set the following parameters to prevent the mutation of the genome for this individual (frozen case), in self.params dictionary in Individual Recipe’s set_params
method:
prob_add_genes = 0.0
prob_prune_genes = 0.0
prob_prune_by_features = 0.0
prob_addbest_genes = 0.0
prob_prune_by_top_features = 0.0
Also, uncomment the following code in the Individual Recipe’s set_params
method:
self.config_dict.update(self.config_dict_individual)
self.config_dict.update(self.config_dict_experiment)
Note that the validation scores can have some randomness, but when the parent experiment has a reproducible level set with the relevant reproducible levels, the test scores should match.