kdd2017---踩坑

最新推荐文章于 2024-01-21 09:29:07 发布

zjy3496

最新推荐文章于 2024-01-21 09:29:07 发布

阅读量2.9k

点赞数

分类专栏： python 文章标签：机器学习入门

本文链接：https://blog.csdn.net/zjy3496/article/details/70215022

版权

python 专栏收录该内容

2 篇文章 0 订阅

订阅专栏

kdd2017

1、XGBRegressor之objective和eval_metric

更改模型的目标函数和评价指标如下所示：

def objective(preds, dtrain):
    labels = dtrain#.get_label()
    grad = (preds - labels)*-1~~~#此处省略
    hess = np.ones_like(grad)
    return grad, hess


def eval_metric(preds,dtrain):
    truth = dtrain
    return 'error', np.mean(np.abs(preds-truth)/truth)

按照比赛评价指标更改后的目标函数和评价指标对模型训练没有效果。
目标函数无法直接求导，因此先对其进行了平方。一阶导与rmse求出的目标函数只有系数不同，二阶导则同样为常数。

比赛评价指标如下所示：

Evaluation Metrics
We choose Mean Absolute Percentage Error (MAPE) to evaluate the result.
Task 1: Let drt and prt be the actual and predicted average travel time for route r during time window t. The MAPE for travel time prediction is defined as:

$M A P E = 1 R \sum r = 1 R (1 T ∣ ∣ ∣ d r t - p r t d r t ∣ ∣ ∣)$ $MAPE = \frac{1}{R} \sum_{r=1}^{R} \left ( \frac{1}{T} \left | \frac{d_{rt} - p_{rt}}{d_{rt}} \right | \right )$
R and T are the number of routes and number of to-predict time windows in the testing period respectively.

Task 2: Let C be the number of tollgate-direction pairs (as aforementioned: 1-entry, 1-exit, 2-entry, 3-entry and 3-exit), T be the number of time windows in the testing period, and fct and pct be the actual and predicted traffic volume for a specific tollgate-direction pair c during time window t. The MAPE for traffic volume prediction is defined as:

$M A P E = 1 C \sum c = 1 C (1 T ∣ ∣ ∣ f c t - p c t f c t ∣ ∣ ∣)$ $MAPE = \frac{1}{C} \sum_{c=1}^{C} \left ( \frac{1}{T} \left | \frac{f_{ct} - p_{ct}}{f_{ct}} \right | \right )$