Fixed keras has no attribute get_session

alibi/explainers/cfproto.py

@@ -5,6 +5,7 @@
 
 import numpy as np
 import tensorflow.compat.v1 as tf
+import tensorflow.python.keras.backend as K
 
 from alibi.api.defaults import DEFAULT_DATA_CFP, DEFAULT_META_CFP
 from alibi.api.interfaces import Explainer, Explanation, FitMixin
@@ -120,7 +121,7 @@ def __init__(self,
 
         # check if the passed object is a model and get session
         is_model = isinstance(predict, tf.keras.Model)
-        model_sess = tf.compat.v1.keras.backend.get_session()
+        model_sess = K.get_session()
         is_ae = isinstance(ae_model, tf.keras.Model)
         is_enc = isinstance(enc_model, tf.keras.Model)
         self.meta['params'].update(is_model=is_model, is_ae=is_ae, is_enc=is_enc)

alibi/explainers/counterfactual.py

@@ -4,6 +4,7 @@
 
 import numpy as np
 import tensorflow.compat.v1 as tf
+import tensorflow.python.keras.backend as K
 
 from alibi.api.defaults import DEFAULT_DATA_CF, DEFAULT_META_CF
 from alibi.api.interfaces import Explainer, Explanation
@@ -167,7 +168,7 @@ def __init__(self,
 
         # check if the passed object is a model and get session
         is_model = isinstance(predict_fn, tf.keras.Model)
-        model_sess = tf.compat.v1.keras.backend.get_session()
+        model_sess = K.get_session()
 
         self.meta['params'].update(is_model=is_model)
 

alibi/tests/test_counterfactual_proto.py

@@ -0,0 +1,56 @@
+# alibi/tests/test_counterfactual_proto.py
+import os
+os.environ["TF_CPP_MIN_LOG_LEVEL"] = "2"
+
+import numpy as np
+import tensorflow.compat.v1 as tf
+tf.disable_v2_behavior()
+
+import tensorflow.python.keras.backend as K
+
+from alibi.api.interfaces import Explanation
+from alibi.explainers.cfproto import CounterfactualProto
+
+
+def test_cfproto_uses_k_session_blackbox():
+    tf.reset_default_graph()
+    sess = tf.Session()
+    K.set_session(sess)
+    K.set_learning_phase(0)
+
+    with sess.as_default():
+        # Simple TF1 graph: softmax(Wx+b)
+        x_ph = tf.placeholder(tf.float32, shape=(None, 4), name="x")
+        W = tf.get_variable("W", shape=(4, 2),
+                            initializer=tf.random_normal_initializer(stddev=0.1))
+        b = tf.get_variable("b", shape=(2,),
+                            initializer=tf.zeros_initializer())
+        logits = tf.matmul(x_ph, W) + b
+        probs = tf.nn.softmax(logits)
+
+        sess.run(tf.global_variables_initializer())
+
+        def predict_fn(x: np.ndarray) -> np.ndarray:
+            return sess.run(probs, feed_dict={x_ph: x})
+
+        explainer = CounterfactualProto(
+            predict=predict_fn,
+            shape=(1, 4),
+            max_iterations=5,
+            c_steps=1,
+            c_init=0.0,
+            kappa=0.0,
+            beta=0.1,
+            gamma=0.0,
+            use_kdtree=False,
+        )
+
+        assert explainer.sess is K.get_session()
+
+        x0 = np.zeros((1, 4), dtype="float32")
+        explanation = explainer.explain(x0)
+
+        assert isinstance(explanation, Explanation)
+        assert "orig_class" in explanation.data
+        if explanation.data.get("cf") and "X" in explanation.data["cf"]:
+            assert explanation.data["cf"]["X"].shape == x0.shape