topic AutoML split with dt column not working properly in Machine Learning

AutoML split with dt column not working properly

Noura_azza — Wed, 24 Jan 2024 13:17:53 GMT

I am using AutoML and want to split my data to train/validation and test using a dt column (one date for train one different date for validation and a third date for test. The problem that the autoML fails, there are only training metrics (no valiation nor test ones) and when I check the data exploratory notebook it seems that all samples are considered as training eventhough the corresponding dt are different. When I look to model artifacts, I see that the column dt were taken into consideration as feature by the model

Re: AutoML split with dt column not working properly

Noura_azza — Wed, 24 Jan 2024 13:25:49 GMT

this is what I see in my data exploration notebook. All dates are considered part of the training split

Re: AutoML split with dt column not working properly

maggiewang — Wed, 28 Aug 2024 23:50:38 GMT

Hello! Did you try specify a column name as manual split column?

Then you can fully control which rows are in train / validate / test splits: https://docs.databricks.com/en/machine-learning/automl/automl-data-preparation.html#split-data-for-regression-and-classification