search

nsenno-dbr / many-model-forecasting

Bootstrap your large scale forecasting solution on Databricks with Many Models Forecasting (MMF) Project.
Other
0 stars 0 forks source link

Consider making registering to model registry configurable #7

Open nsenno-dbr opened 1 month ago

nsenno-dbr commented 1 month ago

Currently the solution creates a new registered model every time it is run. This can get confusing when running many cycles in development. It would be better if we just saved the model in the experiment then push to the registry when we were happy with the solution

ryuta-yoshimatsu commented 1 month ago

Reflecting on this issue, I'm thinking to add a parameter "register" or "dev" to the run_forecast() function that prevents MMF from registering the model. If we take the "dev" route, we can switch off bunch of other stuffs that are not really important during the iteration: e.g. scoring_output table.