High-dimensional neural network¶
SIMPLE-NN use High-Dimensional Neural Network(HDNN)  as a default machine learning model.
Some parameters in neural_network may need to decrease exponentially during the optimization process. In those cases, you can use this format instead of float value. More information can be found in Tensorflow homepage
parameter_name: learning_rate: 1. decay_rate: 0.95 decay_steps: 10000 staircase: false
global_step (see the link above) of save points is also loaded.
Thus, you need to consider the
global_step to calculate the values from
On the contrary,
global_step is reset when
Initiator of Neural_network class.
train(self, user_optimizer=None, aw_modifier=None)¶
user_optimizer: User defined optimizer. Can be set in the script run.py
aw_modifier: scale function for atomic weights.
Method for optimizing neural network potential.
|||J. Behler, M. Parrinello, Phys. Rev. Lett. 98 (2007) 146401|