This message was deleted.

Hey. The data argument was deprecated in 0.8.1 and removed in 0.11.0, you should now use

df

, as in statsforecast. We raised deprecation warnings when you provided data as a keyword argument, didn't you see them? (Not complaining, just wondering if the warnings may not be shown and whether we should change our approach to deprecate things).

In Databricks I do not see any warning. I updated that and figured out the window_size was made consistent with SF (e.g., now 'h') and it's running. Thanks @José Morales!

Nice! Sorry for the confusion. We changed: • data -> df • horizon -> h • window_size -> h • dynamic_dfs -> X_df In case you run into a similar error

Thank you! I'll look into the built in tranformations and provide any feedback I might have. Also, is there any documentation anywhere on tuning the ML models (xgboost and lgb) when using spark that I might have missed in my searches? I'm still relatively new to python and I can probably figure it out, but if there's documentation that'd be really helpful. Appreciate the quick respones too!

It should be similar to a regular optimization, you just have to return the CV score from the DistributedMLForecast. A good starting point is this example from optuna

@José Morales Following up on this. I finally got back to update my code to try the built-in window_ops transforations. Worked great for me. Very fast.

Thanks. Do you mean the coreforecast ones? The ones in mlforecast.lag_transforms?

Yes, exactly.

Great to hear!