Merge pull request #9 from antoinedemathelin/master

fix: coral, kliep, GradientHandler

Author: antoinedemathelin

adapt/feature_based/_coral.py

@@ -196,8 +196,17 @@ def fit_embeddings(self, Xs, Xt):
 
         self.Cs_ = self.lambda_ * cov_Xs + np.eye(Xs.shape[1])
         self.Ct_ = self.lambda_ * cov_Xt + np.eye(Xt.shape[1])
-        Xs_emb = np.matmul(Xs, linalg.inv(linalg.sqrtm(self.Cs_)))
-        Xs_emb = np.matmul(Xs_emb, linalg.sqrtm(self.Ct_))
+        
+        Cs_sqrt_inv = linalg.inv(linalg.sqrtm(self.Cs_))
+        Ct_sqrt = linalg.sqrtm(self.Ct_)
+        
+        if np.iscomplexobj(Cs_sqrt_inv):
+            Cs_sqrt_inv = Cs_sqrt_inv.real
+        if np.iscomplexobj(Ct_sqrt):
+            Ct_sqrt = Ct_sqrt.real
+        
+        Xs_emb = np.matmul(Xs, Cs_sqrt_inv)
+        Xs_emb = np.matmul(Xs_emb, Ct_sqrt)
 
         if self.verbose:
             new_cov_Xs = np.cov(Xs_emb, rowvar=False)
@@ -280,8 +289,16 @@ def predict_features(self, X, domain="tgt"):
         if domain in ["tgt", "target"]:
             X_emb = X
         elif domain in ["src", "source"]:
-            X_emb = np.matmul(X, linalg.inv(linalg.sqrtm(self.Cs_)))
-            X_emb = np.matmul(X_emb, linalg.sqrtm(self.Ct_))
+            Cs_sqrt_inv = linalg.inv(linalg.sqrtm(self.Cs_))
+            Ct_sqrt = linalg.sqrtm(self.Ct_)
+
+            if np.iscomplexobj(Cs_sqrt_inv):
+                Cs_sqrt_inv = Cs_sqrt_inv.real
+            if np.iscomplexobj(Ct_sqrt):
+                Ct_sqrt = Ct_sqrt.real
+            
+            X_emb = np.matmul(X, Cs_sqrt_inv)
+            X_emb = np.matmul(X_emb, Ct_sqrt)
         else:
             raise ValueError("`domain `argument "
                              "should be `tgt` or `src`, "

adapt/feature_based/_mdd.py

@@ -73,7 +73,7 @@ class MDD(BaseDeepFeature):
     task_ : tensorflow Model
         Principal task network.
         
-    task_adv_ : tensorflow Model
+    discriminator_ : tensorflow Model
         Adversarial task network.
     
     model_ : tensorflow Model

adapt/instance_based/_kliep.py

@@ -288,6 +288,9 @@ def fit_estimator(self, X, y,
         else:
             if sample_weight is not None:
                 sample_weight /= (sample_weight.sum() + EPS)
+            if sample_weight.sum() <= 0:
+                sample_weight = np.ones(len(sample_weight))
+                sample_weight /= (sample_weight.sum() + EPS)
             bootstrap_index = np.random.choice(
             len(X), size=len(X), replace=True,
             p=sample_weight)
@@ -359,7 +362,7 @@ def _fit(self, Xs, Xt, sigma):
             alpha += b * ((((1-np.dot(np.transpose(b), alpha)) /
                             (np.dot(np.transpose(b), b) + EPS))))
             alpha = np.maximum(0, alpha)
-            alpha /= np.dot(np.transpose(b), alpha)
+            alpha /= (np.dot(np.transpose(b), alpha) + EPS)
             objective = np.mean(np.log(np.dot(A, alpha) + EPS))
             k += 1
 
@@ -389,7 +392,7 @@ def _cross_val_jscore(self, Xs, Xt, sigma, cv):
                                         centers,
                                         sigma),
                     alphas
-                )
+                ) + EPS
             ))
             cv_scores.append(j_score)
         return cv_scores

adapt/utils.py

@@ -364,6 +364,7 @@ class GradientHandler(Layer):
     """
     def __init__(self, lambda_init=1., name="g_handler"):
         super().__init__(name=name)
+        self.lambda_init=lambda_init
         self.lambda_ = tf.Variable(lambda_init,
                                    trainable=False,
                                    dtype="float32")
@@ -384,6 +385,21 @@ def call(self, x):
         return _grad_handler(x, self.lambda_)
 
 
+    def get_config(self):
+        """
+        Return config dictionnary.
+        
+        Returns
+        -------
+        dict
+        """
+        config = super().get_config().copy()
+        config.update({
+            'lambda_init': self.lambda_init
+        })
+        return config
+
+
 def make_classification_da(n_samples=100, 
                            n_features=2,
                            random_state=2):

src_docs/_templates/index.html

@@ -135,18 +135,18 @@ <h2> Examples </h2>
 		<div class="three-cards">
 		  <div class="card h-100">
 			<div class="card-body">
-			  <a href="{{ pathto("examples/Two_Moons") }}"><h4 class="card-title">Two Moons</h4></a>
+			  <a href="{{ pathto("examples/Two_moons") }}"><h4 class="card-title">Two Moons</h4></a>
 			  <p class="card-text">Here are some examples of domain adaptation methods applied on the Two Moons dataset.
 			  Please look at these examples to learn how to use the algorithms provided by the ADAPT package.
 			  </p>
 			</div>
 
 			<div class="image">
-			  <a href="{{ pathto("examples/Two_Moons") }}">
+			  <a href="{{ pathto("examples/Two_moons") }}">
 			  <img src="_static/images/two_moons_setup.png" class="sk-index-img" alt="Two Moons">
 			  </a>
 			</div>
-			  <a class="btn btn-below" role="button" href="{{ pathto("examples/Two_Moons") }}">See examples</a>
+			  <a class="btn btn-below" role="button" href="{{ pathto("examples/Two_moons") }}">See examples</a>
 		  </div>
 	  </div>
 

tests/test_coral.py

@@ -4,6 +4,7 @@
 
 import numpy as np
 from sklearn.linear_model import LogisticRegression
+from scipy import linalg
 import tensorflow as tf
 from tensorflow.keras import Sequential, Model
 from tensorflow.keras.layers import Dense
@@ -62,6 +63,17 @@ def test_fit_coral():
     assert np.abs(model.estimator_.coef_[0][0] /
            model.estimator_.coef_[0][1] + 0.5) < 0.1
     assert (model.predict(Xt) == yt).sum() / len(Xt) >= 0.99
+    
+    
+def test_fit_coral_complex():
+    np.random.seed(0)
+    model = CORAL(LogisticRegression(), lambda_=10000.)
+    Xs_ = np.random.randn(10, 100)
+    Xt_ = np.random.randn(10, 100)
+    model.fit(Xs_, ys[:10], Xt_)
+    assert np.iscomplexobj(linalg.inv(linalg.sqrtm(model.Cs_)))
+    assert np.iscomplexobj(linalg.sqrtm(model.Ct_))
+    model.predict(Xs_, domain="src")
 
 
 def test_fit_deepcoral():

tests/test_kliep.py

@@ -4,9 +4,21 @@
 
 import numpy as np
 from sklearn.linear_model import LinearRegression
+from sklearn.base import BaseEstimator
 
 from adapt.instance_based import KLIEP
 
+
+class DummyEstimator(BaseEstimator):
+    
+    def __init__(self):
+        pass
+    
+    def fit(self, X, y):
+        self.y = y
+        return self
+
+
 np.random.seed(0)
 Xs = np.concatenate((
     np.random.randn(50)*0.1,
@@ -34,3 +46,21 @@ def test_fit():
     assert model.weights_[:50].sum() > 90
     assert model.weights_[50:].sum() < 0.5
     assert np.abs(model.predict(Xt) - yt).sum() < 20
+    
+
+def test_fit_estimator_bootstrap_index():
+    np.random.seed(0)
+    ys_ = np.random.randn(100)
+    model = KLIEP(DummyEstimator(),
+                  sigmas=[10, 100])
+    model.fit_estimator(Xs, ys_, sample_weight=np.random.random(len(ys)))
+    assert len(set(list(model.estimator_.y.ravel())) & set(list(ys_.ravel()))) > 33
+    
+    
+def test_fit_estimator_sample_weight_zeros():
+    np.random.seed(0)
+    ys_ = np.random.randn(100)
+    model = KLIEP(DummyEstimator(),
+                  sigmas=[10, 100])
+    model.fit_estimator(Xs, ys_, sample_weight=np.zeros(len(ys)))
+    assert len(set(list(model.estimator_.y.ravel())) & set(list(ys_.ravel()))) > 33

tests/test_utils.py

@@ -433,6 +433,9 @@ def test_gradienthandler(lambda_):
         gradient = tape.gradient(grad_handler(inputs),
                                  inputs)
     assert np.all(gradient == lambda_ * np.ones(3))
+    config = grad_handler.get_config()
+    assert config['lambda_init'] == lambda_
+    
 
 
 def test_make_classification_da():