What is BigQuery ML Evaluation based on when DATA_SPLIT_METHOD = 'NO_SPLIT'

Question

Context: creating (training) a model in BigQuery ML (logistic regression, to be more precise).

Problem: A default value for DATA_SPLIT_METHOD is 'AUTO_SPLIT'. In this case, I understand what results are presented in the Evaluation tab in the BigQuery UI. However, even with the value set to 'NO_SPLIT', I still get the Evaluation results, as indicated in the screenshot below:

Could anyone confirm whether this is the evaluation on the training set, or...? I'm not sure what else it could be.

Thanks in advance for any suggestions!

Answer 1

With NO_SPLIT, both training and the evaluation numbers are based on the entire training set.

Note that this means the evaluation number may be overly optimistic, as the model can just 'memorize' what it saw during training.