"""Module for the Causal PINN solver."""
import torch
from ...problem import TimeDependentProblem
from .pinn import PINN
from ...utils import check_consistency
[docs]
class CausalPINN(PINN):
r"""
Causal Physics-Informed Neural Network (CausalPINN) solver class.
This class implements the Causal Physics-Informed Neural Network solver,
using a user specified ``model`` to solve a specific ``problem``.
It can be used to solve both forward and inverse problems.
The Causal Physics-Informed Neural Network solver aims to find the solution
:math:`\mathbf{u}:\Omega\rightarrow\mathbb{R}^m` of a differential problem:
.. math::
\begin{cases}
\mathcal{A}[\mathbf{u}](\mathbf{x})=0\quad,\mathbf{x}\in\Omega\\
\mathcal{B}[\mathbf{u}](\mathbf{x})=0\quad,
\mathbf{x}\in\partial\Omega
\end{cases}
minimizing the loss function:
.. math::
\mathcal{L}_{\rm{problem}} = \frac{1}{N_t}\sum_{i=1}^{N_t}
\omega_{i}\mathcal{L}_r(t_i),
where:
.. math::
\mathcal{L}_r(t) = \frac{1}{N}\sum_{i=1}^N
\mathcal{L}(\mathcal{A}[\mathbf{u}](\mathbf{x}_i, t)) +
\frac{1}{N}\sum_{i=1}^N
\mathcal{L}(\mathcal{B}[\mathbf{u}](\mathbf{x}_i, t))
and,
.. math::
\omega_i = \exp\left(\epsilon \sum_{k=1}^{i-1}\mathcal{L}_r(t_k)\right).
:math:`\epsilon` is an hyperparameter, set by default to :math:`100`, while
:math:`\mathcal{L}` is a specific loss function, typically the MSE:
.. math::
\mathcal{L}(v) = \| v \|^2_2.
.. seealso::
**Original reference**: Wang, Sifan, Shyam Sankaran, and Paris
Perdikaris.
*Respecting causality for training physics-informed
neural networks.*
Computer Methods in Applied Mechanics and Engineering 421 (2024):116813.
DOI: `10.1016 <https://doi.org/10.1016/j.cma.2024.116813>`_.
.. note::
This class is only compatible with problems that inherit from the
:class:`~pina.problem.time_dependent_problem.TimeDependentProblem`
class.
"""
def __init__(
self,
problem,
model,
optimizer=None,
scheduler=None,
weighting=None,
loss=None,
eps=100,
):
"""
Initialization of the :class:`CausalPINN` class.
:param AbstractProblem problem: The problem to be solved. It must
inherit from at least
:class:`~pina.problem.time_dependent_problem.TimeDependentProblem`.
:param torch.nn.Module model: The neural network model to be used.
:param Optimizer optimizer: The optimizer to be used.
If ``None``, the :class:`torch.optim.Adam` optimizer is used.
Default is ``None``.
:param torch.optim.LRScheduler scheduler: Learning rate scheduler.
If ``None``, the :class:`torch.optim.lr_scheduler.ConstantLR`
scheduler is used. Default is ``None``.
:param WeightingInterface weighting: The weighting schema to be used.
If ``None``, no weighting schema is used. Default is ``None``.
:param torch.nn.Module loss: The loss function to be minimized.
If ``None``, the :class:`torch.nn.MSELoss` loss is used.
Default is `None`.
:param float eps: The exponential decay parameter. Default is ``100``.
:raises ValueError: If the problem is not a TimeDependentProblem.
"""
super().__init__(
model=model,
problem=problem,
optimizer=optimizer,
scheduler=scheduler,
weighting=weighting,
loss=loss,
)
# checking consistency
check_consistency(eps, (int, float))
self._eps = eps
if not isinstance(self.problem, TimeDependentProblem):
raise ValueError(
"Casual PINN works only for problems"
"inheriting from TimeDependentProblem."
)
[docs]
def loss_phys(self, samples, equation):
"""
Computes the physics loss for the physics-informed solver based on the
provided samples and equation.
:param LabelTensor samples: The samples to evaluate the physics loss.
:param EquationInterface equation: The governing equation.
:return: The computed physics loss.
:rtype: LabelTensor
"""
# split sequentially ordered time tensors into chunks
chunks, labels = self._split_tensor_into_chunks(samples)
# compute residuals - this correspond to ordered loss functions
# values for each time step. Apply `flatten` to ensure obtaining
# a tensor of shape #chunks after concatenating the residuals
time_loss = []
for chunk in chunks:
chunk.labels = labels
# classical PINN loss
residual = self.compute_residual(samples=chunk, equation=equation)
loss_val = self._loss_fn(
torch.zeros_like(residual, requires_grad=True), residual
)
time_loss.append(loss_val)
# concatenate residuals
time_loss = torch.stack(time_loss)
# compute weights without storing the gradient
with torch.no_grad():
weights = self._compute_weights(time_loss)
return (weights * time_loss).mean()
@property
def eps(self):
"""
The exponential decay parameter.
:return: The exponential decay parameter.
:rtype: float
"""
return self._eps
@eps.setter
def eps(self, value):
"""
Set the exponential decay parameter.
:param float value: The exponential decay parameter.
"""
check_consistency(value, float)
self._eps = value
def _sort_label_tensor(self, tensor):
"""
Sort the tensor with respect to the temporal variables.
:param LabelTensor tensor: The tensor to be sorted.
:return: The tensor sorted with respect to the temporal variables.
:rtype: LabelTensor
"""
# labels input tensors
labels = tensor.labels
# extract time tensor
time_tensor = tensor.extract(self.problem.temporal_domain.variables)
# sort the time tensors (this is very bad for GPU)
_, idx = torch.sort(time_tensor.tensor.flatten())
tensor = tensor[idx]
tensor.labels = labels
return tensor
def _split_tensor_into_chunks(self, tensor):
"""
Split the tensor into chunks based on time.
:param LabelTensor tensor: The tensor to be split.
:return: A tuple containing the list of tensor chunks and the
corresponding labels.
:rtype: tuple[list[LabelTensor], list[str]]
"""
# extract labels
labels = tensor.labels
# sort input tensor based on time
tensor = self._sort_label_tensor(tensor)
# extract time tensor
time_tensor = tensor.extract(self.problem.temporal_domain.variables)
# count unique tensors in time
_, idx_split = time_tensor.unique(return_counts=True)
# split the tensor based on time
chunks = torch.split(tensor, tuple(idx_split))
return chunks, labels
def _compute_weights(self, loss):
"""
Compute the weights for the physics loss based on the cumulative loss.
:param LabelTensor loss: The physics loss values.
:return: The computed weights for the physics loss.
:rtype: LabelTensor
"""
# compute comulative loss and multiply by epsilon
cumulative_loss = self._eps * torch.cumsum(loss, dim=0)
# return the exponential of the negative weighted cumulative sum
return torch.exp(-cumulative_loss)
