Merge pull request #78 from antoinedemathelin/master

feat: Add LinInt

Author: antoinedemathelin

adapt/feature_based/_pred.py

@@ -75,7 +75,7 @@ def __init__(self,
         super().__init__(**kwargs)
 
 
-    def fit_transform(self, Xs, Xt, ys, yt, domains=None, **kwargs):
+    def fit_transform(self, Xs, Xt, ys, yt, **kwargs):
         """
         Fit embeddings.
         

adapt/parameter_based/__init__.py

@@ -6,10 +6,12 @@
 from ._finetuning import FineTuning
 from ._transfer_tree import TransferTreeClassifier
 from ._transfer_tree import TransferForestClassifier
+from ._linint import LinInt
 
 __all__ = ["RegularTransferLR",
            "RegularTransferLC",
            "RegularTransferNN",
            "FineTuning",
            "TransferTreeClassifier",
-           "TransferForestClassifier"]
+           "TransferForestClassifier",
+           "LinInt"]

adapt/parameter_based/_linint.py

@@ -0,0 +1,190 @@
+"""
+Frustratingly Easy Domain Adaptation module.
+"""
+
+import warnings
+
+import numpy as np
+from sklearn.utils import check_array
+from sklearn.exceptions import NotFittedError
+from sklearn.linear_model import LinearRegression
+from sklearn.metrics import r2_score
+
+from adapt.base import BaseAdaptEstimator, make_insert_doc
+from adapt.utils import check_arrays, set_random_seed
+
+
+@make_insert_doc(supervised=True)
+class LinInt(BaseAdaptEstimator):
+    """
+    LinInt: Linear Interpolation between SrcOnly and TgtOnly.
+    
+    LinInt linearly interpolates the predictions of the SrcOnly and
+    TgtOnly models. The interpolation parameter is adjusted based on
+    a small amount of target data removed from the training set
+    of TgtOnly.
+
+    Parameters
+    ----------
+    prop : float (default=0.5)
+        Proportion between 0 and 1 of the data used
+        to fit the TgtOnly model. The rest of the 
+        target data are used to estimate the interpolation
+        parameter.
+
+    Attributes
+    ----------
+    estimator_src_ : object
+        Fitted source estimator.
+    
+    estimator_ : object
+        Fitted estimator.
+          
+    See also
+    --------
+    FA
+    PRED
+        
+    Examples
+    --------
+    >>> from sklearn.linear_model import Ridge
+    >>> from adapt.utils import make_regression_da
+    >>> from adapt.parameter_based import LinInt
+    >>> Xs, ys, Xt, yt = make_regression_da()
+    >>> model = LinInt(Ridge(), Xt=Xt[:6], yt=yt[:6], prop=0.5,            
+    ...              verbose=0, random_state=0)
+    >>> model.fit(Xs, ys)
+    >>> model.score(Xt, yt)
+    0.68...
+
+    References
+    ----------
+    .. [1] `[1] <https://arxiv.org/pdf/0907.1815\
+.pdf>`_ Daume III, H. "Frustratingly easy domain adaptation". In ACL, 2007.
+    """
+    def __init__(self,
+                 estimator=None,
+                 Xt=None,
+                 yt=None,
+                 prop=0.5,
+                 copy=True,
+                 verbose=1,
+                 random_state=None,
+                 **params):
+        
+        names = self._get_param_names()
+        kwargs = {k: v for k, v in locals().items() if k in names}
+        kwargs.update(params)
+        super().__init__(**kwargs)
+
+
+    def fit(self, Xs, ys, Xt=None, yt=None, **kwargs):
+        """
+        Fit LinInt.
+        
+        Parameters
+        ----------
+        Xs : array
+            Source input data.
+            
+        ys : array
+            Source output data.
+            
+        Xt : array
+            Target input data.
+            
+        yt : array
+            Target output data.
+        
+        kwargs : key, value argument
+            Not used, present here for adapt consistency.
+        
+        Returns
+        -------
+        Xt_aug, yt : augmented input and output target data
+        """        
+        set_random_seed(self.random_state)
+        
+        Xs, ys = check_arrays(Xs, ys, accept_sparse=True)
+        Xt, yt = self._get_target_data(Xt, yt)
+        Xt, yt = check_arrays(Xt, yt, accept_sparse=True)
+        
+        shuffle_index = np.random.choice(len(Xt), len(Xt), replace=False)
+        cut = int(len(Xt)*self.prop)
+        Xt_train = Xt[shuffle_index[:cut]]
+        Xt_test = Xt[shuffle_index[cut:]]
+        yt_train = yt[shuffle_index[:cut]]
+        yt_test = yt[shuffle_index[cut:]]
+        
+        self.estimator_src_ = self.fit_estimator(Xs, ys,
+                                            warm_start=False,
+                                            random_state=None)
+        
+        self.estimator_ = self.fit_estimator(Xt_train, yt_train,
+                                            warm_start=False,
+                                            random_state=None)
+            
+        self.interpolator_ = LinearRegression(fit_intercept=False)
+        
+        yp_src = self.estimator_src_.predict(Xt_test)
+        yp_tgt = self.estimator_.predict(Xt_test)
+        
+        if len(yp_src.shape) < 2:
+            yp_src = yp_src.reshape(-1, 1)
+        if len(yp_tgt.shape) < 2:
+            yp_tgt = yp_tgt.reshape(-1, 1)
+        
+        Xp = np.concatenate((yp_src, yp_tgt), axis=1)
+        
+        self.interpolator_.fit(Xp, yt_test)
+        
+        return self
+
+
+    def predict(self, X):
+        """
+        Return LinInt predictions.
+        
+        Parameters
+        ----------
+        X : array
+            Input data.
+
+        Returns
+        -------
+        y : array
+            Predictions
+        """
+        yp_src = self.estimator_src_.predict(X)
+        yp_tgt = self.estimator_.predict(X)
+        
+        if len(yp_src.shape) < 2:
+            yp_src = yp_src.reshape(-1, 1)
+        if len(yp_tgt.shape) < 2:
+            yp_tgt = yp_tgt.reshape(-1, 1)
+        
+        Xp = np.concatenate((yp_src, yp_tgt), axis=1)
+        
+        return self.interpolator_.predict(Xp)
+    
+    
+    def score(self, X, y):
+        """
+        Compute R2 score
+        
+        Parameters
+        ----------
+        X : array
+            input data
+            
+        y : array
+            output data
+            
+        Returns
+        -------
+        score : float
+            estimator score.
+        """
+        yp = self.predict(X)
+        score = r2_score(y, yp)
+        return score

tests/test_linint.py

@@ -0,0 +1,13 @@
+from sklearn.linear_model import Ridge
+from adapt.utils import make_regression_da
+from adapt.parameter_based import LinInt
+
+Xs, ys, Xt, yt = make_regression_da()
+
+def test_linint():
+    model = LinInt(Ridge(), Xt=Xt[:6], yt=yt[:6],              
+                 verbose=0, random_state=0)
+    model.fit(Xs, ys)
+    model.fit(Xs, ys, Xt[:6], yt[:6])
+    model.predict(Xt)
+    model.score(Xt, yt)