Regenerate KerasRS examples to apply changes. (#2095)

- Apply changes to examples.
- Apply fix to progress bars.
- Also fix some preprocessing issue with `deep_recommender` example.

The `scann` and `data_parallel_retrieval` examples will be done separately.

Author: hertschuh

examples/keras_rs/deep_recommender.py

@@ -429,9 +429,7 @@ def preprocess_rating(x):
             "raw_user_age": features["raw_user_age"],
             "user_gender": features["user_gender"],
             "user_occupation_label": features["user_occupation_label"],
-            "user_gender_X_raw_user_age": tf.squeeze(
-                features["user_gender_X_raw_user_age"], axis=-1
-            ),
+            "user_gender_X_raw_user_age": features["user_gender_X_raw_user_age"],
             # Movie inputs are movie ID, vectorized title and genres
             "movie_id": int(x["movie_id"]),
             "movie_title_vector": features["movie_title"],

examples/keras_rs/img/dcn/dcn_16_2.png

@@ -155,11 +155,7 @@
     "    \"num_epochs\": 10,\n",
     "    \"batch_size\": 1024,\n",
     "}\n",
-    "\n",
-    "LOOKUP_LAYERS = {\n",
-    "    \"int\": keras.layers.IntegerLookup,\n",
-    "    \"str\": keras.layers.StringLookup,\n",
-    "}"
+    ""
    ]
   },
   {
@@ -234,9 +230,7 @@
     "    if num_trials == 1:\n",
     "        print(f\"{model_name}: RMSE = {avg_rmse}; #params = {num_params}\")\n",
     "    else:\n",
-    "        print(\n",
-    "            f\"{model_name}: RMSE = {avg_rmse} \u00b1 {std_rmse}; \" \"#params = {num_params}\"\n",
-    "        )\n",
+    "        print(f\"{model_name}: RMSE = {avg_rmse} \u00b1 {std_rmse}; #params = {num_params}\")\n",
     ""
    ]
   },
@@ -351,6 +345,7 @@
     "        keras.layers.Dense(512, activation=\"relu\"),\n",
     "        keras.layers.Dense(256, activation=\"relu\"),\n",
     "        keras.layers.Dense(128, activation=\"relu\"),\n",
+    "        keras.layers.Dense(1),\n",
     "    ]\n",
     ")"
    ]

examples/keras_rs/img/dcn/dcn_31_2.png

@@ -734,9 +734,7 @@
     "            \"raw_user_age\": features[\"raw_user_age\"],\n",
     "            \"user_gender\": features[\"user_gender\"],\n",
     "            \"user_occupation_label\": features[\"user_occupation_label\"],\n",
-    "            \"user_gender_X_raw_user_age\": tf.squeeze(\n",
-    "                features[\"user_gender_X_raw_user_age\"], axis=-1\n",
-    "            ),\n",
+    "            \"user_gender_X_raw_user_age\": features[\"user_gender_X_raw_user_age\"],\n",
     "            # Movie inputs are movie ID, vectorized title and genres\n",
     "            \"movie_id\": int(x[\"movie_id\"]),\n",
     "            \"movie_title_vector\": features[\"movie_title\"],\n",
@@ -1029,7 +1027,7 @@
     "        self.update_candidates()  # Provide an initial set of candidates\n",
     "        self.loss_fn = keras.losses.MeanSquaredError()\n",
     "        self.top_k_metric = keras.metrics.SparseTopKCategoricalAccuracy(\n",
-    "            k=100, from_sorted_ids=True\n",
+    "            k=retrieval_k, from_sorted_ids=True\n",
     "        )\n",
     "\n",
     "    def update_candidates(self):\n",

examples/keras_rs/img/deep_recommender/deep_recommender_57_0.png

@@ -258,16 +258,19 @@
     "    }\n",
     "    labels = []\n",
     "    for user_id, user_list in sequences.items():\n",
-    "        sampled_list = sample_sublist_from_list(\n",
-    "            user_list,\n",
-    "            num_examples_per_list,\n",
-    "        )\n",
-    "\n",
-    "        inputs[\"user_id\"].append(user_id)\n",
-    "        inputs[\"movie_id\"].append(\n",
-    "            tf.convert_to_tensor([f[\"movie_id\"] for f in sampled_list])\n",
-    "        )\n",
-    "        labels.append(tf.convert_to_tensor([f[\"user_rating\"] for f in sampled_list]))\n",
+    "        for _ in range(num_list_per_user):\n",
+    "            sampled_list = sample_sublist_from_list(\n",
+    "                user_list,\n",
+    "                num_examples_per_list,\n",
+    "            )\n",
+    "\n",
+    "            inputs[\"user_id\"].append(user_id)\n",
+    "            inputs[\"movie_id\"].append(\n",
+    "                tf.convert_to_tensor([f[\"movie_id\"] for f in sampled_list])\n",
+    "            )\n",
+    "            labels.append(\n",
+    "                tf.convert_to_tensor([f[\"user_rating\"] for f in sampled_list])\n",
+    "            )\n",
     "\n",
     "    return (\n",
     "        {\"user_id\": inputs[\"user_id\"], \"movie_id\": inputs[\"movie_id\"]},\n",

examples/keras_rs/img/deep_recommender/deep_recommender_61_0.png

@@ -233,7 +233,7 @@
     "        )\n",
     "\n",
     "        # Rating model.\n",
-    "        self.rating_model = tf.keras.Sequential(\n",
+    "        self.rating_model = keras.Sequential(\n",
     "            [\n",
     "                keras.layers.Dense(layer_size, activation=\"relu\")\n",
     "                for layer_size in layer_sizes\n",
@@ -252,7 +252,7 @@
     "\n",
     "        # Top-k accuracy for retrieval\n",
     "        self.top_k_metric = keras.metrics.SparseTopKCategoricalAccuracy(\n",
-    "            k=100, from_sorted_ids=True\n",
+    "            k=10, from_sorted_ids=True\n",
     "        )\n",
     "        # RMSE for ranking\n",
     "        self.rmse_metric = keras.metrics.RootMeanSquaredError()\n",
@@ -410,7 +410,7 @@
     "    ranking_loss_wt=1.0,\n",
     "    retrieval_loss_wt=0.0,\n",
     ")\n",
-    "model.compile(optimizer=tf.keras.optimizers.Adagrad(0.1))\n",
+    "model.compile(optimizer=keras.optimizers.Adagrad(0.1))\n",
     "model.fit(train_ratings, epochs=5)\n",
     "\n",
     "model.evaluate(test_ratings)\n",
@@ -422,7 +422,7 @@
     "    ranking_loss_wt=0.0,\n",
     "    retrieval_loss_wt=1.0,\n",
     ")\n",
-    "model.compile(optimizer=tf.keras.optimizers.Adagrad(0.1))\n",
+    "model.compile(optimizer=keras.optimizers.Adagrad(0.1))\n",
     "model.fit(train_ratings, epochs=5)\n",
     "\n",
     "model.evaluate(test_ratings)\n",
@@ -434,7 +434,7 @@
     "    ranking_loss_wt=1.0,\n",
     "    retrieval_loss_wt=1.0,\n",
     ")\n",
-    "model.compile(optimizer=tf.keras.optimizers.Adagrad(0.1))\n",
+    "model.compile(optimizer=keras.optimizers.Adagrad(0.1))\n",
     "model.fit(train_ratings, epochs=5)\n",
     "\n",
     "model.evaluate(test_ratings)"

examples/keras_rs/ipynb/dcn.ipynb

@@ -96,7 +96,7 @@
     "TEST_BATCH_SIZE = 2048\n",
     "EMBEDDING_DIM = 32\n",
     "NUM_EPOCHS = 5\n",
-    "LEARNING_RATE = 0.05"
+    "LEARNING_RATE = 0.005"
    ]
   },
   {
@@ -516,7 +516,7 @@
    "outputs": [],
    "source": [
     "model = SequentialRetrievalModel(\n",
-    "    movies_count=movies_count + 1, embedding_dimension=EMBEDDING_DIM\n",
+    "    movies_count=movies_count, embedding_dimension=EMBEDDING_DIM\n",
     ")\n",
     "\n",
     "# Compile.\n",