Add helper method to get best HyperparameterTuningJob trial

[zchenyu]

Currently there's no way to get the best trial from a HyperparameterTuningJob besides inspecting each individual trial and implementing minimization or maximization (depending on the study config).

It would be nice if we had a helper method like HPTuningJob.best_trial() or HPTuningJob.best_trials(n)

Or just sorting the trials returned by trials() might be sufficient.

[jaycee-li]

Hi @zchenyu ,

There could be multiple metrics for each trial, so the helper method needs several more fields E.g,HPTuningJob.best_trials(num_trials=10, metric_id='accuracy', goal='maximize')

An alternative approach is to add a HPTuningJob.get_trails_df() method which returns all the trials' ids, training steps, params, and metrics in a pandas DataFrame (like what we show in cloud console UI. In this case users can easily sort/filter by every param or metric.

Which option do your prefer?