English  |  正體中文  |  简体中文  |  全文筆數/總筆數 : 6507/11669
造訪人次 : 30120626      線上人數 : 488
RC Version 3.2 © Powered By DSPACE, MIT. Enhanced by NTU Library IR team.
搜尋範圍 進階搜尋

請使用永久網址來引用或連結此文件: http://ir.ncue.edu.tw/ir/handle/987654321/14054

題名: 樣條函數中平滑參數選取方法
Selection Method of the Smoothing Parameter for Spline Function
作者: 陳春樹
貢獻者: 統計資訊研究所
日期: 2008
上傳時間: 2012-09-10T04:40:00Z
出版者: 行政院國家科學委員會
摘要: 在無母數迴歸方法中,平滑樣條(smoothing spline)的方法常被使用做曲線的配適。而此方法的關鍵在於平滑參數的選取。過去的文獻中,有許多學者提出選取平滑參數的準則。例如:Cp (Mallows 1973),generalized maximum likelihood (GML) (Wecker and Ansley 1983; Wahba 1985),和extended exponential (EE) 準則 (Kou and Efron 2002)。雖然Li (1986) 已經證明Cp 在誤差平方和之下具有近似最佳的性質,但是Kou 和Efron (2002) 研究上述三個選取準則的幾何性質,說明了Cp 準則在有限樣本的情形下具有較GML 和EE 高的選取變異,因而在有限樣本的情形下表現的較不理想。相反的,雖然GML 和EE 的幾何性質較穩定,但二者卻沒有和Cp 一樣具有近似最佳的性質。因此本計畫的主要構想是:提出一個選取平滑參數的準則,使其在有限樣本的情形下能有好的表現,同時也能具有與Cp 一樣的近似最佳性質。
In nonparametric regression, smoothing splines are a popular method for curve fitting, in which selection of the smoothing parameter is crucial. In the past literatures, there are many scholars who proposed some criteria for selecting the smoothing parameter, such as Cp (Mallows 1973), generalized maximum likelihood (GML) (Wecker and Ansley 1983; Wahba 1985), and the extended exponential (EE) criterion (Kou and Efron 2002). Although Cp has been shown to be asymptotically optimal under the squared error loss (Li 1986), Kou and Efron (2002) utilized a geometric approach to show that Cp has a higher variability than GML and EE for small to moderately large sample sizes. On the other hand, GML and EE are more stable than Cp, but they do not possess the same asymptotic optimality as Cp. Therefore, in this research project, a new selection criterion would be proposed which is expected to be superior and more stable than Cp for small to moderately large sample sizes, and possesses the same asymptotic optimality as Cp. 表
關聯: 國科會計畫, 計畫編號: NSC97-2118-M018-003; 研究期間: 9710-9807
顯示於類別:[統計資訊研究所] 國科會計畫

文件中的檔案:

檔案 大小格式瀏覽次數
2020800312002.pdf50KbAdobe PDF537檢視/開啟


在NCUEIR中所有的資料項目都受到原著作權保護.

 


DSpace Software Copyright © 2002-2004  MIT &  Hewlett-Packard  /   Enhanced by   NTU Library IR team Copyright ©   - 回饋