# Custom Objective and Evaluation Metric

Contents

## Overview

XGBoost is designed to be an extensible library. One way to extend it is by providing our own objective function for training and corresponding metric for performance monitoring. This document introduces implementing a customized elementwise evaluation metric and objective for XGBoost. Although the introduction uses Python for demonstration, the concepts should be readily applicable to other language bindings.

Note

• The ranking task does not support customized functions.

• Breaking change was made in XGBoost 1.6.

In the following two sections, we will provide a step by step walk through of implementing the Squared Log Error (SLE) objective function:

$\frac{1}{2}[log(pred + 1) - log(label + 1)]^2$

and its default metric Root Mean Squared Log Error(RMSLE):

$\sqrt{\frac{1}{N}[log(pred + 1) - log(label + 1)]^2}$

Although XGBoost has native support for said functions, using it for demonstration provides us the opportunity of comparing the result from our own implementation and the one from XGBoost internal for learning purposes. After finishing this tutorial, we should be able to provide our own functions for rapid experiments. And at the end, we will provide some notes on non-identy link function along with examples of using custom metric and objective with scikit-learn interface. with scikit-learn interface.

## Customized Objective Function

During model training, the objective function plays an important role: provide gradient information, both first and second order gradient, based on model predictions and observed data labels (or targets). Therefore, a valid objective function should accept two inputs, namely prediction and labels. For implementing SLE, we define:

import numpy as np
import xgboost as xgb
from typing import Tuple

def gradient(predt: np.ndarray, dtrain: xgb.DMatrix) -> np.ndarray:
'''Compute the gradient squared log error.'''
y = dtrain.get_label()
return (np.log1p(predt) - np.log1p(y)) / (predt + 1)

def hessian(predt: np.ndarray, dtrain: xgb.DMatrix) -> np.ndarray:
'''Compute the hessian for squared log error.'''
y = dtrain.get_label()
return ((-np.log1p(predt) + np.log1p(y) + 1) /
np.power(predt + 1, 2))

def squared_log(predt: np.ndarray,
dtrain: xgb.DMatrix) -> Tuple[np.ndarray, np.ndarray]:
'''Squared Log Error objective. A simplified version for RMSLE used as
objective function.
'''
predt[predt < -1] = -1 + 1e-6
hess = hessian(predt, dtrain)


In the above code snippet, squared_log is the objective function we want. It accepts a numpy array predt as model prediction, and the training DMatrix for obtaining required information, including labels and weights (not used here). This objective is then used as a callback function for XGBoost during training by passing it as an argument to xgb.train:

xgb.train({'tree_method': 'hist', 'seed': 1994},  # any other tree method is fine.
dtrain=dtrain,
num_boost_round=10,
obj=squared_log)


Notice that in our definition of the objective, whether we subtract the labels from the prediction or the other way around is important. If you find the training error goes up instead of down, this might be the reason.

## Customized Metric Function

So after having a customized objective, we might also need a corresponding metric to monitor our model’s performance. As mentioned above, the default metric for SLE is RMSLE. Similarly we define another callback like function as the new metric:

def rmsle(predt: np.ndarray, dtrain: xgb.DMatrix) -> Tuple[str, float]:
''' Root mean squared log error metric.'''
y = dtrain.get_label()
predt[predt < -1] = -1 + 1e-6
elements = np.power(np.log1p(y) - np.log1p(predt), 2)
return 'PyRMSLE', float(np.sqrt(np.sum(elements) / len(y)))


Since we are demonstrating in Python, the metric or objective need not be a function, any callable object should suffice. Similar to the objective function, our metric also accepts predt and dtrain as inputs, but returns the name of the metric itself and a floating point value as the result. After passing it into XGBoost as argument of feval parameter:

xgb.train({'tree_method': 'hist', 'seed': 1994,
'disable_default_eval_metric': 1},
dtrain=dtrain,
num_boost_round=10,
obj=squared_log,
feval=rmsle,
evals=[(dtrain, 'dtrain'), (dtest, 'dtest')],
evals_result=results)


We will be able to see XGBoost printing something like:

 dtrain-PyRMSLE:1.37153  dtest-PyRMSLE:1.31487
 dtrain-PyRMSLE:1.26619  dtest-PyRMSLE:1.20899
 dtrain-PyRMSLE:1.17508  dtest-PyRMSLE:1.11629
 dtrain-PyRMSLE:1.09836  dtest-PyRMSLE:1.03871
 dtrain-PyRMSLE:1.03557  dtest-PyRMSLE:0.977186
 dtrain-PyRMSLE:0.985783 dtest-PyRMSLE:0.93057
...


Notice that the parameter disable_default_eval_metric is used to suppress the default metric in XGBoost.

For fully reproducible source code and comparison plots, see Demo for defining a custom regression objective and metric.

## Scikit-Learn Interface

The scikit-learn interface of XGBoost has some utilities to improve the integration with standard scikit-learn functions. For instance, after XGBoost 1.6.0 users can use the cost function (not scoring functions) from scikit-learn out of the box:

from sklearn.datasets import load_diabetes
from sklearn.metrics import mean_absolute_error
X, y = load_diabetes(return_X_y=True)
reg = xgb.XGBRegressor(
tree_method="hist",
eval_metric=mean_absolute_error,
)
reg.fit(X, y, eval_set=[(X, y)])


Also, for custom objective function, users can define the objective without having to access DMatrix:

def softprob_obj(labels: np.ndarray, predt: np.ndarray) -> Tuple[np.ndarray, np.ndarray]:
rows = labels.shape
grad = np.zeros((rows, classes), dtype=float)
hess = np.zeros((rows, classes), dtype=float)
eps = 1e-6
for r in range(predt.shape):
target = labels[r]
p = softmax(predt[r, :])
for c in range(predt.shape):
g = p[c] - 1.0 if c == target else p[c]
h = max((2.0 * p[c] * (1.0 - p[c])).item(), eps)
grad[r, c] = g
hess[r, c] = h