Question about the custom metric

[xpye98]

Hi all,

The site here shows an example of how we can use custom metric for FLAML, and the alpha in def custom_metric seems a weight value for training and validating set.

alpha = 0.5
    return val_loss * (1 + alpha) - alpha * train_loss, {
        "val_loss": val_loss,
        "train_loss": train_loss,
        "pred_time": pred_time,
}

My question is that for default built-in metrics, say, accuracy for binary classification, what is the alpha? Is it zero, which means we only use validation information, or the opposite, we don't use validation loss at all?

What I want to do is to create a f2 score like the built-in f1 score, but I don't know how to decide the alpha as shown in the example.

Thank you very much!

[sonichi]

It's 0.

[xpye98]

Thank you