legacy optimizer 4

Author: antoinedemathelin

adapt/base.py

@@ -14,6 +14,11 @@
 from sklearn.exceptions import NotFittedError
 from tensorflow.keras import Model
 from tensorflow.keras.wrappers.scikit_learn import KerasClassifier, KerasRegressor
+try:
+    from tensorflow.keras.optimizers.legacy import RMSprop
+except:
+    from tensorflow.keras.optimizers import RMSprop
+
 
 from adapt.utils import (check_estimator,
                          check_network,
@@ -1308,7 +1313,7 @@ def compile(self,
 
         if ((not "optimizer" in compile_params) or 
             (compile_params["optimizer"] is None)):
-            compile_params["optimizer"] = "rmsprop"
+            compile_params["optimizer"] = RMSprop()
         else:
             if optimizer is None:
                 if not isinstance(compile_params["optimizer"], str):

tests/test_ccsa.py

@@ -4,6 +4,10 @@
 from adapt.utils import make_classification_da
 from adapt.feature_based import CCSA
 from tensorflow.keras.initializers import GlorotUniform
+try:
+    from tensorflow.keras.optimizers.legacy import Adam
+except:
+    from tensorflow.keras.optimizers import Adam
 
 np.random.seed(0)
 tf.random.set_seed(0)
@@ -18,13 +22,13 @@
 
 def test_ccsa():
     ccsa = CCSA(task=task, loss="categorical_crossentropy",
-                optimizer="adam", metrics=["acc"], gamma=0.1, random_state=0)
+                optimizer=Adam(), metrics=["acc"], gamma=0.1, random_state=0)
     ccsa.fit(Xs, tf.one_hot(ys, 2).numpy(), Xt=Xt[ind],
              yt=tf.one_hot(yt, 2).numpy()[ind], epochs=100, verbose=0)
     assert np.mean(ccsa.predict(Xt).argmax(1) == yt) > 0.8
 
     ccsa = CCSA(task=task, loss="categorical_crossentropy",
-                optimizer="adam", metrics=["acc"], gamma=1., random_state=0)
+                optimizer=Adam(), metrics=["acc"], gamma=1., random_state=0)
     ccsa.fit(Xs, tf.one_hot(ys, 2).numpy(), Xt=Xt[ind],
              yt=tf.one_hot(yt, 2).numpy()[ind], epochs=100, verbose=0)
 

tests/test_finetuning.py

@@ -5,6 +5,10 @@
 from adapt.utils import make_classification_da
 from adapt.parameter_based import FineTuning
 from tensorflow.keras.initializers import GlorotUniform
+try:
+    from tensorflow.keras.optimizers.legacy import Adam
+except:
+    from tensorflow.keras.optimizers import Adam
 
 np.random.seed(0)
 tf.random.set_seed(0)
@@ -21,14 +25,14 @@
 
 
 def test_finetune():
-    model = FineTuning(encoder=encoder, task=task, loss="bce", optimizer="adam", random_state=0)
+    model = FineTuning(encoder=encoder, task=task, loss="bce", optimizer=Adam(), random_state=0)
     model.fit(Xs, ys, epochs=100, verbose=0)
 
     assert np.mean((model.predict(Xt).ravel()>0.5) == yt) < 0.7
 
     fine_tuned = FineTuning(encoder=model.encoder_, task=model.task_,
                             training=False,
-                            loss="bce", optimizer="adam", random_state=0)
+                            loss="bce", optimizer=Adam(), random_state=0)
     fine_tuned.fit(Xt[ind], yt[ind], epochs=100, verbose=0)
 
     assert np.abs(fine_tuned.encoder_.get_weights()[0] - model.encoder_.get_weights()[0]).sum() == 0.

tests/test_tradaboost.py

@@ -8,6 +8,10 @@
 from sklearn.linear_model import LinearRegression, LogisticRegression, Ridge, RidgeClassifier
 from sklearn.metrics import r2_score, accuracy_score
 import tensorflow as tf
+try:
+    from tensorflow.keras.optimizers.legacy import Adam
+except:
+    from tensorflow.keras.optimizers import Adam
 
 from adapt.instance_based import (TrAdaBoost,
                                   TrAdaBoostR2,
@@ -50,14 +54,14 @@ def test_tradaboost_fit_keras_model():
     np.random.seed(0)
     est = tf.keras.Sequential()
     est.add(tf.keras.layers.Dense(1, activation="sigmoid"))
-    est.compile(loss="bce", optimizer="adam")
+    est.compile(loss="bce", optimizer=Adam())
     model = TrAdaBoost(est, n_estimators=2, random_state=0)
     model.fit(Xs, ys_classif, Xt=Xt[:10], yt=yt_classif[:10])
     yp = model.predict(Xt)
 
     est = tf.keras.Sequential()
     est.add(tf.keras.layers.Dense(2, activation="softmax"))
-    est.compile(loss="mse", optimizer="adam")
+    est.compile(loss="mse", optimizer=Adam())
     model = TrAdaBoost(est, n_estimators=2, random_state=0)
     model.fit(Xs, np.random.random((100, 2)),
               Xt=Xt[:10], yt=np.random.random((10, 2)))