elicit.optimization module

elicit.optimization.sgd_training(expert_elicited_statistics, prior_model_init, trainer, optimizer, model, targets)

Wrapper that runs the optimization algorithms for E epochs.

Parameters:

expert_elicited_statisticsdict

expert data or simulated data representing a prespecified ground truth.

prior_model_initclass instance

instance of a class that initializes and samples from the prior distributions.

one_forward_simulationcallable

one forward simulation cycle including: sampling from priors, simulating model predictions, computing target quantities and elicited statistics.

compute_losscallable

sub-dag to compute the loss value including: compute loss components of model simulations and expert data, compute loss per component, compute total loss.

global_dictdict

dictionary including all user-input settings.