EconML MetaLearner Estimators and Categorical Variables - Error: None of Index are in the Columns

[kirant1729]

I have been trying to use EconML MetaLearner Estimators with DoWhy, but I get the following error message when estimating the effect. This only happens when I have categorical variables as confounders. With continous variables, it all works as expected.

""None of [Index(['Factor2', 'Factor1'], dtype='object')] are in the [columns]""

This is based on simulated data

Data

np.random.seed(42)

num_rows = 1000

data = {
'Treatment': np.random.choice([0, 1], size=num_rows),
'Factor1': np.random.choice(['A', 'B', 'C'], size=num_rows),
'Factor2': np.random.choice(['D', 'E', 'F', 'G'], size=num_rows),
#'Outcomes': np.random.choice([1, 2, 3, 4, 5], size=num_rows)
'Outcomes': np.random.choice([0,1], size=num_rows)
}`

DAG

Estimate

# Get estimate (S-Learner) estimate = model.estimate_effect( identified_estimand=estimand, method_name='backdoor.econml.metalearners.SLearner', target_units='ate', method_params={ 'init_params': { 'overall_model': LGBMRegressor(n_estimators=500, max_depth=10) }, 'fit_params': {} })

Full Error Message

`---------------------------------------------------------------------------
KeyError Traceback (most recent call last)
Cell In[10], line 2
1 # Get estimate (S-Learner)
----> 2 estimate = model.estimate_effect(
3 identified_estimand=estimand,
4 method_name='backdoor.econml.metalearners.SLearner',
5 target_units='ate',
6 method_params={
7 'init_params': {
8 'overall_model': LGBMRegressor(n_estimators=500, max_depth=10)
9 },
10 'fit_params': {}
11 })

File \AppData\Local\Programs\Python\Python312\Lib\site-packages\dowhy\causal_model.py:361, in CausalModel.estimate_effect(self, identified_estimand, method_name, control_value, treatment_value, test_significance, evaluate_effect_strength, confidence_intervals, target_units, effect_modifiers, fit_estimator, method_params)
350 causal_estimator = causal_estimator_class(
351 identified_estimand,
352 test_significance=test_significance,
(...)
356 **extra_args,
357 )
359 self._estimator_cache[method_name] = causal_estimator
--> 361 return estimate_effect(
362 self._data,
363 self._treatment,
364 self._outcome,
365 identifier_name,
366 causal_estimator,
367 control_value,
368 treatment_value,
369 target_units,
370 effect_modifiers,
371 fit_estimator,
372 method_params,
373 )

File \AppData\Local\Programs\Python\Python312\Lib\site-packages\dowhy\causal_estimator.py:758, in estimate_effect(data, treatment, outcome, identifier_name, estimator, control_value, treatment_value, target_units, effect_modifiers, fit_estimator, method_params)
751 if fit_estimator:
752 estimator.fit(
753 data=data,
754 effect_modifier_names=effect_modifiers,
755 **method_params["fit_params"] if "fit_params" in method_params else {},
756 )
--> 758 estimate = estimator.estimate_effect(
759 data,
760 treatment_value=treatment_value,
761 control_value=control_value,
762 target_units=target_units,
763 confidence_intervals=estimator._confidence_intervals,
764 )
766 if estimator._significance_test:
767 estimator.test_significance(data, estimate.value, method=estimator._significance_test)

File \AppData\Local\Programs\Python\Python312\Lib\site-packages\dowhy\causal_estimators\econml.py:244, in Econml.estimate_effect(self, data, treatment_value, control_value, target_units, **_)
241 # Changing shape to a list for a singleton value
242 # Note that self._control_value is assumed to be a singleton value
243 self._treatment_value = parse_state(self._treatment_value)
--> 244 est = self.effect(X_test)
245 ate = np.mean(est, axis=0) # one value per treatment value
247 if len(ate) == 1:

File \AppData\Local\Programs\Python\Python312\Lib\site-packages\dowhy\causal_estimators\econml.py:327, in Econml.effect(self, df, *args, **kwargs)
324 def effect_fun(filtered_df, T0, T1, *args, **kwargs):
325 return self.estimator.effect(filtered_df, T0=T0, T1=T1, *args, **kwargs)
--> 327 Xdf = df[self._effect_modifier_names] if df is not None else df
328 return self.apply_multitreatment(Xdf, effect_fun, *args, **kwargs)

File \AppData\Local\Programs\Python\Python312\Lib\site-packages\pandas\core\frame.py:4108, in DataFrame.getitem(self, key)
4106 if is_iterator(key):
4107 key = list(key)
-> 4108 indexer = self.columns._get_indexer_strict(key, "columns")[1]
4110 # take() does not accept boolean indexers
4111 if getattr(indexer, "dtype", None) == bool:

File \AppData\Local\Programs\Python\Python312\Lib\site-packages\pandas\core\indexes\base.py:6200, in Index._get_indexer_strict(self, key, axis_name)
6197 else:
6198 keyarr, indexer, new_indexer = self._reindex_non_unique(keyarr)
-> 6200 self._raise_if_missing(keyarr, indexer, axis_name)
6202 keyarr = self.take(indexer)
6203 if isinstance(key, Index):
6204 # GH 42790 - Preserve name from an Index

File \AppData\Local\Programs\Python\Python312\Lib\site-packages\pandas\core\indexes\base.py:6249, in Index._raise_if_missing(self, key, indexer, axis_name)
6247 if nmissing:
6248 if nmissing == len(indexer):
-> 6249 raise KeyError(f"None of [{key}] are in the [{axis_name}]")
6251 not_found = list(ensure_index(key)[missing_mask.nonzero()[0]].unique())
6252 raise KeyError(f"{not_found} not in index")

KeyError: "None of [Index(['Factor2', 'Factor1'], dtype='object')] are in the [columns]"`