Define $D_{\mathtt{subset}}\leftarrow \mathtt{Sample}(D,p)$

Define $D_{\mathtt{train}},D_{\mathtt{val}}\leftarrow \mathtt{Split}(D_{\mathtt{subset}})$

Define search space for hyperparameters θ

Define objective function $f(\theta )$ using validation accuracy

Initialize $E\leftarrow E_{\mathtt{init}}$, $T\leftarrow 0$

while $T<n$ do

Train model with hyperparameters θ for E epochs

Implement early stopping based on

Update hyperparameters θ using Bayesian optimization

Update $E_{\mathtt{opt}}$ using the best validation accuracy so far

$T\leftarrow T+1$

end while

Train with optimal hyperparameters ${\theta }^{\ast }$ for $E_{\mathtt{opt}}$ epochs

Save the model with the best hyperparameters ${\theta }^{\ast }$