🐛 Bug

ReduceLROnPlateau will fail when add new parameter group to the optimizer.
The _reduce_lr function will raise list index out of range error

To Reproduce

Steps to reproduce the behavior:

1. initialize optimizer: optimizer = Adam(filter(lambda p: p.requires_grad, model.parameters()), args.lr)
2. initialize scheduler: scheduler = ReduceLROnPlateau(optimizer, patience=1, factor=0.1, verbose=True, mode='max')
3. add parameter group to the optimizer: optimizer.add_param_group({'params':unfreezed_params, 'lr':lr})
4. raise error when the learning rate should be changed

Expected behavior

when add a parameter group to the optimizer, the min_lrs attribute of the scheduler should be updated to avoid this error.

Environment

• PyTorch Version (1.1.0):