Merge pull request #163 from CITCOM-project/functional_form

Functional form

Author: jmafoster1

causal_testing/testing/estimators.py

@@ -50,18 +50,18 @@ def doubling_beta_CATE_on_csv(observational_data_path: str, simulate_counterfact
     linear_regression_estimator = LinearRegressionEstimator('beta', 0.032, 0.016,
                                                             {'avg_age', 'contacts'},  # We use custom adjustment set
                                                             'cum_infections',
-                                                            df=past_execution_df)
+                                                            df=past_execution_df,
+                                                            formula="cum_infections ~ beta + np.power(beta, 2) + avg_age + contacts")
 
     # Add squared terms for beta, since it has a quadratic relationship with cumulative infections
-    linear_regression_estimator.add_squared_term_to_df('beta')
     causal_test_result = causal_test_engine.execute_test(linear_regression_estimator, causal_test_case, 'ate')
 
     # Repeat for association estimate (no adjustment)
     no_adjustment_linear_regression_estimator = LinearRegressionEstimator('beta', 0.032, 0.016,
                                                                           set(),
                                                                           'cum_infections',
-                                                                          df=past_execution_df)
-    no_adjustment_linear_regression_estimator.add_squared_term_to_df('beta')
+                                                                          df=past_execution_df,
+                                                                          formula="cum_infections ~ beta + np.power(beta, 2)")
     association_test_result = causal_test_engine.execute_test(no_adjustment_linear_regression_estimator, causal_test_case, 'ate')
 
     # Store results for plotting
@@ -82,8 +82,8 @@ def doubling_beta_CATE_on_csv(observational_data_path: str, simulate_counterfact
         counterfactual_linear_regression_estimator = LinearRegressionEstimator('beta', 0.032, 0.016,
                                                                                {'avg_age', 'contacts'},
                                                                                'cum_infections',
-                                                                               df=counterfactual_past_execution_df)
-        counterfactual_linear_regression_estimator.add_squared_term_to_df('beta')
+                                                                               df=counterfactual_past_execution_df,
+                                                                               formula="cum_infections ~ beta + np.power(beta, 2) + avg_age + contacts")
         counterfactual_causal_test_result = causal_test_engine.execute_test(linear_regression_estimator, causal_test_case, 'ate')
         results_dict['counterfactual'] = {'ate': counterfactual_causal_test_result.test_value.value,
                                           'cis': counterfactual_causal_test_result.confidence_intervals,

examples/covasim_/doubling_beta/causal_test_beta.py

@@ -0,0 +1,2 @@
+intensity_num_shapes_results_random_1000.csv
+width_num_shapes_results_random_1000.csv

examples/poisson-line-process/.gitignore

@@ -117,20 +117,18 @@ def test_intensity_num_shapes(
             effect_modifiers=causal_test_case.effect_modifier_configuration,
         )
     else:
+        square_terms = [f"np.power({t}, 2)" for t in square_terms]
+        inverse_terms = [f"np.float_power({t}, -1)" for t in inverse_terms]
         estimator = LinearRegressionEstimator(
             treatment=treatment,
             control_value=causal_test_case.control_value,
             treatment_value=causal_test_case.treatment_value,
             adjustment_set=set(),
             outcome=outcome,
             df=data,
-            intercept=0,
             effect_modifiers=causal_test_case.effect_modifier_configuration,
+            formula=f"{outcome} ~ {treatment} + {'+'.join(square_terms + inverse_terms + list([e.name for e in causal_test_case.effect_modifier_configuration]))} -1"
         )
-        for t in square_terms:
-            estimator.add_squared_term_to_df(t)
-        for t in inverse_terms:
-            estimator.add_inverse_term_to_df(t)
 
     # 10. Execute the test
     causal_test_result = causal_test_engine.execute_test(

examples/poisson-line-process/causal_test_poisson.py

@@ -7,5 +7,3 @@ To run this case study:
    (instructions are provided in the project README).
 2. Change directory to `causal_testing/examples/poisson`.
 3. Run the command `python run_causal_tests.py --data_path data.csv --dag_path dag.dot --json_path causal_tests.json`
-
-This should print a series of causal test results and produce two CSV files. `intensity_num_shapes_results_random_1000.csv` corresponds to table 1, and `width_num_shapes_results_random_1000.csv` relates to our findings regarding the relationship of width and `P_u`.

examples/poisson-line-process/intensity_num_shapes_results_random_1000.csv

@@ -150,11 +150,7 @@ class MyJsonUtility(JsonUtility):
     def add_modelling_assumptions(self, estimation_model: Estimator):
         # Add squared intensity term as a modelling assumption if intensity is the treatment of the test
         if "intensity" in estimation_model.treatment[0]:
-            estimation_model.add_squared_term_to_df("intensity")
             estimation_model.intercept = 0
-        if isinstance(estimation_model, WidthHeightEstimator):
-            estimation_model.add_product_term_to_df("width", "intensity")
-            estimation_model.add_product_term_to_df("height", "intensity")
 
 
 if __name__ == "__main__":

examples/poisson-line-process/width_num_shapes_results_random_1000.csv

@@ -67,8 +67,6 @@ def test_argparse(self):
 
     def test_setup_modelling_scenario(self):
         self.json_class.setup()
-        print(type(self.json_class.modelling_scenario))
-        print(self.json_class.modelling_scenario)
         self.assertIsInstance(self.json_class.modelling_scenario, Scenario)
 
     def test_setup_causal_specification(self):

examples/poisson/README.md

@@ -218,8 +218,8 @@ def test_execute_test_observational_linear_regression_estimator_squared_term(sel
             self.minimal_adjustment_set,
             "C",
             self.causal_test_engine.scenario_execution_data_df,
+            formula=f"C ~ A + {'+'.join(self.minimal_adjustment_set)} + (D ** 2)",
         )
-        estimation_model.add_squared_term_to_df("D")
         causal_test_result = self.causal_test_engine.execute_test(estimation_model, self.causal_test_case)
         self.assertAlmostEqual(round(causal_test_result.test_value.value, 1), 4, delta=1)
 
@@ -240,7 +240,7 @@ def test_execute_observational_causal_forest_estimator_cates(self):
             self.minimal_adjustment_set,
             "C",
             self.causal_test_engine.scenario_execution_data_df,
-            effect_modifiers={Input("M", int): None},
+            effect_modifiers={"M": None},
         )
         causal_test_result = self.causal_test_engine.execute_test(
             estimation_model, self.causal_test_case, estimate_type="cate"