fix documentation issue, and format with ruff

Author: paulbkoch

docs/interpret/python/examples/quantile-regression.ipynb

@@ -3,23 +3,7 @@
   {
    "cell_type": "markdown",
    "metadata": {},
-   "source": [
-    "# EBM Internals - Quantile Regression\n",
-    "\n",
-    "This notebook covers quantile regression using pinball loss with Explainable Boosting Machines. For standard regression internals, see [Part 1](./ebm-internals-regression.ipynb). For classification, see [Part 2](./ebm-internals-classification.ipynb). For multiclass, see [Part 3](./ebm-internals-multiclass.ipynb).\n",
-    "\n",
-    "Standard regression models (e.g. with RMSE) predict the conditional mean of the target. Quantile regression instead predicts a specific quantile (e.g. median, 10th percentile, 90th percentile). This is useful for:\n",
-    "\n",
-    "- **Prediction intervals**: Fit models at the 10th and 90th percentiles to get an 80% prediction interval.\n",
-    "- **Asymmetric risk**: When over-predicting and under-predicting have different costs.\n",
-    "- **Robustness**: Median regression (alpha=0.5) is more robust to outliers than mean regression.\n",
-    "\n",
-    "EBMs support quantile regression via the `\"quantile\"` objective, which uses the pinball loss (also called the quantile loss). The pinball loss for quantile alpha is:\n",
-    "\n",
-    "$$L(y, \\hat{y}) = \\begin{cases} \\alpha \\cdot (y - \\hat{y}) & \\text{if } y \\geq \\hat{y} \\\\ (1 - \\alpha) \\cdot (\\hat{y} - y) & \\text{if } y < \\hat{y} \\end{cases}$$\n",
-    "\n",
-    "This loss penalizes under-predictions by a factor of alpha and over-predictions by a factor of (1 - alpha), causing the model to learn the alpha-quantile of the conditional distribution."
-   ]
+   "source": "# EBM Internals - Quantile Regression\n\nThis notebook covers quantile regression using pinball loss with Explainable Boosting Machines.\n\nStandard regression models (e.g. with RMSE) predict the conditional mean of the target. Quantile regression instead predicts a specific quantile (e.g. median, 10th percentile, 90th percentile). This is useful for:\n\n- **Prediction intervals**: Fit models at the 10th and 90th percentiles to get an 80% prediction interval.\n- **Asymmetric risk**: When over-predicting and under-predicting have different costs.\n- **Robustness**: Median regression (alpha=0.5) is more robust to outliers than mean regression.\n\nEBMs support quantile regression via the `\"quantile\"` objective, which uses the pinball loss (also called the quantile loss). The pinball loss for quantile alpha is:\n\n$$L(y, \\hat{y}) = \\begin{cases} \\alpha \\cdot (y - \\hat{y}) & \\text{if } y \\geq \\hat{y} \\\\ (1 - \\alpha) \\cdot (\\hat{y} - y) & \\text{if } y < \\hat{y} \\end{cases}$$\n\nThis loss penalizes under-predictions by a factor of alpha and over-predictions by a factor of (1 - alpha), causing the model to learn the alpha-quantile of the conditional distribution."
   },
   {
    "cell_type": "code",
@@ -285,15 +269,7 @@
   {
    "cell_type": "markdown",
    "metadata": {},
-   "source": [
-    "## Summary\n",
-    "\n",
-    "- Use `objective=\"quantile:alpha=0.5\"` for median regression, or any alpha in (0, 1) for other quantiles.\n",
-    "- The prediction mechanism is identical to standard regression EBMs (intercept + additive score lookups). Only the training loss function changes.\n",
-    "- Fitting multiple quantile models (e.g. alpha=0.1 and alpha=0.9) provides interpretable prediction intervals.\n",
-    "- All EBM interpretability tools (global/local explanations) work with quantile models.\n",
-    "- For the complete prediction code that handles interactions, missing values, and all model types, see [Part 3](./ebm-internals-multiclass.ipynb)."
-   ]
+   "source": "## Summary\n\n- Use `objective=\"quantile:alpha=0.5\"` for median regression, or any alpha in (0, 1) for other quantiles.\n- The prediction mechanism is identical to standard regression EBMs (intercept + additive score lookups). Only the training loss function changes.\n- Fitting multiple quantile models (e.g. alpha=0.1 and alpha=0.9) provides interpretable prediction intervals.\n- All EBM interpretability tools (global/local explanations) work with quantile models."
   }
  ],
  "metadata": {
@@ -317,4 +293,4 @@
  },
  "nbformat": 4,
  "nbformat_minor": 4
-}
+}

python/interpret-core/interpret/utils/_misc.py

@@ -28,7 +28,7 @@ def normalize_objective(objective):
                     FutureWarning,
                     stacklevel=2,
                 )
-                objective = new_name + objective[len(old_name):]
+                objective = new_name + objective[len(old_name) :]
                 break
     return objective