changes bernoulli

lds_gnn/lds.py

@@ -152,7 +152,7 @@
 import gcn.metrics
 from sklearn.neighbors import kneighbors_graph
 import tensorflow_probability as tfp
-from tfp.distributions import Bernoulli, Normal
+#from tfp.distributions import Bernoulli, Normal
 
 try:
     from lds_gnn.data import ConfigData, UCI, EdgeDelConfigData
@@ -271,10 +271,10 @@ def noisy_features(features, mask, noise, ogb):
         else:
             print("Invalid noise input encountered while adding noise to features")
     else:
-        print(features)
-        print(features.shape)
-        print(mask)
-        print(mask.shape)
+        #print(features)
+        #print(features.shape)
+        #print(mask)
+        #print(type(mask))
         masked_features = features * (1 - mask)
     
     return masked_features
@@ -302,7 +302,6 @@ def lds(data_conf: ConfigData, config: LDSConfig):
     #nones = torch.sum(features > 0.0).float()
     nones = tf.math.reduce_sum(tf.cast(features > 0.0, dtype=tf.float32))
 
-    print(features.shape)
     nzeros = features.shape[0] * features.shape[1] - nones
     pzeros = nones / nzeros / r * nr
 
@@ -317,8 +316,15 @@ def lds(data_conf: ConfigData, config: LDSConfig):
     #probs[features > 0.0] = 1 / r
     b_broadcast = tf.ones(tf.shape(features), dtype=features.dtype) * (1 / r)
     features = tf.where(tf.greater(features, 0.0), b_broadcast, features)
-
-    mask = Bernoulli(probs)
+  
+    print("the shape of probs, temp are:")
+
+    mask = tfp.distributions.Bernoulli(probs).sample()
+    print(mask.shape)
+    #mask = Bernoulli(probs) 
+      
+    mask = tf.cast(mask, dtype=tf.float32)
+    print(mask.shape)
     #mask = torch.bernoulli(probs)
     ogb = False