Merge pull request #292 from CITCOM-project/sparse-treatment-strategies

Support for treatment strategies with interventions at non-consecutiv…

Author: jmafoster1

causal_testing/estimation/ipcw_estimator.py

@@ -2,7 +2,6 @@
 
 import logging
 from typing import Any
-import numpy as np
 
 from causal_testing.specification.variable import Variable
 from causal_testing.testing.causal_test_outcome import CausalTestOutcome
@@ -81,21 +80,13 @@ def _return_causal_test_results(self, estimator) -> CausalTestResult:
         if not hasattr(estimator, f"estimate_{self.estimate_type}"):
             raise AttributeError(f"{estimator.__class__} has no {self.estimate_type} method.")
         estimate_effect = getattr(estimator, f"estimate_{self.estimate_type}")
-        try:
-            effect, confidence_intervals = estimate_effect(**self.estimate_params)
-            return CausalTestResult(
-                estimator=estimator,
-                test_value=TestValue(self.estimate_type, effect),
-                effect_modifier_configuration=self.effect_modifier_configuration,
-                confidence_intervals=confidence_intervals,
-            )
-        except np.linalg.LinAlgError:
-            return CausalTestResult(
-                estimator=estimator,
-                test_value=TestValue(self.estimate_type, None),
-                effect_modifier_configuration=self.effect_modifier_configuration,
-                confidence_intervals=None,
-            )
+        effect, confidence_intervals = estimate_effect(**self.estimate_params)
+        return CausalTestResult(
+            estimator=estimator,
+            test_value=TestValue(self.estimate_type, effect),
+            effect_modifier_configuration=self.effect_modifier_configuration,
+            confidence_intervals=confidence_intervals,
+        )
 
     def __str__(self):
         treatment_config = {self.treatment_variable.name: self.treatment_value}

causal_testing/specification/capabilities.py

@@ -4,7 +4,6 @@
 import matplotlib.pyplot as plt
 from causal_testing.specification.variable import Input
 from causal_testing.utils.validation import CausalValidator
-from causal_testing.specification.capabilities import TreatmentSequence
 
 from causal_testing.estimation.cubic_spline_estimator import CubicSplineRegressionEstimator
 from causal_testing.estimation.linear_regression_estimator import LinearRegressionEstimator

causal_testing/testing/causal_test_case.py

@@ -4,7 +4,6 @@
 import matplotlib.pyplot as plt
 from causal_testing.specification.variable import Input
 from causal_testing.utils.validation import CausalValidator
-from causal_testing.specification.capabilities import TreatmentSequence
 
 from causal_testing.estimation.instrumental_variable_estimator import InstrumentalVariableEstimator
 

tests/estimation_tests/test_cubic_spline_estimator.py

@@ -4,7 +4,6 @@
 import matplotlib.pyplot as plt
 from causal_testing.specification.variable import Input
 from causal_testing.utils.validation import CausalValidator
-from causal_testing.specification.capabilities import TreatmentSequence
 
 from causal_testing.estimation.ipcw_estimator import IPCWEstimator
 
@@ -16,8 +15,8 @@ class TestIPCWEstimator(unittest.TestCase):
 
     def test_estimate_hazard_ratio(self):
         timesteps_per_intervention = 1
-        control_strategy = TreatmentSequence(timesteps_per_intervention, [("t", 0), ("t", 0), ("t", 0)])
-        treatment_strategy = TreatmentSequence(timesteps_per_intervention, [("t", 1), ("t", 1), ("t", 1)])
+        control_strategy = [[t, "t", 0] for t in range(1, 4, timesteps_per_intervention)]
+        treatment_strategy = [[t, "t", 1] for t in range(1, 4, timesteps_per_intervention)]
         outcome = "outcome"
         fit_bl_switch_formula = "xo_t_do ~ time"
         df = pd.read_csv("tests/resources/data/temporal_data.csv")
@@ -34,12 +33,12 @@ def test_estimate_hazard_ratio(self):
             eligibility=None,
         )
         estimate, intervals = estimation_model.estimate_hazard_ratio()
-        self.assertEqual(estimate["trtrand"], 1.0)
+        self.assertEqual(round(estimate["trtrand"], 3), 1.351)
 
     def test_invalid_treatment_strategies(self):
         timesteps_per_intervention = 1
-        control_strategy = TreatmentSequence(timesteps_per_intervention, [("t", 0), ("t", 0), ("t", 0)])
-        treatment_strategy = TreatmentSequence(timesteps_per_intervention, [("t", 1), ("t", 1), ("t", 1)])
+        control_strategy = [[t, "t", 0] for t in range(1, 4, timesteps_per_intervention)]
+        treatment_strategy = [[t, "t", 1] for t in range(1, 4, timesteps_per_intervention)]
         outcome = "outcome"
         fit_bl_switch_formula = "xo_t_do ~ time"
         df = pd.read_csv("tests/resources/data/temporal_data.csv")
@@ -60,8 +59,8 @@ def test_invalid_treatment_strategies(self):
 
     def test_invalid_fault_t_do(self):
         timesteps_per_intervention = 1
-        control_strategy = TreatmentSequence(timesteps_per_intervention, [("t", 0), ("t", 0), ("t", 0)])
-        treatment_strategy = TreatmentSequence(timesteps_per_intervention, [("t", 1), ("t", 1), ("t", 1)])
+        control_strategy = [[t, "t", 0] for t in range(1, 4, timesteps_per_intervention)]
+        treatment_strategy = [[t, "t", 1] for t in range(1, 4, timesteps_per_intervention)]
         outcome = "outcome"
         fit_bl_switch_formula = "xo_t_do ~ time"
         df = pd.read_csv("tests/resources/data/temporal_data.csv")
@@ -80,3 +79,47 @@ def test_invalid_fault_t_do(self):
         estimation_model.df["fault_t_do"] = 0
         with self.assertRaises(ValueError):
             estimate, intervals = estimation_model.estimate_hazard_ratio()
+
+    def test_no_individual_began_control_strategy(self):
+        timesteps_per_intervention = 1
+        control_strategy = [[t, "t", 0] for t in range(1, 4, timesteps_per_intervention)]
+        treatment_strategy = [[t, "t", 1] for t in range(1, 4, timesteps_per_intervention)]
+        outcome = "outcome"
+        fit_bl_switch_formula = "xo_t_do ~ time"
+        df = pd.read_csv("tests/resources/data/temporal_data.csv")
+        df["t"] = 1
+        df["ok"] = df["outcome"] == 1
+        with self.assertRaises(ValueError):
+            estimation_model = IPCWEstimator(
+                df,
+                timesteps_per_intervention,
+                control_strategy,
+                treatment_strategy,
+                outcome,
+                "ok",
+                fit_bl_switch_formula=fit_bl_switch_formula,
+                fit_bltd_switch_formula=fit_bl_switch_formula,
+                eligibility=None,
+            )
+
+    def test_no_individual_began_treatment_strategy(self):
+        timesteps_per_intervention = 1
+        control_strategy = [[t, "t", 0] for t in range(1, 4, timesteps_per_intervention)]
+        treatment_strategy = [[t, "t", 1] for t in range(1, 4, timesteps_per_intervention)]
+        outcome = "outcome"
+        fit_bl_switch_formula = "xo_t_do ~ time"
+        df = pd.read_csv("tests/resources/data/temporal_data.csv")
+        df["t"] = 0
+        df["ok"] = df["outcome"] == 1
+        with self.assertRaises(ValueError):
+            estimation_model = IPCWEstimator(
+                df,
+                timesteps_per_intervention,
+                control_strategy,
+                treatment_strategy,
+                outcome,
+                "ok",
+                fit_bl_switch_formula=fit_bl_switch_formula,
+                fit_bltd_switch_formula=fit_bl_switch_formula,
+                eligibility=None,
+            )

tests/estimation_tests/test_instrumental_variable_estimator.py

@@ -4,7 +4,6 @@
 import matplotlib.pyplot as plt
 from causal_testing.specification.variable import Input
 from causal_testing.utils.validation import CausalValidator
-from causal_testing.specification.capabilities import TreatmentSequence
 
 from causal_testing.estimation.linear_regression_estimator import LinearRegressionEstimator
 from causal_testing.estimation.genetic_programming_regression_fitter import reciprocal

tests/estimation_tests/test_ipcw_estimator.py

@@ -4,7 +4,6 @@
 import matplotlib.pyplot as plt
 from causal_testing.specification.variable import Input
 from causal_testing.utils.validation import CausalValidator
-from causal_testing.specification.capabilities import TreatmentSequence
 from causal_testing.estimation.logistic_regression_estimator import LogisticRegressionEstimator
 
 

tests/estimation_tests/test_linear_regression_estimator.py

@@ -1,8 +1,7 @@
+import os
 import unittest
 from pathlib import Path
-from statistics import StatisticsError
 import scipy
-import os
 import pandas as pd
 
 from causal_testing.estimation.linear_regression_estimator import LinearRegressionEstimator
@@ -11,12 +10,9 @@
 from causal_testing.testing.causal_test_case import CausalTestCase
 from causal_testing.testing.causal_test_suite import CausalTestSuite
 from causal_testing.testing.causal_test_adequacy import DAGAdequacy
-from causal_testing.testing.causal_test_outcome import NoEffect, Positive, SomeEffect
+from causal_testing.testing.causal_test_outcome import NoEffect, SomeEffect
 from causal_testing.json_front.json_class import JsonUtility, CausalVariables
-from causal_testing.specification.variable import Input, Output, Meta
 from causal_testing.specification.scenario import Scenario
-from causal_testing.specification.causal_specification import CausalSpecification
-from causal_testing.specification.capabilities import TreatmentSequence
 from causal_testing.testing.causal_test_adequacy import DataAdequacy
 
 
@@ -112,8 +108,8 @@ def test_data_adequacy_cateogorical(self):
 
     def test_data_adequacy_group_by(self):
         timesteps_per_intervention = 1
-        control_strategy = TreatmentSequence(timesteps_per_intervention, [("t", 0), ("t", 0), ("t", 0)])
-        treatment_strategy = TreatmentSequence(timesteps_per_intervention, [("t", 1), ("t", 1), ("t", 1)])
+        control_strategy = [[t, "t", 0] for t in range(1, 4, timesteps_per_intervention)]
+        treatment_strategy = [[t, "t", 1] for t in range(1, 4, timesteps_per_intervention)]
         outcome = "outcome"
         fit_bl_switch_formula = "xo_t_do ~ time"
         df = pd.read_csv("tests/resources/data/temporal_data.csv")
@@ -145,11 +141,12 @@ def test_data_adequacy_group_by(self):
         causal_test_result = causal_test_case.execute_test(estimation_model, None)
         adequacy_metric = DataAdequacy(causal_test_case, estimation_model, group_by="id")
         adequacy_metric.measure_adequacy()
-        causal_test_result.adequacy = adequacy_metric
-        print(causal_test_result.adequacy.to_dict())
+        adequacy_dict = adequacy_metric.to_dict()
+        self.assertEqual(round(adequacy_dict["kurtosis"]["trtrand"], 3), -0.857)
+        adequacy_dict.pop("kurtosis")
         self.assertEqual(
-            causal_test_result.adequacy.to_dict(),
-            {"kurtosis": {"trtrand": 0.0}, "bootstrap_size": 100, "passing": 0, "successful": 95},
+            adequacy_dict,
+            {"bootstrap_size": 100, "passing": 32, "successful": 100},
         )
 
     def test_dag_adequacy_dependent(self):