convert scikitlearn models behind the scenes

causalpy/experiments/base.py

@@ -17,8 +17,10 @@
 
 from abc import abstractmethod
 
+from sklearn.base import RegressorMixin
+
 from causalpy.pymc_models import PyMCModel
-from causalpy.skl_models import ScikitLearnModel
+from causalpy.skl_models import create_causalpy_compatible_class
 
 
 class BaseExperiment:
@@ -28,13 +30,18 @@ class BaseExperiment:
     supports_ols: bool
 
     def __init__(self, model=None):
+        # Ensure we've made any provided Scikit Learn model (as identified as being type
+        # RegressorMixin) compatible with CausalPy by appending our custom methods.
+        if isinstance(model, RegressorMixin):
+            model = create_causalpy_compatible_class(model)
+
         if model is not None:
             self.model = model
 
         if isinstance(self.model, PyMCModel) and not self.supports_bayes:
             raise ValueError("Bayesian models not supported.")
 
-        if isinstance(self.model, ScikitLearnModel) and not self.supports_ols:
+        if isinstance(self.model, RegressorMixin) and not self.supports_ols:
             raise ValueError("OLS models not supported.")
 
         if self.model is None:
@@ -57,7 +64,7 @@ def plot(self, *args, **kwargs) -> tuple:
         """
         if isinstance(self.model, PyMCModel):
             return self.bayesian_plot(*args, **kwargs)
-        elif isinstance(self.model, ScikitLearnModel):
+        elif isinstance(self.model, RegressorMixin):
             return self.ols_plot(*args, **kwargs)
         else:
             raise ValueError("Unsupported model type")

causalpy/experiments/diff_in_diff.py

@@ -21,14 +21,14 @@
 import seaborn as sns
 from matplotlib import pyplot as plt
 from patsy import build_design_matrices, dmatrices
+from sklearn.base import RegressorMixin
 
 from causalpy.custom_exceptions import (
     DataException,
     FormulaException,
 )
 from causalpy.plot_utils import plot_xY
 from causalpy.pymc_models import PyMCModel
-from causalpy.skl_models import ScikitLearnModel
 from causalpy.utils import _is_variable_dummy_coded, convert_to_string, round_num
 
 from .base import BaseExperiment
@@ -106,7 +106,7 @@ def __init__(
         if isinstance(self.model, PyMCModel):
             COORDS = {"coeffs": self.labels, "obs_indx": np.arange(self.X.shape[0])}
             self.model.fit(X=self.X, y=self.y, coords=COORDS)
-        elif isinstance(self.model, ScikitLearnModel):
+        elif isinstance(self.model, RegressorMixin):
             self.model.fit(X=self.X, y=self.y)
         else:
             raise ValueError("Model type not recognized")
@@ -181,7 +181,7 @@ def __init__(
                     self.causal_impact = self.model.idata.posterior["beta"].isel(
                         {"coeffs": i}
                     )
-        elif isinstance(self.model, ScikitLearnModel):
+        elif isinstance(self.model, RegressorMixin):
             # This is the coefficient on the interaction term
             # TODO: THIS IS NOT YET CORRECT ?????
             self.causal_impact = (

causalpy/experiments/prepostfit.py

@@ -22,11 +22,11 @@
 import pandas as pd
 from matplotlib import pyplot as plt
 from patsy import build_design_matrices, dmatrices
+from sklearn.base import RegressorMixin
 
 from causalpy.custom_exceptions import BadIndexException
 from causalpy.plot_utils import plot_xY
 from causalpy.pymc_models import PyMCModel
-from causalpy.skl_models import ScikitLearnModel
 from causalpy.utils import round_num
 
 from .base import BaseExperiment
@@ -77,7 +77,7 @@ def __init__(
         if isinstance(self.model, PyMCModel):
             COORDS = {"coeffs": self.labels, "obs_indx": np.arange(self.pre_X.shape[0])}
             self.model.fit(X=self.pre_X, y=self.pre_y, coords=COORDS)
-        elif isinstance(self.model, ScikitLearnModel):
+        elif isinstance(self.model, RegressorMixin):
             self.model.fit(X=self.pre_X, y=self.pre_y)
         else:
             raise ValueError("Model type not recognized")

causalpy/experiments/prepostnegd.py

@@ -23,13 +23,13 @@
 import seaborn as sns
 from matplotlib import pyplot as plt
 from patsy import build_design_matrices, dmatrices
+from sklearn.base import RegressorMixin
 
 from causalpy.custom_exceptions import (
     DataException,
 )
 from causalpy.plot_utils import plot_xY
 from causalpy.pymc_models import PyMCModel
-from causalpy.skl_models import ScikitLearnModel
 from causalpy.utils import _is_variable_dummy_coded, round_num
 
 from .base import BaseExperiment
@@ -115,7 +115,7 @@ def __init__(
         if isinstance(self.model, PyMCModel):
             COORDS = {"coeffs": self.labels, "obs_indx": np.arange(self.X.shape[0])}
             self.model.fit(X=self.X, y=self.y, coords=COORDS)
-        elif isinstance(self.model, ScikitLearnModel):
+        elif isinstance(self.model, RegressorMixin):
             raise NotImplementedError("Not implemented for OLS model")
         else:
             raise ValueError("Model type not recognized")

causalpy/experiments/regression_discontinuity.py

@@ -22,14 +22,14 @@
 import seaborn as sns
 from matplotlib import pyplot as plt
 from patsy import build_design_matrices, dmatrices
+from sklearn.base import RegressorMixin
 
 from causalpy.custom_exceptions import (
     DataException,
     FormulaException,
 )
 from causalpy.plot_utils import plot_xY
 from causalpy.pymc_models import PyMCModel
-from causalpy.skl_models import ScikitLearnModel
 from causalpy.utils import _is_variable_dummy_coded, convert_to_string, round_num
 
 from .base import BaseExperiment
@@ -126,7 +126,7 @@ def __init__(
             # fit the model to the observed (pre-intervention) data
             COORDS = {"coeffs": self.labels, "obs_indx": np.arange(self.X.shape[0])}
             self.model.fit(X=self.X, y=self.y, coords=COORDS)
-        elif isinstance(self.model, ScikitLearnModel):
+        elif isinstance(self.model, RegressorMixin):
             self.model.fit(X=self.X, y=self.y)
         else:
             raise ValueError("Model type not recognized")

causalpy/skl_models.py

@@ -23,7 +23,7 @@
 from causalpy.utils import round_num
 
 
-class ScikitLearnModel:
+class ScikitLearnAdaptor:
     """Base class for scikit-learn models that can be used for causal inference."""
 
     def calculate_impact(self, y_true, y_pred):
@@ -53,7 +53,7 @@ def get_coeffs(self):
         return np.squeeze(self.coef_)
 
 
-class WeightedProportion(ScikitLearnModel, LinearModel, RegressorMixin):
+class WeightedProportion(ScikitLearnAdaptor, LinearModel, RegressorMixin):
     """Weighted proportion model for causal inference. Used for synthetic control
     methods for example"""
 
@@ -82,11 +82,19 @@ def predict(self, X):
 
 def create_causalpy_compatible_class(
     estimator: type[RegressorMixin],
-) -> type[ScikitLearnModel]:
+) -> type[RegressorMixin]:
     """This function takes a scikit-learn estimator and returns a new class that is
     compatible with CausalPy."""
-
-    class Model(ScikitLearnModel, estimator):
-        pass
-
-    return Model
+    _add_mixin_methods(estimator, ScikitLearnAdaptor)
+    return estimator
+
+
+def _add_mixin_methods(model_instance, mixin_class):
+    """Utility function to bind mixin methods to an existing model instance."""
+    for attr_name in dir(mixin_class):
+        attr = getattr(mixin_class, attr_name)
+        if callable(attr) and not attr_name.startswith("__"):
+            # Bind the method to the instance
+            method = attr.__get__(model_instance, model_instance.__class__)
+            setattr(model_instance, attr_name, method)
+    return model_instance

causalpy/tests/test_input_validation.py

@@ -23,7 +23,6 @@
 
 from sklearn.linear_model import LinearRegression
 
-CustomLinearRegression = cp.create_causalpy_compatible_class(LinearRegression)
 
 sample_kwargs = {"tune": 20, "draws": 20, "chains": 2, "cores": 2}
 
@@ -254,7 +253,7 @@ def test_rd_validation_treated_in_formula():
         _ = cp.RegressionDiscontinuity(
             df,
             formula="y ~ 1 + x",
-            model=CustomLinearRegression(),
+            model=LinearRegression(),
             treatment_threshold=0.5,
         )
 
@@ -281,7 +280,7 @@ def test_rd_validation_treated_is_dummy():
         _ = cp.RegressionDiscontinuity(
             df,
             formula="y ~ 1 + x + treated",
-            model=CustomLinearRegression(),
+            model=LinearRegression(),
             treatment_threshold=0.5,
         )
 

causalpy/tests/test_integration_skl_examples.py

@@ -20,9 +20,6 @@
 from sklearn.linear_model import LinearRegression
 
 import causalpy as cp
-from causalpy.skl_models import ScikitLearnModel
-
-CustomLinearRegression = cp.create_causalpy_compatible_class(LinearRegression)
 
 
 @pytest.mark.integration
@@ -42,7 +39,7 @@ def test_did():
         group_variable_name="group",
         treated=1,
         untreated=0,
-        model=CustomLinearRegression(),
+        model=LinearRegression(),
     )
     assert isinstance(data, pd.DataFrame)
     assert isinstance(result, cp.DifferenceInDifferences)
@@ -71,7 +68,7 @@ def test_rd_drinking():
         df,
         formula="all ~ 1 + age + treated",
         running_variable_name="age",
-        model=CustomLinearRegression(),
+        model=LinearRegression(),
         treatment_threshold=21,
         epsilon=0.001,
     )
@@ -103,7 +100,7 @@ def test_its():
         df,
         treatment_time,
         formula="y ~ 1 + t + C(month)",
-        model=CustomLinearRegression(),
+        model=LinearRegression(),
     )
     assert isinstance(df, pd.DataFrame)
     assert isinstance(result, cp.InterruptedTimeSeries)
@@ -165,7 +162,7 @@ def test_rd_linear_main_effects():
     result = cp.RegressionDiscontinuity(
         data,
         formula="y ~ 1 + x + treated",
-        model=CustomLinearRegression(),
+        model=LinearRegression(),
         treatment_threshold=0.5,
         epsilon=0.001,
     )
@@ -191,7 +188,7 @@ def test_rd_linear_main_effects_bandwidth():
     result = cp.skl_experiments.RegressionDiscontinuity(
         data,
         formula="y ~ 1 + x + treated",
-        model=CustomLinearRegression(),
+        model=LinearRegression(),
         treatment_threshold=0.5,
         epsilon=0.001,
         bandwidth=0.3,
@@ -217,7 +214,7 @@ def test_rd_linear_with_interaction():
     result = cp.RegressionDiscontinuity(
         data,
         formula="y ~ 1 + x + treated + x:treated",
-        model=CustomLinearRegression(),
+        model=LinearRegression(),
         treatment_threshold=0.5,
         epsilon=0.001,
     )
@@ -238,18 +235,13 @@ def test_rd_linear_with_gaussian_process():
     1. data is a dataframe
     2. skl_experiements.RegressionDiscontinuity returns correct type
     """
-
-    # create a custom GaussianProcessRegressor class by subclassing
-    # GaussianProcessRegressor and adding the ScikitLearnModel mixin
-    class CustomGaussianProcessRegressor(GaussianProcessRegressor, ScikitLearnModel):
-        pass
-
     data = cp.load_data("rd")
     kernel = 1.0 * ExpSineSquared(1.0, 5.0) + WhiteKernel(1e-1)
     result = cp.RegressionDiscontinuity(
         data,
         formula="y ~ 1 + x + treated",
-        model=CustomGaussianProcessRegressor(kernel=kernel),
+        model=GaussianProcessRegressor(kernel=kernel),
+        model_kwargs={"kernel": kernel},
         treatment_threshold=0.5,
         epsilon=0.001,
     )
@@ -275,7 +267,7 @@ def test_did_deprecation_warning():
             group_variable_name="group",
             treated=1,
             untreated=0,
-            model=CustomLinearRegression(),
+            model=LinearRegression(),
         )
         assert isinstance(result, cp.DifferenceInDifferences)
 
@@ -294,7 +286,7 @@ def test_its_deprecation_warning():
             df,
             treatment_time,
             formula="y ~ 1 + t + C(month)",
-            model=CustomLinearRegression(),
+            model=LinearRegression(),
         )
         assert isinstance(result, cp.InterruptedTimeSeries)
 
@@ -322,7 +314,7 @@ def test_rd_deprecation_warning():
         result = cp.skl_experiments.RegressionDiscontinuity(
             data,
             formula="y ~ 1 + x + treated",
-            model=CustomLinearRegression(),
+            model=LinearRegression(),
             treatment_threshold=0.5,
             epsilon=0.001,
         )