Skip to content

deep_macrofin.pde_model_time_step

This implements the time stepping scheme with neural network. It shares the same abstract base class BasePDEModel as PDEModel, so it has exactly the same variable/equation/constraint definition API; only the training loop and a few sampling helpers differ.

PDEModelTimeStep

class PDEModelTimeStep(BasePDEModel):
'''
PDEModelTimeStep uses time stepping scheme + neural network to solve for optimality

PDEModel class to assign variables, equations & constraints, etc.

Also initialize the neural network architectures for each agent/endogenous variables 
with some config dictionary.
'''

The time-stepping config additionally accepts the experimental stacked flag (default False), which batches all same-architecture agents/endogenous variables into a single vmap forward/derivative call. See Experimental: stacked evaluation.

set_initial_guess

def set_initial_guess(self, initial_guess: Dict[str, float])

Set the initial guess (uniform value across the state variable domain) for agents or endogenous variables. This is the boundary condition at \(t=T\) in the first time iteration.