Several updates to tutorials.

- Add buttons for docs compliance. Only the tutorials shown on tensorflow.org/neural_structured_learning have the 'download' button.
- Update the license header to include 'The TensorFlow Neural Structured Learning Authors' for tutorials shown on tf.org for docs compliance.
- Don't include sigmoid/softmax as an activation function for the last layer of the model. Instead change the loss function so that it operates on unscaled logits to avoid numerical instability.
- Update graph_keras_cnn_flowers.ipynb to accept named inputs. This tutorial was broken before.
- Update adversarial_cnn_transfer_learning_fashionmnist.ipynb to use tf.image.resize instead of cv2.resize because the latter doesn't exist in certain environments by default. Also fix an undefined variable error in one of the functions.

PiperOrigin-RevId: 419622832

Author: arjung

g3doc/tutorials/adversarial_keras_cnn_mnist.ipynb

@@ -6,7 +6,7 @@
         "id": "ZwZNOAMZcxl3"
       },
       "source": [
-        "##### Copyright 2019 Google LLC"
+        "##### Copyright 2019 The TensorFlow Neural Structured Learning Authors"
       ]
     },
     {
@@ -56,6 +56,9 @@
         "  \u003ctd\u003e\n",
         "    \u003ca target=\"_blank\" href=\"https://github.com/tensorflow/neural-structured-learning/blob/master/g3doc/tutorials/adversarial_keras_cnn_mnist.ipynb\"\u003e\u003cimg src=\"https://www.tensorflow.org/images/GitHub-Mark-32px.png\" /\u003eView source on GitHub\u003c/a\u003e\n",
         "  \u003c/td\u003e\n",
+        "  \u003ctd\u003e\n",
+        "    \u003ca href=\"https://storage.googleapis.com/tensorflow_docs/neural-structured-learning/g3doc/tutorials/adversarial_keras_cnn_mnist.ipynb\"\u003e\u003cimg src=\"https://www.tensorflow.org/images/download_logo_32px.png\" /\u003eDownload notebook\u003c/a\u003e\n",
+        "  \u003c/td\u003e\n",
         "\u003c/table\u003e"
       ]
     },
@@ -404,7 +407,7 @@
         "  x = tf.keras.layers.Flatten()(x)\n",
         "  for num_units in hparams.num_fc_units:\n",
         "    x = tf.keras.layers.Dense(num_units, activation='relu')(x)\n",
-        "  pred = tf.keras.layers.Dense(hparams.num_classes, activation='softmax')(x)\n",
+        "  pred = tf.keras.layers.Dense(hparams.num_classes)(x)\n",
         "  model = tf.keras.Model(inputs=inputs, outputs=pred)\n",
         "  return model"
       ]
@@ -438,8 +441,10 @@
       },
       "outputs": [],
       "source": [
-        "base_model.compile(optimizer='adam', loss='sparse_categorical_crossentropy',\n",
-        "                   metrics=['acc'])\n",
+        "base_model.compile(\n",
+        "    optimizer='adam',\n",
+        "    loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),\n",
+        "    metrics=['acc'])\n",
         "base_model.fit(train_dataset, epochs=HPARAMS.epochs)"
       ]
     },
@@ -565,8 +570,10 @@
       },
       "outputs": [],
       "source": [
-        "adv_model.compile(optimizer='adam', loss='sparse_categorical_crossentropy',\n",
-        "                   metrics=['acc'])\n",
+        "adv_model.compile(\n",
+        "    optimizer='adam',\n",
+        "    loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),\n",
+        "    metrics=['acc'])\n",
         "adv_model.fit(train_set_for_adv_model, epochs=HPARAMS.epochs)"
       ]
     },
@@ -620,12 +627,10 @@
       "outputs": [],
       "source": [
         "reference_model = nsl.keras.AdversarialRegularization(\n",
-        "    base_model,\n",
-        "    label_keys=[LABEL_INPUT_NAME],\n",
-        "    adv_config=adv_config)\n",
+        "    base_model, label_keys=[LABEL_INPUT_NAME], adv_config=adv_config)\n",
         "reference_model.compile(\n",
         "    optimizer='adam',\n",
-        "    loss='sparse_categorical_crossentropy',\n",
+        "    loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),\n",
         "    metrics=['acc'])"
       ]
     },
@@ -685,7 +690,7 @@
         "for batch in test_set_for_adv_model:\n",
         "  perturbed_batch = reference_model.perturb_on_batch(batch)\n",
         "  # Clipping makes perturbed examples have the same range as regular ones.\n",
-        "  perturbed_batch[IMAGE_INPUT_NAME] = tf.clip_by_value(                          \n",
+        "  perturbed_batch[IMAGE_INPUT_NAME] = tf.clip_by_value(\n",
         "      perturbed_batch[IMAGE_INPUT_NAME], 0.0, 1.0)\n",
         "  y_true = perturbed_batch.pop(LABEL_INPUT_NAME)\n",
         "  perturbed_images.append(perturbed_batch[IMAGE_INPUT_NAME].numpy())\n",
@@ -742,7 +747,7 @@
         "\n",
         "batch_size = HPARAMS.batch_size\n",
         "n_col = 4\n",
-        "n_row = (batch_size + n_col - 1) / n_col\n",
+        "n_row = (batch_size + n_col - 1) // n_col\n",
         "\n",
         "print('accuracy in batch %d:' % batch_index)\n",
         "for name, pred in batch_pred.items():\n",