Add GAMRegressor and GAMClassifier classes to sklearn_api.py

.gitignore

@@ -45,3 +45,6 @@ _build/
 # PyCharm
 #########
 .idea/
+.vscode/settings.json
+.env
+.vscode/launch.json

pygam/sklearn_api.py

@@ -0,0 +1,262 @@
+"""
+sklearn_api.py
+
+This module provides scikit-learn compatible classes for Generalized Additive Models (GAM) regressors and classifiers.
+It integrates pygam's GAM capabilities with scikit-learn's estimator interface, enabling seamless use in machine learning pipelines.
+"""
+
+# Standard library imports
+import numpy as np
+
+# Third-party imports
+from sklearn.base import BaseEstimator, RegressorMixin, ClassifierMixin
+from sklearn.datasets import make_regression
+from sklearn.model_selection import train_test_split
+from sklearn.metrics import accuracy_score
+
+# Local application imports
+from pygam import GAM
+from pygam.terms import te, TermList, Term  # Import te for interactions
+from pygam.terms import s, f, l, intercept  # Import s, f, l for splines
+
+
+def create_default_terms(X, categorical_features=None):
+    """Generate default terms for each feature in X, handling categoricals."""
+    n_features = X.shape[1]
+    terms = []
+    if categorical_features is None:
+        categorical_features = []
+    for i in range(n_features):
+        if i in categorical_features:
+            terms.append(f(i))
+        elif not np.issubdtype(X[:, i].dtype, np.number):
+            terms.append(f(i))
+        else:
+            terms.append(s(i))
+    return terms
+
+
+class GAMRegressor(BaseEstimator, RegressorMixin):
+    """
+    GAMRegressor
+
+    A scikit-learn compatible regressor using Generalized Additive Models (GAM).
+
+    Parameters
+    ----------
+    distribution : str, default='normal'
+        The distribution of the response variable.
+    link : str, default='identity'
+        The link function.
+    terms : 'auto', None, or list of Term objects, default='auto'
+        The terms to include in the model. If 'auto', terms are automatically inferred based on X.
+        If None, no terms are used. If a list of Term objects, they are used as specified.
+    interactions : None or list of Term objects, optional
+        Interaction terms to include in the model.
+    callbacks : list, default=['deviance', 'diffs']
+        List of callbacks to monitor during training.
+    fit_intercept : bool, default=True
+        Whether to fit an intercept.
+    max_iter : int, default=100
+        Maximum number of iterations.
+    tol : float, default=1e-4
+        Tolerance for stopping criteria.
+    verbose : bool, default=False
+        Verbosity mode.
+    categorical_features : list, optional
+        List of indices of categorical features.
+    **gam_params :
+        Additional parameters for the GAM model.
+
+    Attributes
+    ----------
+    model_ : GAM
+        The underlying pygam GAM model fitted to the data.
+    """
+
+    def __init__(
+        self,
+        distribution='normal',
+        link='identity',
+        terms='auto',
+        interactions=None,
+        callbacks=['deviance', 'diffs'],
+        fit_intercept=True,
+        max_iter=100,
+        tol=1e-4,
+        verbose=False,
+        categorical_features=None,
+        **gam_params,
+    ):
+        self.distribution = distribution
+        self.link = link
+        self.terms = terms
+        self.interactions = interactions
+        self.callbacks = callbacks
+        self.fit_intercept = fit_intercept
+        self.max_iter = max_iter
+        self.tol = tol
+        self.verbose = verbose
+        self.categorical_features = categorical_features
+        self.gam_params = gam_params
+
+    def fit(self, X, y):
+        if self.terms == 'auto':
+            self.terms_ = create_default_terms(X, self.categorical_features)
+        elif self.terms is None:
+            self.terms_ = []
+        else:
+            self.terms_ = self.terms
+
+        if self.interactions is not None:
+            self.interactions_ = [
+                te(*interaction) if isinstance(interaction, tuple) else interaction
+                for interaction in self.interactions
+            ]
+        else:
+            self.interactions_ = []
+
+        # Combine terms and interactions
+        terms = self.terms_ + self.interactions_
+        terms = TermList(*terms)
+
+        # Create the GAM model with the specified terms
+        self.model_ = GAM(
+            distribution=self.distribution,
+            link=self.link,
+            terms=terms,
+            callbacks=self.callbacks,
+            fit_intercept=self.fit_intercept,
+            max_iter=self.max_iter,
+            tol=self.tol,
+            verbose=self.verbose,
+            **self.gam_params,
+        )
+        self.model_.fit(X, y)
+        return self
+
+    def predict(self, X):
+        return self.model_.predict(X)
+
+    def score(self, X, y):
+        return float(self.model_.statistics_.get('pseudo R-squared', 0))
+
+
+class GAMClassifier(BaseEstimator, ClassifierMixin):
+    """
+    GAMClassifier
+
+    A scikit-learn compatible classifier using Generalized Additive Models (GAM).
+
+    Parameters
+    ----------
+    distribution : str, default='binomial'
+        The distribution of the response variable.
+    link : str, default='logit'
+        The link function.
+    terms : 'auto', None, or list of Term objects, default='auto'
+        The terms to include in the model. If 'auto', terms are automatically inferred based on X.
+        If None, no terms are used. If a list of Term objects, they are used as specified.
+    interactions : None or list of Term objects, optional
+        Interaction terms to include in the model.
+    callbacks : list, default=['deviance', 'diffs', 'accuracy']
+        List of callbacks to monitor during training.
+    fit_intercept : bool, default=True
+        Whether to fit an intercept.
+    max_iter : int, default=100
+        Maximum number of iterations.
+    tol : float, default=1e-4
+        Tolerance for stopping criteria.
+    verbose : bool, default=False
+        Verbosity mode.
+    categorical_features : list, optional
+        List of indices of categorical features.
+    **gam_params :
+        Additional parameters for the GAM model.
+
+    Attributes
+    ----------
+    model_ : GAM
+        The underlying pygam GAM model fitted to the data.
+    classes_ : array-like
+        Unique class labels.
+    """
+
+    def __init__(
+        self,
+        distribution='binomial',
+        link='logit',
+        terms='auto',
+        interactions=None,
+        callbacks=['deviance', 'diffs', 'accuracy'],
+        fit_intercept=True,
+        max_iter=100,
+        tol=1e-4,
+        verbose=False,
+        categorical_features=None,
+        **gam_params,
+    ):
+        self.distribution = distribution
+        self.link = link
+        self.terms = terms
+        self.interactions = interactions
+        self.callbacks = callbacks
+        self.fit_intercept = fit_intercept
+        self.max_iter = max_iter
+        self.tol = tol
+        self.verbose = verbose
+        self.categorical_features = categorical_features
+        self.gam_params = gam_params
+
+    def fit(self, X, y):
+        if self.terms == 'auto':
+            self.terms_ = create_default_terms(X, self.categorical_features)
+        elif self.terms is None:
+            self.terms_ = []
+        else:
+            self.terms_ = self.terms
+
+        if self.interactions is not None:
+            self.interactions_ = [
+                te(*interaction) if isinstance(interaction, tuple) else interaction
+                for interaction in self.interactions
+            ]
+        else:
+            self.interactions_ = []
+
+        # Combine terms and interactions
+        terms = self.terms_ + self.interactions_
+        terms = TermList(*terms)
+
+        # Create the GAM model with the specified terms
+        self.model_ = GAM(
+            distribution=self.distribution,
+            link=self.link,
+            terms=terms,
+            callbacks=self.callbacks,
+            fit_intercept=self.fit_intercept,
+            max_iter=self.max_iter,
+            tol=self.tol,
+            verbose=self.verbose,
+            **self.gam_params,
+        )
+        self.model_.fit(X, y)
+        self.classes_ = np.unique(y)
+        return self
+
+    def predict(self, X):
+        proba = self.model_.predict(X)
+        if len(self.classes_) == 2:
+            return (proba >= 0.5).astype(int)
+        else:
+            return self.classes_[np.argmax(proba, axis=1)]
+
+    def predict_proba(self, X):
+        proba = self.model_.predict(X)
+        if len(self.classes_) == 2:
+            return np.vstack([1 - proba, proba]).T
+        else:
+            return proba  # Assume GAM model returns probabilities for each class
+
+    def score(self, X, y):
+        return accuracy_score(y, self.predict(X))

pygam/tests/test_penalties.py

@@ -23,13 +23,13 @@ def test_single_spline_penalty():
     monotonic_ and convexity_ should be 0.
     """
     coef = np.array(1.0)
-    assert np.alltrue(derivative(1, coef).A == 0.0)
-    assert np.alltrue(l2(1, coef).A == 1.0)
-    assert np.alltrue(monotonic_inc(1, coef).A == 0.0)
-    assert np.alltrue(monotonic_dec(1, coef).A == 0.0)
-    assert np.alltrue(convex(1, coef).A == 0.0)
-    assert np.alltrue(concave(1, coef).A == 0.0)
-    assert np.alltrue(none(1, coef).A == 0.0)
+    assert np.all(derivative(1, coef).A == 0.0)
+    assert np.all(l2(1, coef).A == 1.0)
+    assert np.all(monotonic_inc(1, coef).A == 0.0)
+    assert np.all(monotonic_dec(1, coef).A == 0.0)
+    assert np.all(convex(1, coef).A == 0.0)
+    assert np.all(concave(1, coef).A == 0.0)
+    assert np.all(none(1, coef).A == 0.0)
 
 
 def test_wrap_penalty():