Calculators#

A calculator is the central agent that exchanges information between a model and the minimizer.

It uses the computation methods provided by a model to calculate the energy, forces, etc. and pass these properties, together with the corresponding reference data, to Loss to construct a loss function to be minimized by the optimizer.
It also inquires the model to get parameters that are going to be optimized, and provide these parameters to the optimizer, which will be used as the initial values by the optimizer to carry out the optimization.
In the reverse direction, at each optimization step, the calculator grabs the new parameters from the optimizer and update the model parameters with the new ones. So, in the next minimization step, the loss function will be calculated using the new parameters.

A calculator for the physics-motivated potential can be created by:

from kliff.calculators import Calculator

model = ...  # create a model
configs = ...  # get a list of configurations
calc = Calculator(model)
calc.create(configs, use_energy=True, use_forces=True, use_stress=False)

It creates a calculator for a model (discussed in Models), and configs is a list of Configuration (discussed in Dataset), for which the calculator is going to make predictions. use_energy, use_forces, and use_stress inform the calculator whether energy, forces, and stress will be requested from the calculator. If the potential is to be trained on energy only, it would be better to set use_forces and use_stress to False, which turns off the calculations for forces and stress and thus can speed up the fitting process.

Other methods of the calculator include:

Initialization: get_compute_arguments().
Property calculation using a model: compute(), get_compute_arguments(), compute(), get_energy(), get_forces(), get_stress(), get_prediction(), get_reference().
Optimizing parameters: get_opt_params(), get_opt_params_bounds(), update_model_params().