"""
Implementation of the IRM algorithm from `"Invariant Risk Minimization"
<https://arxiv.org/abs/1907.02893>`_ paper
"""
import torch
from torch.autograd import grad
from torch import Tensor
from torch_geometric.data import Batch
from GOOD import register
from GOOD.utils.config_reader import Union, CommonArgs, Munch
from .BaseOOD import BaseOODAlg
[docs]@register.ood_alg_register
class IRM(BaseOODAlg):
r"""
Implementation of the IRM algorithm from `"Invariant Risk Minimization"
<https://arxiv.org/abs/1907.02893>`_ paper
Args:
config (Union[CommonArgs, Munch]): munchified dictionary of args (:obj:`config.device`, :obj:`config.dataset.num_envs`, :obj:`config.ood.ood_param`)
"""
def __init__(self, config: Union[CommonArgs, Munch]):
super(IRM, self).__init__(config)
self.dummy_w = torch.nn.Parameter(torch.Tensor([1.0])).to(config.device)
[docs] def output_postprocess(self, model_output: Tensor, **kwargs) -> Tensor:
r"""
Process the raw output of model; apply the linear classifier
Args:
model_output (Tensor): model raw output
Returns (Tensor):
model raw predictions with the linear classifier applied
"""
raw_pred = self.dummy_w * model_output
return raw_pred
[docs] def loss_postprocess(self, loss: Tensor, data: Batch, mask: Tensor, config: Union[CommonArgs, Munch], **kwargs) -> Tensor:
r"""
Process loss based on IRM algorithm
Args:
loss (Tensor): base loss between model predictions and input labels
data (Batch): input data
mask (Tensor): NAN masks for data formats
config (Union[CommonArgs, Munch]): munchified dictionary of args (:obj:`config.device`, :obj:`config.dataset.num_envs`, :obj:`config.ood.ood_param`)
.. code-block:: python
config = munchify({device: torch.device('cuda'),
dataset: {num_envs: int(10)},
ood: {ood_param: float(0.1)}
})
Returns (Tensor):
loss with IRM penalty
"""
spec_loss_list = []
for i in range(config.dataset.num_envs):
env_idx = data.env_id == i
if loss[env_idx].shape[0] > 0:
grad_all = torch.sum(
grad(loss[env_idx].sum() / mask[env_idx].sum(), self.dummy_w, create_graph=True)[0].pow(2))
spec_loss_list.append(grad_all)
spec_loss = config.ood.ood_param * sum(spec_loss_list) / len(spec_loss_list)
if torch.isnan(spec_loss):
spec_loss = 0
mean_loss = loss.sum() / mask.sum()
loss = spec_loss + mean_loss
self.mean_loss = mean_loss
self.spec_loss = spec_loss
return loss