You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The number of generations executed (not planned as in the generations param) is prominently featured in TPOTEstimator's output, but doesn't appear to be accessible via its public API. I am testing progressively longer hyperparameter optimization of several ML models using TPOT2, and it would be useful to get this information directly from public / documented API that I have some assurance will be durable over time.
The best / only way I've found to get this so far is from estimator._evolver_instance.generation. estimator.evaluated_individuals seems to have the first generation each model was introduced, which isn't informative in recording when / why my optimization hit an early stop limit.
The text was updated successfully, but these errors were encountered:
The number of generations executed (not planned as in the
generations
param) is prominently featured inTPOTEstimator's
output, but doesn't appear to be accessible via its public API. I am testing progressively longer hyperparameter optimization of several ML models using TPOT2, and it would be useful to get this information directly from public / documented API that I have some assurance will be durable over time.The best / only way I've found to get this so far is from
estimator._evolver_instance.generation
.estimator.evaluated_individuals
seems to have the first generation each model was introduced, which isn't informative in recording when / why my optimization hit an early stop limit.The text was updated successfully, but these errors were encountered: