How to automatically change Learning Rate if loss function is not improved?

[cemox35]

Hello there,

This is my code:

loss = PITLossWrapper(singlesrc_neg_sisdr, pit_from="pw_pt")
optimizer = optim.Adam(model.parameters(), lr=1e-3)
system = System(model, optimizer, loss, train_loader, val_loader)
trainer = Trainer(max_epochs=EPOCHS,
                          resume_from_checkpoint=MODEL_PATH[i],
                          gpus=1,
                          precision=16,
                          callbacks=[EarlyStopping('val_loss')])
trainer.fit(system)

How can I change the LR if the output of the loss function in the validation set is not improved in three consecutive epochs?

[mpariente]

Have a look at the recipes, most of them have a learning rate scheduler set up.