Do not leak information

blue-yonder · Aug 3, 2024 · 821895c · 821895c
1 parent 659f63d
commit 821895c
Showing 1 changed file with 20 additions and 19 deletions.
diff --git a/notebooks/01 Feature Extraction and Selection.ipynb b/notebooks/01 Feature Extraction and Selection.ipynb
@@ -157,7 +157,18 @@
    "source": [
     "Using the hypothesis tests implemented in `tsfresh` (see [here](https://tsfresh.readthedocs.io/en/latest/text/feature_filtering.html) for more information) it is now possible to select only the relevant features out of this large dataset.\n",
     "\n",
-    "`tsfresh` will do a hypothesis test for each of the features to check, if it is relevant for your given target."
+    "`tsfresh` will do a hypothesis test for each of the features to check, if it is relevant for your given target.\n",
+    "\n",
+    "To not leak information between the train and the test set, we will only perform the selection on the train set"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "X_full_train, X_full_test, y_train, y_test = train_test_split(X, y, test_size=.4, random_state=42)"
    ]
   },
   {
@@ -166,7 +177,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "X_filtered = select_features(X, y)"
+    "X_filtered_train = select_features(X_full_train, y_train)"
    ]
   },
   {
@@ -177,7 +188,7 @@
    },
    "outputs": [],
    "source": [
-    "X_filtered.head()"
+    "X_filtered_train.head()"
    ]
   },
   {
@@ -186,7 +197,7 @@
    "source": [
     "<div class=\"alert alert-info\">\n",
     "\n",
-    "Currently, 669 non-NaN features survive the feature selection given this target.\n",
+    "Currently, 423 non-NaN features survive the feature selection given this target.\n",
     "Again, this number will vary depending on your data, your target and the `tsfresh` version.\n",
     "    \n",
     "</div>"
@@ -214,8 +225,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "X_full_train, X_full_test, y_train, y_test = train_test_split(X, y, test_size=.4)\n",
-    "X_filtered_train, X_filtered_test = X_full_train[X_filtered.columns], X_full_test[X_filtered.columns]"
+    "X_filtered_train, X_filtered_test = X_full_train[X_filtered_train.columns], X_full_test[X_filtered_train.columns]"
    ]
   },
   {
@@ -246,7 +256,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "Compared to using all features (`classifier_full`), using only the relevant features (`classifier_filtered`) achieves better classification performance with less data."
+    "Compared to using all features (`classifier_full`), using only the relevant features (`classifier_filtered`) achieves similar or better classification performance with much less data."
    ]
   },
   {
@@ -284,17 +294,8 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "X_filtered_2 = extract_relevant_features(df, y, column_id='id', column_sort='time',\n",
-    "                                         default_fc_parameters=extraction_settings)"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "(X_filtered.columns == X_filtered_2.columns).all()"
+    "extract_relevant_features(df, y, column_id='id', column_sort='time',\n",
+    "                          default_fc_parameters=extraction_settings)"
    ]
   }
  ],
@@ -314,7 +315,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.8.2"
+   "version": "3.10.13"
   }
  },
  "nbformat": 4,