Representation of parameter batching in OPIL

[jgladwig]

I have been converting the strateos experiment params from being output as json to instead be output as OPIL. The previous json representation would output parameters in 'batches' where each batch represented some grouping that should be executed together due to some shared parameter value within that batch (e.g. identical media or identical temperature). My understanding is that part of the goal here is to ensure that one batch is processed in its entirety before the next batch is processed.

@danbryce can provide an extended description of what the batches represent and how they are currently used.

The end goal for this issue is to determine how to best represent this batching behavior within OPIL.

[danbryce]

XPlan produces an experiment design where each measurement (row in the experiment design table) has a batch id. The batch id corresponds to a single invocation at a lab. We can express the measurements as a table and add a column for batch_id or somehow group the measurements by batch (e.g., have a table for each batch).

[jakebeal]

The model here appears to be that Intent Parser is giving one ExperimentRequest object, which then XPlan then splits into several ExperimentRequest documents. The split is guided by hints provided by the user in a column.

A near-term kludge that works for this is having a batch column, which goes as a custom marker.

A better longer-term approach will be to have Intent Parser able to export multiple experiment requests, but we aren't there yet.

[danbryce]

Roger that.

…

On Feb 24, 2021, at 12:54 PM, Jacob Beal ***@***.***> wrote: The model here appears to be that Intent Parser is giving one ExperimentRequest object, which then XPlan then splits into several ExperimentRequest documents. The split is guided by hints provided by the user in a column. A near-term kludge that works for this is having a batch column, which goes as a custom marker. A better longer-term approach will be to have Intent Parser able to export multiple experiment requests, but we aren't there yet. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.