the difference of trials in parallel search workflow

[Phu Ha-Van]

Hi everyone,

I am working with parallel search workflow on Hyper-opt. I see there are two options for this purpose (MongoDB and Apache Spark). As far as I understand about the mechanism of parallelizing tasks in Hyper-opt is to create a number of trials (this number equals to the number of workers) and pass those trials to available workers for evaluation. A trial corresponds to one setting of hyperparameters which is suggested based on current model built by pass trials.

So, my question is what is the difference among trials passed to workers? Are they generated by different models?

Thanks

Phu