Greetings dear community,
I am using AutoML for the first time ands was wondering whether it is possible to have early stopping or incorporate any approach in my code to make the training of a model stop when the performance plateaus. Early stopping is something one can implement in the traditional way of training models (without auto ML). Additionally tracking loss function, performance evolution, etc...
I would be interested to have you thoughts on this since I am doing a client demo in the coming days.
# Run AutoML with manual's data split (0.8/0.2)
automl_result_manual_split = automl.classify(
dataset=train_df,
target_col="cae_type",
primary_metric="f1",
timeout_minutes=30,
experiment_dir=f"{group_workspace_base}/manual_split",
experiment_name=experiment_auto_ml_manual_split
)
2) My second question is regarding data split. As you can see here, I did a manual split (0.8 training data/ 0.2 testing data) but I am aware that data splitting can be done automatically by AutoML. Are there any resources that recommend the one or the other? (I also have class imbalance but I did not consider this in this first demo trial)
Best regards
