Clarifications in the tutorial

Author: Lucas Zimmer

examples/basics/Auto-PyTorch Tutorial.ipynb

@@ -50,7 +50,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "Upon initialization of a class, you can specify its configuration. The *config_preset* allows to constrain the search space to one of *tiny_cs, medium_cs* or *full_cs*. These presets can be seen in *core/presets/*."
+    "Upon initialization of a class, you can specify its configuration. Later, you can override its configuration in each fit call. The *config_preset* allows to constrain the search space to one of *tiny_cs, medium_cs* or *full_cs*. These presets can be seen in *core/presets/*."
    ]
   },
   {
@@ -93,7 +93,7 @@
    "source": [
     "The most important methods for using Auto-PyTorch are **fit**, **refit**, **score** and **predict**.\n",
     "\n",
-    "**fit** is used to search for a configuration:"
+    "**fit** is used to search for a good configuration by fitting different parameter configurations. The incumbent configuration is then returned and stored in the class:"
    ]
   },
   {
@@ -151,21 +151,19 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "**refit** allows you to fit a configuration of your choice for a defined time:"
+    "**refit** allows you to fit a configuration of your choice for a defined time. You can specify a hyperparameter configuration to fit (if you do not specify a configuration the incumbent configuration from the last fit call will be used):"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": null,
-   "metadata": {
-    "scrolled": true
-   },
+   "metadata": {},
    "outputs": [],
    "source": [
     "# Create an autonet, use tensorboard during fitting\n",
     "autonet_config = {\n",
     "    \"result_logger_dir\" : \"logs/\",\n",
-    "    \"budget_type\" : \"epochs\",\n",
+    "    \"budget_type\" : \"time\",\n",
     "    \"log_level\" : \"info\", \n",
     "    \"use_tensorboard_logger\" : True\n",
     "    }\n",
@@ -174,14 +172,14 @@
     "# This samples a random hyperparameter configuration as an example\n",
     "hyperparameter_config = autonet.get_hyperparameter_search_space().sample_configuration().get_dictionary()\n",
     "\n",
-    "# Refit with sampled hyperparameter config for 10 epochs. This time on the full dataset.\n",
+    "# Refit with sampled hyperparameter config for 120 s. This time on the full dataset.\n",
     "results_refit = autonet.refit(X_train=X_train,\n",
     "                              Y_train=Y_train,\n",
     "                              X_valid=None,\n",
     "                              Y_valid=None,\n",
     "                              hyperparameter_config=hyperparameter_config,\n",
     "                              autonet_config=autonet.get_current_autonet_config(),\n",
-    "                              budget=10)\n",
+    "                              budget=120)\n",
     "\n",
     "# Save json\n",
     "with open(\"logs/results_refit.json\", \"w\") as file:\n",
@@ -201,18 +199,19 @@
    "metadata": {},
    "outputs": [],
    "source": [
+    "# Print score of found configuration\n",
     "score = autonet.score(X_test=X_test, Y_test=Y_test)\n",
     "pred = autonet.predict(X=X_test)\n",
     "\n",
-    "print(\"Model prediction:\", pred)\n",
+    "print(\"Model prediction:\", pred[0:10])\n",
     "print(\"Accuracy score\", score)"
    ]
   },
   {
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "Finall, you can also get the incumbent model as PyTorch Sequential model via"
+    "Finally, you can also get the incumbent model as PyTorch Sequential model via"
    ]
   },
   {
@@ -321,9 +320,9 @@
     "autonet_image_classification.fit(X_train=X_train,\n",
     "                                 Y_train=Y_train,\n",
     "                                 images_shape=[3,32,32],\n",
-    "                                 min_budget=100,\n",
-    "                                 max_budget=200,\n",
-    "                                 max_runtime=400,\n",
+    "                                 min_budget=900,\n",
+    "                                 max_budget=1200,\n",
+    "                                 max_runtime=3000,\n",
     "                                 save_checkpoints=True,\n",
     "                                 images_root_folders=[os.path.abspath(\"../../datasets/example_images\")])"
    ]