search

dssg / triage

General Purpose Risk Modeling and Prediction Toolkit for Policy and Social Good Problems
Other
187 stars 61 forks source link

Should timechop have a cron-like interface? #443

Open nanounanue opened 6 years ago

nanounanue commented 6 years ago

If we plan to share triage with the partners, the current implementation of timechop recreates all the matrices (because you are moving the last day) everytime that you update the experiment's temporal configuration.

If the partner (or ourselves) want to run the experiment "the 5th of everymonth" timechop should take that information don't create new matrices and just train with the new data. Also this interface potentially is more easy to explain that the actual setup.

But what are your ideas?

thcrock commented 6 years ago

I'd like to understand more about the 'last day'. Are you referring to the 'label end time' in timechop parlance?

thcrock commented 6 years ago

What exactly would we do with the new date ('5th of every month')? What time range should it train on? I interpret the given date as the 'train end date'...what is the train start date? Are we always just creating one train matrix in these scenarios? In other words, how much of the experiment's original time config is this new operation subject to?