Minor fix to tutorials.

PiperOrigin-RevId: 463145196

tutorials/lm_dpsgd_tutorial.py

@@ -73,8 +73,8 @@ def rnn_model_fn(features, labels, mode):  # pylint: disable=unused-argument
   x = tf.reshape(x, [-1, SEQ_LEN])
   input_layer = x[:, :-1]
   input_one_hot = tf.one_hot(input_layer, 256)
-  lstm = tf.keras.layers.LSTM(256, return_sequences=True).apply(input_one_hot)
+  lstm = tf.keras.layers.LSTM(256, return_sequences=True)(input_one_hot)
-  logits = tf.keras.layers.Dense(256).apply(lstm)
+  logits = tf.keras.layers.Dense(256)(lstm)
 
   # Calculate loss as a vector (to support microbatches in DP-SGD).
   vector_loss = tf.nn.softmax_cross_entropy_with_logits(

tutorials/mnist_dpsgd_tutorial.py

@@ -85,7 +85,7 @@ def cnn_model_fn(features, labels, mode, params):  # pylint: disable=unused-argu
   elif mode == tf_estimator.ModeKeys.EVAL:
     eval_metric_ops = {
         'accuracy':
-            tf.metrics.accuracy(
+            tf.compat.v1.metrics.accuracy(
                 labels=labels, predictions=tf.argmax(input=logits, axis=1))
     }
     return tf_estimator.EstimatorSpec(

tutorials/mnist_dpsgd_tutorial_common.py

@@ -21,14 +21,16 @@ def get_cnn_model(features):
   """Given input features, returns the logits from a simple CNN model."""
   input_layer = tf.reshape(features, [-1, 28, 28, 1])
   y = tf.keras.layers.Conv2D(
-      16, 8, strides=2, padding='same', activation='relu').apply(input_layer)
+      16, 8, strides=2, padding='same', activation='relu')(
-  y = tf.keras.layers.MaxPool2D(2, 1).apply(y)
+          input_layer)
+  y = tf.keras.layers.MaxPool2D(2, 1)(y)
   y = tf.keras.layers.Conv2D(
-      32, 4, strides=2, padding='valid', activation='relu').apply(y)
+      32, 4, strides=2, padding='valid', activation='relu')(
-  y = tf.keras.layers.MaxPool2D(2, 1).apply(y)
+          y)
-  y = tf.keras.layers.Flatten().apply(y)
+  y = tf.keras.layers.MaxPool2D(2, 1)(y)
-  y = tf.keras.layers.Dense(32, activation='relu').apply(y)
+  y = tf.keras.layers.Flatten()(y)
-  logits = tf.keras.layers.Dense(10).apply(y)
+  y = tf.keras.layers.Dense(32, activation='relu')(y)
+  logits = tf.keras.layers.Dense(10)(y)
 
   return logits
 

tutorials/mnist_dpsgd_tutorial_vectorized.py

@@ -70,14 +70,16 @@ def cnn_model_fn(features, labels, mode):
   # Define CNN architecture using tf.keras.layers.
   input_layer = tf.reshape(features['x'], [-1, 28, 28, 1])
   y = tf.keras.layers.Conv2D(
-      16, 8, strides=2, padding='same', activation='relu').apply(input_layer)
+      16, 8, strides=2, padding='same', activation='relu')(
-  y = tf.keras.layers.MaxPool2D(2, 1).apply(y)
+          input_layer)
+  y = tf.keras.layers.MaxPool2D(2, 1)(y)
   y = tf.keras.layers.Conv2D(
-      32, 4, strides=2, padding='valid', activation='relu').apply(y)
+      32, 4, strides=2, padding='valid', activation='relu')(
-  y = tf.keras.layers.MaxPool2D(2, 1).apply(y)
+          y)
-  y = tf.keras.layers.Flatten().apply(y)
+  y = tf.keras.layers.MaxPool2D(2, 1)(y)
-  y = tf.keras.layers.Dense(32, activation='relu').apply(y)
+  y = tf.keras.layers.Flatten()(y)
-  logits = tf.keras.layers.Dense(10).apply(y)
+  y = tf.keras.layers.Dense(32, activation='relu')(y)
+  logits = tf.keras.layers.Dense(10)(y)
 
   # Calculate loss as a vector (to support microbatches in DP-SGD).
   vector_loss = tf.nn.sparse_softmax_cross_entropy_with_logits(
@@ -115,7 +117,7 @@ def cnn_model_fn(features, labels, mode):
   elif mode == tf_estimator.ModeKeys.EVAL:
     eval_metric_ops = {
         'accuracy':
-            tf.metrics.accuracy(
+            tf.compat.v1.metrics.accuracy(
                 labels=labels, predictions=tf.argmax(input=logits, axis=1))
     }
 

tutorials/mnist_lr_tutorial.py

@@ -56,12 +56,12 @@ def lr_model_fn(features, labels, mode, nclasses, dim):
   logits = tf.keras.layers.Dense(
       units=nclasses,
       kernel_regularizer=tf.keras.regularizers.L2(l2=FLAGS.regularizer),
-      bias_regularizer=tf.keras.regularizers.L2(
+      bias_regularizer=tf.keras.regularizers.L2(l2=FLAGS.regularizer))(
-          l2=FLAGS.regularizer)).apply(input_layer)
+          input_layer)
 
   # Calculate loss as a vector (to support microbatches in DP-SGD).
   vector_loss = tf.nn.sparse_softmax_cross_entropy_with_logits(
-      labels=labels, logits=logits) + tf.losses.get_regularization_loss()
+      labels, logits) + tf.compat.v1.losses.get_regularization_loss()
   # Define mean of loss across minibatch (for reporting through tf.Estimator).
   scalar_loss = tf.reduce_mean(vector_loss)
 
@@ -94,7 +94,7 @@ def lr_model_fn(features, labels, mode, nclasses, dim):
   elif mode == tf_estimator.ModeKeys.EVAL:
     eval_metric_ops = {
         'accuracy':
-            tf.metrics.accuracy(
+            tf.compat.v1.metrics.accuracy(
                 labels=labels, predictions=tf.argmax(input=logits, axis=1))
     }
     return tf_estimator.EstimatorSpec(
@@ -165,7 +165,7 @@ def print_privacy_guarantees(epochs, batch_size, samples, noise_multiplier):
     # Using RDP accountant to compute eps. Doing computation analytically is
     # an option.
     rdp = [order * coef for order in orders]
-    eps = dp_accounting.rdp.compute_epsilon(orders, rdp, delta)
+    eps, _ = dp_accounting.rdp.compute_epsilon(orders, rdp, delta)
     print('\t{:g}% enjoy at least ({:.2f}, {})-DP'.format(p * 100, eps, delta))
 
   accountant = dp_accounting.rdp.RdpAccountant(orders)