This message was deleted.

Very cool! I’m just experimenting with Hamilton, and very much liking the readability improvements it’s making to the code base I’m working on. It would appear that Ray has a good amount of overhead, and probably isn’t worth it for executions operations that are less than a minute or so… Does that seem right? When timing local execution on my 2020 MBP

dr = driver.Driver(…)
 dr.execute(output_columns)

Without Ray: 3.76 seconds With Ray: 30.02 seconds

yep, also data set sizes matter a bit too and how often some common data is used — Hamilton isn’t for instance telling Ray what should be in the object store as an optimization — but we’re looking for feedback here; if you have some code to share, I could make an issue to look at why it’s that slow 🙂