Fix bug inspect signature

Author: antoinedemathelin

adapt/base.py

@@ -265,7 +265,11 @@ def score_adapt(self, Xs, Xt, src_index=None):
         if src_index is None:
             src_index = np.arange(len(Xs))
         if hasattr(self, "transform"):
-            args = inspect.getfullargspec(self.transform).args
+            args = [
+                p.name
+                for p in inspect.signature(self.transform).parameters.values()
+                if p.name != "self" and p.kind != p.VAR_KEYWORD
+            ]
             if "domain" in args:
                 Xt = self.transform(Xt, domain="tgt")
                 Xs = self.transform(Xs, domain="src")
@@ -333,7 +337,11 @@ def _get_param_names(self):
 
     def _filter_params(self, func, override={}, prefix=""):
         kwargs = {}
-        args = inspect.getfullargspec(func).args
+        args = [
+            p.name
+            for p in inspect.signature(func).parameters.values()
+            if p.name != "self" and p.kind != p.VAR_KEYWORD
+        ]
         for key, value in self.__dict__.items():
             new_key = key.replace(prefix+"__", "")
             if new_key in args and prefix in key:
@@ -575,7 +583,12 @@ def fit_estimator(self, X, y, sample_weight=None,
 
         fit_params = self._filter_params(self.estimator_.fit, fit_params)
 
-        if "sample_weight" in inspect.getfullargspec(self.estimator_.fit).args:
+        fit_args = [
+            p.name
+            for p in inspect.signature(self.estimator_.fit).parameters.values()
+            if p.name != "self" and p.kind != p.VAR_KEYWORD
+        ]
+        if "sample_weight" in fit_args:
             sample_weight = check_sample_weight(sample_weight, X)
             with warnings.catch_warnings():
                 warnings.simplefilter("ignore")
@@ -718,7 +731,11 @@ def _get_legal_params(self, params):
 
         legal_params = ["domain", "val_sample_size"]
         for func in legal_params_fct:
-            args = list(inspect.getfullargspec(func).args)
+            args = [
+                p.name
+                for p in inspect.signature(func).parameters.values()
+                if p.name != "self" and p.kind != p.VAR_KEYWORD
+            ]
             legal_params = legal_params + args
 
         # Add kernel params for kernel based algorithm
@@ -1190,7 +1207,11 @@ def _get_legal_params(self, params):
 
         legal_params = ["domain", "val_sample_size"]
         for func in legal_params_fct:
-            args = list(inspect.getfullargspec(func).args)
+            args = [
+                p.name
+                for p in inspect.signature(func).parameters.values()
+                if p.name != "self" and p.kind != p.VAR_KEYWORD
+            ]
             legal_params = legal_params + args
 
         if "pretrain" in legal_params:
@@ -1200,7 +1221,11 @@ def _get_legal_params(self, params):
                     legal_params_fct.append(params["pretrain__optimizer"].__init__)
 
             for func in legal_params_fct:
-                args = list(inspect.getfullargspec(func).args)
+                args = [
+                    p.name
+                    for p in inspect.signature(func).parameters.values()
+                    if p.name != "self" and p.kind != p.VAR_KEYWORD
+                ]
                 legal_params = legal_params + ["pretrain__"+name for name in args]
         return legal_params
 

adapt/feature_based/_adda.py

@@ -288,7 +288,7 @@ def _initialize_weights(self, shape_X):
         self.encoder_(np.zeros((1,) + shape_X))
 
         # Set same weights to encoder_src
-        self.encoder_src_ = check_network(self.encoder,
+        self.encoder_src_ = check_network(self.encoder_,
                                           copy=True,
                                           name="encoder_src")
 

adapt/feature_based/_cdan.py

@@ -248,8 +248,6 @@ def train_step(self, data):
             disc_loss += sum(self.discriminator_.losses)
             enc_loss += sum(self.encoder_.losses)
 
-        print(task_loss.shape, enc_loss.shape, disc_loss.shape)
-            
         # Compute gradients
         trainable_vars_task = self.task_.trainable_variables
         trainable_vars_enc = self.encoder_.trainable_variables
@@ -295,8 +293,10 @@ def _initialize_weights(self, shape_X):
                                         self.max_features])
             self._random_enc = tf.random.normal([Xs_enc.get_shape()[1],
                                            self.max_features])
+            self.discriminator_(np.zeros((1, self.max_features)))
         else:
             self.is_overloaded_ = False
+            self.discriminator_(np.zeros((1, Xs_enc.get_shape()[1] * ys_pred.get_shape()[1])))
 
 
     def _initialize_networks(self):
@@ -318,6 +318,7 @@ def _initialize_networks(self):
             self.discriminator_ = check_network(self.discriminator,
                                                 copy=self.copy,
                                                 name="discriminator")
+        
 
 
     # def _initialize_networks(self, shape_Xt):

adapt/feature_based/_mcd.py

@@ -6,7 +6,7 @@
 import tensorflow as tf
 
 from adapt.base import BaseAdaptDeep, make_insert_doc
-from adapt.utils import check_network
+from adapt.utils import check_network, get_default_encoder, get_default_task
 
 EPS = np.finfo(np.float32).eps
 
@@ -27,10 +27,10 @@ class MCD(BaseAdaptDeep):
     
     Parameters
     ----------       
-    pretrain_steps : int (default=None)
-        Specify the number of pretraining of source encoder
-        and task networks. If `None` the number of pretrain
-        steps is equal to half the total number of training steps.
+    pretrain : bool (default=True)
+        Weither to pretrain the networks or not.
+        If True, the three networks are fitted on source
+        labeled data.
     
     Attributes
     ----------

adapt/feature_based/_mdd.py

@@ -6,7 +6,7 @@
 import tensorflow as tf
 
 from adapt.base import BaseAdaptDeep, make_insert_doc
-from adapt.utils import check_network
+from adapt.utils import check_network, get_default_encoder, get_default_task
 
 EPS = np.finfo(np.float32).eps
 

adapt/feature_based/_wdgrl.py

@@ -71,10 +71,6 @@ class WDGRL(BaseAdaptDeep):
         
     discriminator_ : tensorflow Model
         discriminator network.
-    
-    model_ : tensorflow Model
-        Fitted model: the union of ``encoder_``,
-        ``task_`` and ``discriminator_`` networks.
         
     history_ : dict
         history of the losses and metrics across the epochs.

adapt/instance_based/_wann.py

@@ -28,7 +28,12 @@ class WANN(BaseAdaptDeep):
           between the reweighted source and target distributions: the Y-discrepancy
     
     Parameters
-    ----------        
+    ----------
+    pretrain : bool (default=True)
+        Weither to perform pretraining of the ``weighter``
+        network or not. If True, the ``weighter`` is 
+        pretrained in order to predict 1 for each source.
+    
     C : float (default=1.)
         Clipping constant for the weighting networks
         regularization. Low value of ``C`` produce smoother
@@ -40,6 +45,7 @@ def __init__(self,
                  weighter=None,
                  Xt=None,
                  yt=None,
+                 pretrain=True,
                  C=1.,
                  verbose=1,
                  copy=True,

src_docs/examples/Two_moons.ipynb

@@ -16,7 +16,7 @@
 from sklearn.compose import TransformedTargetRegressor
 from sklearn.base import BaseEstimator, RegressorMixin, ClassifierMixin
 from sklearn.tree._tree import Tree
-from tensorflow.keras.wrappers.scikit_learn import KerasClassifier, KerasRegressor
+# from tensorflow.keras.wrappers.scikit_learn import KerasClassifier, KerasRegressor
 from tensorflow.keras import Model, Sequential
 from tensorflow.keras.layers import Input, Dense, Flatten, Reshape
 from tensorflow.python.keras.engine.input_layer import InputLayer
@@ -28,7 +28,7 @@ def is_equal_estimator(v1, v2):
     assert type(v2) == type(v1)
     if isinstance(v1, np.ndarray):
         assert np.array_equal(v1, v2)
-    elif isinstance(v1, (BaseEstimator, KerasClassifier, KerasRegressor)):
+    elif isinstance(v1, BaseEstimator):  # KerasClassifier, KerasRegressor
         assert is_equal_estimator(v1.__dict__, v2.__dict__)
     elif isinstance(v1, Model):
         assert is_equal_estimator(v1.get_config(),
@@ -198,7 +198,7 @@ def test_check_network_high_dataset():
     TransformedTargetRegressor(regressor=Ridge(alpha=25), transformer=StandardScaler()),
     MultiOutputRegressor(Ridge(alpha=0.3)),
     make_pipeline(StandardScaler(), Ridge(alpha=0.2)),
-    KerasClassifier(_get_model_Sequential, input_shape=(1,)),
+    # KerasClassifier(_get_model_Sequential, input_shape=(1,)),
     CustomEstimator()
 ]
 
@@ -212,10 +212,10 @@ def test_check_estimator_estimators(est):
     else:
         est.fit(np.linspace(0, 1, 10).reshape(-1, 1),
                 (np.linspace(0, 1, 10)<0.5).astype(float))
-    if isinstance(est, KerasClassifier):
-        new_est = check_estimator(est, copy=False)
-    else:
-        new_est = check_estimator(est, copy=True, force_copy=True)
+    # if isinstance(est, KerasClassifier):
+    #     new_est = check_estimator(est, copy=False)
+    # else:
+    new_est = check_estimator(est, copy=True, force_copy=True)
     assert is_equal_estimator(est, new_est)