- changed MAD Outlier Removal - Function

- added regulation to xgboost

model_training/MAD_outlier_removal/mad_outlier_removal.ipynb

@@ -8,6 +8,64 @@
     "Im folgenden wird auf die Daten das MAD Outlier removal angewendet."
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "46bd036d",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import numpy as np\n",
+    "import pandas as pd\n",
+    "\n",
+    "def calculate_mad_params(df, columns):\n",
+    "    \"\"\"\n",
+    "    Calculate median and MAD parameters for each column.\n",
+    "    This should be run ONLY on the training data.\n",
+    "    \n",
+    "    Returns a dictionary: {col: (median, mad)}\n",
+    "    \"\"\"\n",
+    "    params = {}\n",
+    "    for col in columns:\n",
+    "        median = df[col].median()\n",
+    "        mad = np.median(np.abs(df[col] - median))\n",
+    "        params[col] = (median, mad)\n",
+    "    return params"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "e0691732",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def apply_mad_filter(df, params, threshold=3.5):\n",
+    "    \"\"\"\n",
+    "    Apply MAD-based outlier removal using precomputed parameters.\n",
+    "    Works on training, validation, and test data.\n",
+    "    \n",
+    "    df: DataFrame to filter\n",
+    "    params: dictionary {col: (median, mad)} from training data\n",
+    "    threshold: cutoff for robust Z-score\n",
+    "    \"\"\"\n",
+    "    df_clean = df.copy()\n",
+    "\n",
+    "    for col, (median, mad) in params.items():\n",
+    "        if mad == 0:\n",
+    "            continue  # no spread; nothing to remove for this column\n",
+    "\n",
+    "        robust_z = 0.6745 * (df_clean[col] - median) / mad\n",
+    "        outlier_mask = np.abs(robust_z) > threshold\n",
+    "\n",
+    "        # Remove values only in this specific column\n",
+    "        df_clean.loc[outlier_mask, col] = np.nan\n",
+    "        print(df_clean.shape)\n",
+    "        \n",
+    "    print(df_clean.shape)\n",
+    "    return df_clean"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,
@@ -15,6 +73,9 @@
    "metadata": {},
    "outputs": [],
    "source": [
+    "# old removal - with this we were able to get 85% accuracy\n",
+    "# the values of the old validation & test data set is stored privately on the Cluster\n",
+    "\n",
     "import numpy as np\n",
     "import pandas as pd\n",
     "from sklearn.preprocessing import StandardScaler, MinMaxScaler\n",
@@ -32,8 +93,9 @@
     "            continue  # keine Streuung, keine Ausreißer\n",
     "        robust_z = 0.6745 * (df_clean[col] - median) / mad\n",
     "        mask = np.abs(robust_z) <= threshold\n",
-    "        df_clean = df_clean[mask]\n",
+    "        output = df_clean[mask]\n",
-    "    return df_clean"
+    "\n",
+    "    return output"
    ]
   }
  ],

model_training/xgboost/xgboost_regulated.ipynb

@@ -0,0 +1,538 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "e3be057e-8d2a-4d05-bd42-6b1dc75df5ed",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import pandas as pd\n",
+    "from pathlib import Path\n",
+    "from sklearn.preprocessing import StandardScaler, MinMaxScaler"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "13ad96f5",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# data_path = Path(r\"~/Fahrsimulator_MSY2526_AI/model_training/xgboost/output_windowed.parquet\")\n",
+    "data_path = Path(r\"~/data-paulusjafahrsimulator-gpu/first_AU_dataset/output_windowed.parquet\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "95e1a351",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "df = pd.read_parquet(path=data_path)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "68afd83e",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "subjects = df['subjectID'].unique()\n",
+    "print(subjects)\n",
+    "print(len(subjects))\n",
+    "print(len(subjects)*0.66)\n",
+    "print(len(subjects)*0.33)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "52dfd885",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "low_all = df[\n",
+    "    ((df[\"PHASE\"] == \"baseline\") |\n",
+    "     ((df[\"STUDY\"] == \"n-back\") & (df[\"PHASE\"] != \"baseline\") & (df[\"LEVEL\"].isin([1, 4]))))\n",
+    "]\n",
+    "print(f\"low all: {low_all.shape}\")\n",
+    "\n",
+    "high_nback = df[\n",
+    "    (df[\"STUDY\"]==\"n-back\") &\n",
+    "    (df[\"LEVEL\"].isin([2, 3, 5, 6])) &\n",
+    "    (df[\"PHASE\"].isin([\"train\", \"test\"]))\n",
+    "]\n",
+    "print(f\"high n-back: {high_nback.shape}\")\n",
+    "\n",
+    "high_kdrive = df[\n",
+    "    (df[\"STUDY\"] == \"k-drive\") & (df[\"PHASE\"] != \"baseline\")\n",
+    "]\n",
+    "print(f\"high k-drive: {high_kdrive.shape}\")\n",
+    "\n",
+    "high_all = pd.concat([high_nback, high_kdrive])\n",
+    "print(f\"high all: {high_all.shape}\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "8fba6edf",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def fit_normalizer(train_data, au_columns, method='standard', scope='global'):\n",
+    "    if method == 'standard':\n",
+    "        Scaler = StandardScaler\n",
+    "    elif method == 'minmax':\n",
+    "        Scaler = MinMaxScaler\n",
+    "    else:\n",
+    "        raise ValueError(\"method must be 'standard' or 'minmax'\")\n",
+    "    \n",
+    "    scalers = {}\n",
+    "    \n",
+    "    if scope == 'subject':\n",
+    "        for subject in train_data['subjectID'].unique():\n",
+    "            subject_mask = train_data['subjectID'] == subject\n",
+    "            scaler = Scaler()\n",
+    "            scaler.fit(train_data.loc[subject_mask, au_columns])\n",
+    "            scalers[subject] = scaler\n",
+    "\n",
+    "    elif scope == 'global':\n",
+    "        scaler = Scaler()\n",
+    "        scaler.fit(train_data[au_columns])\n",
+    "        scalers['global'] = scaler\n",
+    "\n",
+    "    else:\n",
+    "        raise ValueError(\"scope must be 'subject' or 'global'\")\n",
+    "    \n",
+    "    return {'scalers': scalers, 'method': method, 'scope': scope}\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "24e3a77b",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "%pip install xgboost"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "8e7fa0fa",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import numpy as np\n",
+    "from sklearn.model_selection import train_test_split,StratifiedKFold, GridSearchCV\n",
+    "from sklearn.metrics import accuracy_score, f1_score, roc_auc_score, classification_report, confusion_matrix\n",
+    "import xgboost as xgb\n",
+    "import joblib\n",
+    "import matplotlib.pyplot as plt"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "325ef71c",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "low = low_all.copy()\n",
+    "high = high_all.copy()\n",
+    "\n",
+    "low[\"label\"] = 0\n",
+    "high[\"label\"] = 1\n",
+    "\n",
+    "data = pd.concat([low, high], ignore_index=True)\n",
+    "data = data.drop_duplicates()\n",
+    "\n",
+    "print(\"Label distribution:\")\n",
+    "print(data[\"label\"].value_counts())"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "67d70e84",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "au_columns = [col for col in data.columns if col.lower().startswith(\"au\")]\n",
+    "print(\"Gefundene AU-Spalten:\", au_columns)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "960bb8c7",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "subjects = np.random.permutation(data[\"subjectID\"].unique())\n",
+    "\n",
+    "n = len(subjects)\n",
+    "n_train = int(n * 0.66)\n",
+    "\n",
+    "train_subjects = subjects[:n_train]\n",
+    "test_subjects  = subjects[n_train:]\n",
+    "train_subs, val_subs = train_test_split(train_subjects, test_size=0.2, random_state=42)\n",
+    "\n",
+    "train_df = data[data.subjectID.isin(train_subs)]\n",
+    "val_df   = data[data.subjectID.isin(val_subs)]\n",
+    "test_df  = data[data.subjectID.isin(test_subjects)]\n",
+    "\n",
+    "print(train_df.shape, val_df.shape, test_df.shape)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "802a45c9",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def apply_normalizer(df_to_transform, normalizer_dict, au_columns):\n",
+    "    scalers = normalizer_dict[\"scalers\"]\n",
+    "    scope = normalizer_dict[\"scope\"]\n",
+    "    df_out = df_to_transform.copy()\n",
+    "\n",
+    "    if scope == \"global\":\n",
+    "        scaler = scalers[\"global\"]\n",
+    "        df_out[au_columns] = scaler.transform(df_out[au_columns])\n",
+    "\n",
+    "    elif scope == \"subject\":\n",
+    "        for subj, subdf in df_out.groupby(\"subjectID\"):\n",
+    "            if subj in scalers:\n",
+    "                df_out.loc[subdf.index, au_columns] = scalers[subj].transform(subdf[au_columns])\n",
+    "            elif \"global\" in scalers:\n",
+    "                df_out.loc[subdf.index, au_columns] = scalers[\"global\"].transform(subdf[au_columns])\n",
+    "\n",
+    "    return df_out"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "dbb58abd",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import numpy as np\n",
+    "import pandas as pd\n",
+    "\n",
+    "def calculate_mad_params(df, columns):\n",
+    "    \"\"\"\n",
+    "    Calculate median and MAD parameters for each column.\n",
+    "    This should be run ONLY on the training data.\n",
+    "    \n",
+    "    Returns a dictionary: {col: (median, mad)}\n",
+    "    \"\"\"\n",
+    "    params = {}\n",
+    "    for col in columns:\n",
+    "        median = df[col].median()\n",
+    "        mad = np.median(np.abs(df[col] - median))\n",
+    "        params[col] = (median, mad)\n",
+    "    return params\n",
+    "\n",
+    "def apply_mad_filter(df, params, threshold=3.5):\n",
+    "    \"\"\"\n",
+    "    Apply MAD-based outlier removal using precomputed parameters.\n",
+    "    Works on training, validation, and test data.\n",
+    "    \n",
+    "    df: DataFrame to filter\n",
+    "    params: dictionary {col: (median, mad)} from training data\n",
+    "    threshold: cutoff for robust Z-score\n",
+    "    \"\"\"\n",
+    "    df_clean = df.copy()\n",
+    "\n",
+    "    for col, (median, mad) in params.items():\n",
+    "        if mad == 0:\n",
+    "            continue  # no spread; nothing to remove for this column\n",
+    "\n",
+    "        robust_z = 0.6745 * (df_clean[col] - median) / mad\n",
+    "        outlier_mask = np.abs(robust_z) > threshold\n",
+    "\n",
+    "        # Remove values only in this specific column\n",
+    "        df_clean.loc[outlier_mask, col] = np.nan\n",
+    "        \n",
+    "    return df_clean"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "0f03f1b4",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Step 1: Fit parameters on training data\n",
+    "params = calculate_mad_params(train_df, au_columns)\n",
+    "\n",
+    "# Step 2: Apply filter consistently\n",
+    "train_outlier_removed = apply_mad_filter(train_df, params, threshold=3.5)\n",
+    "val_outlier_removed   = apply_mad_filter(val_df, params, threshold=50)\n",
+    "test_outlier_removed  = apply_mad_filter(test_df, params, threshold=50)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "289f6b89",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "normalizer = fit_normalizer(train_outlier_removed, au_columns, method=\"standard\", scope=\"global\")\n",
+    "\n",
+    "train_scaled = apply_normalizer(train_outlier_removed, normalizer, au_columns)\n",
+    "val_scaled   = apply_normalizer(val_df, normalizer, au_columns)\n",
+    "test_scaled  = apply_normalizer(test_df, normalizer, au_columns)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "5df30e8d",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "X_train, y_train = train_scaled[au_columns].values, train_scaled[\"label\"].values\n",
+    "X_val,   y_val   = val_scaled[au_columns].values, val_scaled[\"label\"].values\n",
+    "X_test,  y_test  = test_scaled[au_columns].values, test_scaled[\"label\"].values"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "6fb7c86a",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import matplotlib.pyplot as plt\n",
+    "from sklearn.metrics import RocCurveDisplay, log_loss, accuracy_score\n",
+    "early_stop = xgb.callback.EarlyStopping(\n",
+    "    rounds=30, metric_name='auc', data_name='validation_0', save_best=True\n",
+    ")\n",
+    "\n",
+    "# Basis-Modell\n",
+    "xgb_clf = xgb.XGBClassifier(\n",
+    "    objective=\"binary:logistic\",\n",
+    "    scale_pos_weight=1100/1550,\n",
+    "    eval_metric=[\"logloss\",\"auc\",\"error\"],\n",
+    "    use_label_encoder=False,\n",
+    "    random_state=42,\n",
+    "    callbacks=[early_stop],\n",
+    "    verbosity=0,\n",
+    ")\n",
+    "\n",
+    "# Parameter-Raster\n",
+    "param_grid = {\n",
+    "    \"learning_rate\": [0.01, 0.02, 0.05],\n",
+    "    \"max_depth\": [2, 3, 4],\n",
+    "    # \"n_estimators\": [200, 500, 800],\n",
+    "    \"subsample\": [0.4, 0.5],\n",
+    "    \"colsample_bytree\": [0.7, 0.8],\n",
+    "    \"reg_alpha\": [0, 0.1, 1, 10],           # L1 regularization\n",
+    "    \"reg_lambda\": [0.5, 1, 5, 10]           # L2 regularization\n",
+    "}\n",
+    "\n",
+    "# old values - acc 100%\n",
+    "    # \"learning_rate\": [0.01, 0.05, 0.1],\n",
+    "    # \"max_depth\": [4, 6, 8],\n",
+    "    # \"n_estimators\": [200, 500, 800],\n",
+    "    # \"subsample\": [0.8, 1.0],\n",
+    "    # \"colsample_bytree\": [0.8, 1.0]\n",
+    "\n",
+    "# K-Fold Cross Validation\n",
+    "cv = StratifiedKFold(n_splits=5, shuffle=True, random_state=42)\n",
+    "\n",
+    "# Grid Search Setup\n",
+    "grid_search = GridSearchCV(\n",
+    "    estimator=xgb_clf,\n",
+    "    param_grid=param_grid,\n",
+    "    scoring=\"roc_auc\",\n",
+    "    n_jobs=-1,\n",
+    "    cv=cv,\n",
+    "    verbose=0\n",
+    ")\n",
+    "\n",
+    "# Training mit Cross Validation\n",
+    "grid_search.fit(\n",
+    "    X_train, y_train,\n",
+    "    eval_set=[(X_train, y_train), (X_val, y_val)],\n",
+    "    verbose=False,\n",
+    ")\n",
+    "\n",
+    "print(\"Beste Parameter:\", grid_search.best_params_)\n",
+    "print(\"Bestes AUC:\", grid_search.best_score_)\n",
+    "\n",
+    "# Bestes Modell extrahieren\n",
+    "model = grid_search.best_estimator_"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "d2681022",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Plots\n",
+    "\n",
+    "results = model.evals_result()\n",
+    "epochs = len(results['validation_0']['auc'])\n",
+    "x_axis = range(0, epochs)\n",
+    "\n",
+    "# --- Plot Loss ---\n",
+    "plt.figure(figsize=(8,6))\n",
+    "plt.plot(x_axis, results['validation_0']['logloss'], label='Validation Loss')\n",
+    "plt.plot(x_axis, results['validation_1']['logloss'], label='Training Loss')\n",
+    "plt.legend()\n",
+    "plt.xlabel('Epochs')\n",
+    "plt.ylabel('Logloss')\n",
+    "plt.title('XGBoost Loss during Training')\n",
+    "plt.grid(True)\n",
+    "plt.show()\n",
+    "\n",
+    "# --- Plot Accuracy ---\n",
+    "plt.figure(figsize=(8,6))\n",
+    "plt.plot(x_axis, [1-e for e in results['validation_0']['error']], label='Validation Accuracy')\n",
+    "plt.plot(x_axis, [1-e for e in results['validation_1']['error']], label='Training Accuracy')\n",
+    "plt.legend()\n",
+    "plt.xlabel('Epochs')\n",
+    "plt.ylabel('Accuracy')\n",
+    "plt.title('XGBoost Accuracy during Training')\n",
+    "plt.grid(True)\n",
+    "plt.show()\n",
+    "\n",
+    "# Plot AUC\n",
+    "\n",
+    "plt.figure(figsize=(8,6))\n",
+    "plt.plot(x_axis, results['validation_0']['auc'], label='Validation AUC')\n",
+    "plt.plot(x_axis, results['validation_0']['auc'], marker='o')\n",
+    "plt.legend()\n",
+    "plt.xlabel('Epochs')\n",
+    "plt.ylabel('AUC')\n",
+    "plt.title('XGBoost AUC during Training')\n",
+    "plt.grid(True)\n",
+    "plt.show()\n",
+    "\n",
+    "# ROC-Kurve plotten\n",
+    "y_pred_proba = model.predict_proba(X_val)[:, 1]\n",
+    "RocCurveDisplay.from_predictions(y_val, y_pred_proba)\n",
+    "plt.title(\"ROC Curve (Validation Set)\")\n",
+    "plt.grid(True)\n",
+    "plt.show()\n",
+    "\n",
+    "# Test: Loss und Accuracy\n",
+    "y_test_proba = model.predict_proba(X_test)[:,1]\n",
+    "y_test_pred = (y_test_proba > 0.5).astype(int)\n",
+    "\n",
+    "print(\"Test Loss:\", log_loss(y_test, y_test_proba))\n",
+    "print(\"Test Accuracy:\", accuracy_score(y_test, y_test_pred))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "09a8cd21",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from sklearn.metrics import confusion_matrix, accuracy_score, f1_score, roc_auc_score, classification_report, ConfusionMatrixDisplay\n",
+    "\n",
+    "def evaluate(model, X, y, title=\"Evaluation\"):\n",
+    "    # Vorhersagen\n",
+    "    preds_proba = model.predict_proba(X)[:, 1]\n",
+    "    preds = (preds_proba > 0.5).astype(int)\n",
+    "\n",
+    "    # Metriken ausgeben\n",
+    "    print(\"Accuracy:\", accuracy_score(y, preds))\n",
+    "    print(\"F1:\", f1_score(y, preds))\n",
+    "    print(\"AUC:\", roc_auc_score(y, preds))\n",
+    "    print(\"Confusion:\\n\", confusion_matrix(y, preds))\n",
+    "    print(classification_report(y, preds))\n",
+    "\n",
+    "    # Confusion Matrix plotten\n",
+    "    def plot_confusion_matrix(true_labels, predictions, label_names):\n",
+    "        for normalize in [None, 'true']:\n",
+    "            cm = confusion_matrix(true_labels, predictions, normalize=normalize)\n",
+    "            cm_disp = ConfusionMatrixDisplay(cm,  display_labels=label_names)\n",
+    "            cm_disp.plot(cmap=\"Blues\")\n",
+    "    #cm = confusion_matrix(y, preds)\n",
+    "    plot_confusion_matrix(y,preds, label_names=['Low','High'])\n",
+    "    # plt.figure(figsize=(5,4))\n",
+    "    # sns.heatmap(cm, annot=True, fmt=\"d\", cmap=\"Blues\", cbar=False,\n",
+    "    #             xticklabels=[\"Predicted low\", \"Predicted high\"],\n",
+    "    #             yticklabels=[\"Actual low\", \"Actual high\"])\n",
+    "    # plt.title(f\"Confusion Matrix - {title}\")\n",
+    "    # plt.ylabel(\"True label\")\n",
+    "    # plt.xlabel(\"Predicted label\")\n",
+    "    # plt.show()\n",
+    "\n",
+    "# Aufrufen für Train/Val/Test\n",
+    "print(\"TRAIN:\")\n",
+    "evaluate(model, X_train, y_train, title=\"Train\")\n",
+    "\n",
+    "print(\"VAL:\")\n",
+    "evaluate(model, X_val, y_val, title=\"Validation\")\n",
+    "\n",
+    "print(\"TEST:\")\n",
+    "evaluate(model, X_test, y_test, title=\"Test\")\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "c43b0c80",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "joblib.dump(model, \"xgb_model_with_MAD.joblib\")\n",
+    "joblib.dump(normalizer, \"normalizer_with_MAD.joblib\")\n",
+    "print(\"Model gespeichert.\")\n",
+    "\n",
+    "model.save_model(\"xgb_model_with_MAD.json\")   # als JSON (lesbar, portabel)\n",
+    "model.save_model(\"xgb_model_with_MAD.bin\")    # als Binärdatei (kompakt)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "3195cc84",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import os\n",
+    "os.getcwd()\n",
+    "\n"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.12.10"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

model_training/xgboost/xgboost_with_MAD.ipynb

@@ -76,34 +76,6 @@
     "print(f\"high all: {high_all.shape}\")"
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "dbb58abd",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "import numpy as np\n",
-    "import pandas as pd\n",
-    "from sklearn.preprocessing import StandardScaler, MinMaxScaler\n",
-    "\n",
-    "def mad_outlier_removal(df, columns, threshold=3.5):\n",
-    "    \"\"\"\n",
-    "    Entfernt Ausreißer basierend auf Median Absolute Deviation (MAD).\n",
-    "    threshold: typischer Wert ist 3.5 (entspricht robustem Z-Score Cutoff).\n",
-    "    \"\"\"\n",
-    "    df_clean = df.copy()\n",
-    "    for col in columns:\n",
-    "        median = df_clean[col].median()\n",
-    "        mad = np.median(np.abs(df_clean[col] - median))\n",
-    "        if mad == 0:\n",
-    "            continue  # keine Streuung, keine Ausreißer\n",
-    "        robust_z = 0.6745 * (df_clean[col] - median) / mad\n",
-    "        mask = np.abs(robust_z) <= threshold\n",
-    "        df_clean = df_clean[mask]\n",
-    "    return df_clean"
-   ]
-  },
   {
    "cell_type": "code",
    "execution_count": null,
@@ -244,6 +216,58 @@
     "    return df_out"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "dbb58abd",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import numpy as np\n",
+    "import pandas as pd\n",
+    "\n",
+    "def calculate_mad_params(df, columns):\n",
+    "    \"\"\"\n",
+    "    Calculate median and MAD parameters for each column.\n",
+    "    This should be run ONLY on the training data.\n",
+    "    \n",
+    "    Returns a dictionary: {col: (median, mad)}\n",
+    "    \"\"\"\n",
+    "    params = {}\n",
+    "    for col in columns:\n",
+    "        median = df[col].median()\n",
+    "        mad = np.median(np.abs(df[col] - median))\n",
+    "        params[col] = (median, mad)\n",
+    "    return params\n",
+    "\n",
+    "\n",
+    "def apply_mad_filter(df, params, threshold=3.5):\n",
+    "    \"\"\"\n",
+    "    Apply MAD-based outlier removal using precomputed parameters.\n",
+    "    Works on training, validation, and test data.\n",
+    "    \n",
+    "    df: DataFrame to filter\n",
+    "    params: dictionary {col: (median, mad)} from training data\n",
+    "    threshold: cutoff for robust Z-score\n",
+    "    \"\"\"\n",
+    "    df_clean = df.copy()\n",
+    "\n",
+    "    for col, (median, mad) in params.items():\n",
+    "        if mad == 0:\n",
+    "            continue  # no spread; nothing to remove for this column\n",
+    "\n",
+    "        robust_z = 0.6745 * (df_clean[col] - median) / mad\n",
+    "        outlier_mask = np.abs(robust_z) > threshold\n",
+    "\n",
+    "        # Remove values only in this specific column\n",
+    "        df_clean.loc[outlier_mask, col] = np.nan\n",
+    "        print(df_clean.shape)\n",
+    "        \n",
+    "    print(df_clean.shape)\n",
+    "    return df_clean\n",
+    "\n"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,
@@ -251,11 +275,13 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "train_outlier_removed = mad_outlier_removal(train_df, au_columns, 50)\n",
+    "# Step 1: Fit parameters on training data\n",
-    "val_outlier_removed = mad_outlier_removal(val_df, au_columns, 50)\n",
+    "params = calculate_mad_params(train_df, au_columns)\n",
-    "test_outlier_removed = mad_outlier_removal(test_df, au_columns, 50)\n",
+    "\n",
-    "print(train_df.shape)\n",
+    "# Step 2: Apply filter consistently\n",
-    "print(train_outlier_removed.shape)"
+    "train_outlier_removed = apply_mad_filter(train_df, params, threshold=3.5)\n",
+    "val_outlier_removed   = apply_mad_filter(val_df, params, threshold=50)\n",
+    "test_outlier_removed  = apply_mad_filter(test_df, params, threshold=50)"
    ]
   },
   {
@@ -296,7 +322,8 @@
     "    objective=\"binary:logistic\",\n",
     "    eval_metric=\"auc\",\n",
     "    use_label_encoder=False,\n",
-    "    random_state=42\n",
+    "    random_state=42,\n",
+    "    verbosity=0,\n",
     ")\n",
     "\n",
     "# Parameter-Raster\n",
@@ -318,11 +345,14 @@
     "    scoring=\"roc_auc\",\n",
     "    n_jobs=-1,\n",
     "    cv=cv,\n",
-    "    verbose=2\n",
+    "    verbose=0\n",
     ")\n",
     "\n",
     "# Training mit Cross Validation\n",
-    "grid_search.fit(X_train, y_train)\n",
+    "grid_search.fit(\n",
+    "  X_train, y_train, \n",
+    "  verbose=False,\n",
+    "                )\n",
     "\n",
     "print(\"Beste Parameter:\", grid_search.best_params_)\n",
     "print(\"Bestes AUC:\", grid_search.best_score_)\n",