Update exercise12_HyperParameterTuning.ipynb

Author: fs446

exercise12_HyperParameterTuning.ipynb

@@ -63,8 +63,8 @@
     "from sklearn.model_selection import train_test_split\n",
     "from sklearn.preprocessing import OneHotEncoder, LabelBinarizer\n",
     "import tensorflow as tf\n",
-    "import tensorflow.keras as keras\n",
-    "import tensorflow.keras.backend as K\n",
+    "from tensorflow import keras\n",
+    "from keras import backend as K\n",
     "import time\n",
     "\n",
     "print(\n",
@@ -119,7 +119,9 @@
     "print(root_logdir)\n",
     "print(kt_logdir)  # folder for keras tuner results\n",
     "print(tf_kt_logdir)  # folder for TF checkpoints while keras tuning\n",
-    "print(tf_logdir)  # folder for TF checkpoint for best model training"
+    "print(tf_logdir)  # folder for TF checkpoint for best model training\n",
+    "\n",
+    "os.makedirs(tf_logdir, exist_ok=True)"
    ]
   },
   {
@@ -157,7 +159,7 @@
     "    flip_y=1e-2,\n",
     "    random_state=None,\n",
     ")\n",
-    "encoder = OneHotEncoder(sparse=False)\n",
+    "encoder = OneHotEncoder(sparse_output=False)\n",
     "# we encode as one-hot for TF model\n",
     "Y = encoder.fit_transform(Y.reshape(-1, 1))\n",
     "\n",
@@ -209,7 +211,7 @@
     "def build_model(hp):  # with hyper parameter ranges\n",
     "    model = keras.Sequential()\n",
     "    # input layer\n",
-    "    model.add(keras.Input(shape=nx))\n",
+    "    model.add(keras.Input(shape=(nx, )))\n",
     "    # hidden layers\n",
     "    for layer in range(hp.Int(\"no_layers\", 1, 4)):\n",
     "        model.add(\n",
@@ -260,7 +262,7 @@
     "model = build_model(kt.HyperParameters())\n",
     "hptuner = kt.RandomSearch(\n",
     "    hypermodel=build_model,\n",
-    "    objective=\"val_categorical_accuracy\",  # check performance on val data!\n",
+    "    objective='val_loss',  # check performance on val data!\n",
     "    max_trials=max_trials,\n",
     "    executions_per_trial=executions_per_trial,\n",
     "    overwrite=True,\n",
@@ -319,7 +321,7 @@
     "# we might check the best XX models in detail\n",
     "# for didactical purpose we choose only the very best one, located in [0]:\n",
     "model = hptuner.get_best_models(num_models=1)[0]\n",
-    "model.save(tf_logdir + \"/best_model\")"
+    "model.save(tf_logdir + \"/best_model.keras\")"
    ]
   },
   {
@@ -356,7 +358,7 @@
    "outputs": [],
    "source": [
     "# load best model and reset weights\n",
-    "model = keras.models.load_model(tf_logdir + \"/best_model\")\n",
+    "model = keras.models.load_model(tf_logdir + \"/best_model.keras\")\n",
     "reset_weights(model)  # start training from scratch\n",
     "print(model.summary())"
    ]
@@ -391,7 +393,7 @@
     "    callbacks=[earlystopping_cb, tensorboard_cb],\n",
     "    verbose=verbose,\n",
     ")\n",
-    "model.save(tf_logdir + \"/trained_best_model\")\n",
+    "model.save(tf_logdir + \"/trained_best_model.keras\")\n",
     "print(model.summary())"
    ]
   },
@@ -484,7 +486,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.10.6"
+   "version": "3.12.3"
   }
  },
  "nbformat": 4,