ADAM optimizer support for training

[ATirelli]

Hello! I was wondering whether you it would be possible to integrate the ADAM optimiser as an option for the training process. By looking at the code, it looks like the only learning rate schedule available at the moment is Exponential Decay.

Thanks in advance!

[njzjz]

deepmd-kit/source/train/Trainer.py

Line 278 in b2662e2

optimizer = tf.train.AdamOptimizer(learning_rate = self.learning_rate)

The default optimizer is Adam. Exponential Decay is used for the learning rate of Adam. You can see tensorflow API docs https://www.tensorflow.org/api_docs/python/tf/keras/optimizers/Adam.

[ATirelli]

Oh, I see. Thank you!