PyTorch DiffEq solver

TorchDiffEqSolver

Bases: ForwardSolver

ODE-based forward solver using the TorchDiffEq library.

This solver integrates tumor growth models over specified timepoints using advanced ODE solvers and handles both radiotherapy and chemotherapy schedules.

Attributes:

Name	Type	Description
model	TumorGrowthModel3D	The tumor growth model to solve.
solver_options	TorchDiffEqSolverOptions	Configuration options for the solver.

__init__(model, solver_options)

Initializes the TorchDiffEqSolver.

Parameters:

Name	Type	Description	Default
model	TumorGrowthModel3D	The tumor growth model to solve.	required
solver_options	TorchDiffEqSolverOptions	Configuration options for the solver.	required

grid_constructor(func, y0, t)

Constructs a grid of timesteps considering treatment schedules.

Parameters:

Name	Type	Description	Default
func		The ODE function (unused in this method but required by the API).	required
y0		Initial state of the system (unused in this method but required by the API).	required
t		Original list of timepoints requested by the solver.	required

Returns:

Type	Description
Tensor	torch.Tensor: Tensor containing refined timepoints for integration.

solve(timepoints, u_initial)

Solves the tumor growth model over the specified timepoints.

Parameters:

Name	Type	Description	Default
timepoints	List[datetime]	List of timepoints at which the solution is desired.	required
u_initial	Tensor	Initial tumor density field.	required

Returns:

Type	Description
Tuple[List[datetime], List[Tensor]]	Tuple[List[datetime], List[torch.Tensor]]: - A list of datetime objects corresponding to the solution timepoints. - A list of torch.Tensor objects representing the tumor density at each timepoint.

TorchDiffEqSolverOptions dataclass

Configuration options for the TorchDiffEqSolver.

Attributes:

Name	Type	Description
step_size	timedelta	The integration step size for the solver.
method	str	The ODE solver method to use (e.g., "rk4", "dopri5").
device	device	The device on which to perform computations (e.g., CPU or GPU).
use_adjoint	bool	Whether to use the adjoint method for memory-efficient backpropagation.