val_accuracy objective goes for very high epoch count with very small (sub 0.001) incremental improvement.

[istavnit]

During the search for network architecture Automodel with objective of val_accuracy gets stuck (see picture). I imagine that happens because min_delta on objective is set to some very small number. Is it possible to provide the min_delta to Automodel so it looks for a minimum amount of increment to keep going?

[haifeng-jin]

Not sure if you are using 1.0.9.
But this line would stop the training if it doesn't decrease enought (>=0.0001) after 10 epochs.