Add causal analysis

azure-pipelines-steps.yml

@@ -53,6 +53,11 @@ jobs:
     displayName: 'Install graphviz on Linux'
     condition: and(succeeded(), eq(variables['Agent.OS'], 'Linux'))
 
+  # Install OpenMP on Mac to support lightgbm
+  - script: 'brew install libomp'
+    displayName: 'Install OpenMP on Mac'
+    condition: and(succeeded(), eq(variables['Agent.OS'], 'Darwin'))
+
   # Install the package
   - script: 'python -m pip install --upgrade pip && pip install --upgrade setuptools wheel Cython && pip install ${{ parameters.package }}'
     displayName: 'Install dependencies'

doc/reference.rst

@@ -252,6 +252,17 @@ Utilities
 
     econml.utilities
 
+Solutions
+---------
+
+Causal Feature Importance
+^^^^^^^^^^^^^^^^^^^^^^^^^
+
+.. autosummary::
+    :toctree: _autosummary
+
+    econml.solutions.causal_analysis
+
 Private Module Reference
 ========================
 

econml/__init__.py

@@ -18,6 +18,7 @@
            'ortho_iv',
            'policy',
            'score',
+           'solutions',
            'sklearn_extensions',
            'tree',
            'two_stage_least_squares',

econml/inference/_inference.py

@@ -867,6 +867,19 @@ def _expand_outputs(self, n_rows):
         """
         pass
 
+    def translate(self, offset):
+        """
+        Update the results in place by translating by an offset.
+
+        Parameters
+        ----------
+        offset: array-like
+            The offset by which to translate these results
+        """
+        # NOTE: use np.asarray(offset) becuase if offset is a pd.Series direct addition would make the sum
+        #       a Series as well, which would subsequently break summary_frame because flatten isn't supported
+        self.pred = self.pred + np.asarray(offset)
+
 
 class NormalInferenceResults(InferenceResults):
     """
@@ -1081,6 +1094,14 @@ def _expand_outputs(self, n_rows):
         return EmpiricalInferenceResults(self.d_t, self.d_y, pred, pred_dist, self.inf_type, self.fname_transformer,
                                          self.feature_names, self.output_names, self.treatment_names)
 
+    def translate(self, other):
+        # offset preds
+        super().translate(other)
+        # offset the distribution, too
+        self.pred_dist = self.pred_dist + np.asarray(other)
+
+    translate.__doc__ = InferenceResults.translate.__doc__
+
 
 class PopulationSummaryResults:
     """

econml/solutions/causal_analysis/__init__.py

@@ -0,0 +1,6 @@
+# Copyright (c) Microsoft Corporation. All rights reserved.
+# Licensed under the MIT License.
+
+from ._causal_analysis import CausalAnalysis
+
+__all__ = ["CausalAnalysis"]