catenets.models.torch.utils.weight_utils module

Implement different reweighting/balancing strategies as in Li et al (2018)

compute_importance_weights(propensity: torch.Tensor, w: torch.Tensor, weighting_strategy: str, weight_args: Optional[dict] = None) → torch.Tensor

compute_ipw(propensity: torch.Tensor, w: torch.Tensor) → torch.Tensor

compute_matching_weights(propensity: torch.Tensor, w: torch.Tensor) → torch.Tensor

compute_overlap_weights(propensity: torch.Tensor, w: torch.Tensor) → torch.Tensor

compute_trunc_ipw(propensity: torch.Tensor, w: torch.Tensor, cutoff: float = 0.05) → torch.Tensor