Evaluation using other trial's model parameters

Ray Libraries (Data, Train, Tune, Serve) Ray Tune
1

How severe does this issue affect your experience of using Ray?

  • None: Just asking a question out of curiosity

Current ray tune supports to report the metric by each trial. However, if I want to acquire the model parameters from other trials to help current trial’s evaluation, any suggestions to implement this?

2

Hi! Can you elaborate a bit how this would exactly look like? I.e. what do you mean by model parameters etc.

Ray Tune can use search algorithms (e.g. Bayesopt, or Tree-Parzen estimators (hyperopt/optuna)) to inform the search for new configurations. Here, the search algorithm learns from the results from previous trials to generate configurations for future trials. Is this what you mean?

3

Thanks for your reply. Yes, what i want to do is as follows.

RayTune manage three trials, trialA, trialB, trialC.
When I want to evaluate a model performance in trial_A. I want to notify trialB and trialC to send their DNN’s model parameters to trialA. Because I want to reconstruct my expected DNN’s parameters using all trial’s DNN’s parameters. It has nothing with searching configurations.