Hello, community,

I hope this message finds you all well. I am currently working on a project involving Hyperparameter Optimization (HPO) using the Optuna optimizer. Specifically, I've been trying to navigate the parameters 'min_iteration_per_job' and 'max_iteration_per_job' which, as I've come to understand, are typically tied to the number of batch iterations.

My conundrum is that my training regimen is structured around epochs rather than batches. To complicate matters, I'm looking to optimize the batch size as part of the HPO process. However, with a shifting batch size, determining a fixed number of batch iterations becomes problematic. For instance, I initially set 'min_iteration_per_job' to 50 and 'max_iteration_per_job' to 250. Despite these settings, I've observed Optuna terminating tasks after only 10 epochs. This behavior led me to believe that Optuna might be interpreting each batch, rather than each epoch, as a single iteration.

I am reaching out to seek advice or solutions from anyone who may have faced a similar challenge. Is there a way to configure Optuna to perform HPO with an emphasis on epochs? Or is there another approach to integrating batch size adjustment as a variable parameter without conflicting with the established iteration parameters?

Any insights or guidance on this matter would be greatly appreciated.