Describe the bug
When trying to use a distance_metric that requires additional distance_metric_params, distance matching does not seem able to parse out additional kwargs. In the __init__ method of DistanceMatchingEstimator here, the code looks like this:
def __init__(
self,
identified_estimand: IdentifiedEstimand,
test_significance: Union[bool, str] = False,
evaluate_effect_strength: bool = False,
confidence_intervals: bool = False,
num_null_simulations: int = CausalEstimator.DEFAULT_NUMBER_OF_SIMULATIONS_STAT_TEST,
num_simulations: int = CausalEstimator.DEFAULT_NUMBER_OF_SIMULATIONS_CI,
sample_size_fraction: int = CausalEstimator.DEFAULT_SAMPLE_SIZE_FRACTION,
confidence_level: float = CausalEstimator.DEFAULT_CONFIDENCE_LEVEL,
need_conditional_estimates: Union[bool, str] = "auto",
num_quantiles_to_discretize_cont_cols: int = CausalEstimator.NUM_QUANTILES_TO_DISCRETIZE_CONT_COLS,
num_matches_per_unit: int = 1,
distance_metric: str = "minkowski",
**kwargs,
):
...
self.distance_metric_params = {}
for param_name in self.Valid_Dist_Metric_Params:
param_val = getattr(self, param_name, None)
if param_val is not None:
self.distance_metric_params[param_name] = param_val
...Steps to reproduce the behavior
import dowhy
from dowhy import CausalModel
from causaldata import black_politicians
data = black_politicians.load_pandas().data
X = data.drop("responded", axis=1).values
y = data["responded"].values
model = CausalModel(
data=data,
treatment=["leg_black"],
outcome="responded",
common_causes=["medianhhincom", "blackpercent, "leg_democrat"]
)
identified_estimand = model.identify_effect(proceed_when_unindentifiable=True)
# this will error
model.estimate_effect(
identifed_estimand=identified_estimand,
method_name="backdoor.distance_matching",
target_units="att",
method_params={
"distance_metric": "mahalanobis",
"V": np.cov(X.T)
}
)The error will be:
ValueError: Must provide either V or VI for Mahalanobis distance
Expected behavior
When method_name="backdoor.distance_matching" for an instance of CausalModel, it should be possible to pass additional parameters required for distance matching. These additional parameters can be found here.
Version information:
Describe the bug
When trying to use a
distance_metricthat requires additionaldistance_metric_params, distance matching does not seem able to parse out additionalkwargs. In the__init__method ofDistanceMatchingEstimatorhere, the code looks like this:Steps to reproduce the behavior
The error will be:
Expected behavior
When
method_name="backdoor.distance_matching"for an instance ofCausalModel, it should be possible to pass additional parameters required for distance matching. These additional parameters can be found here.Version information: