查詢結果分析
來源資料
相關文獻
- 最短敘述長度與線性模型選取
- 階層線性模式在教育縱貫研究上的應用: 以數學學習機會為例
- 臺灣汽保費率之估計--對數線性費率模式與類神經網路之比較
- Some Statistical and Logical Considerations When Rescoring Tests
- Asymptotic Properties of the Empirical BLUP and BLUE in Mixed Linear Models
- 施肥效應之統計分析與探討--以扁柏苗木施肥為例
- 非線性模式下最佳階梯式收費結構之擁擠減少效果之研究
- Finite Mixture Multivariate Generalized Linear Models Using Gibbs Sampling and E-M Algorithms
- 窄裙活動舒適性之客觀評估(Ⅲ)--簡單模式
- Fitting Non-linear Models with AR(1) Structure: A Marginal Likelihood Approach
頁籤選單縮合
題 名 | 最短敘述長度與線性模型選取=Minimum Description Length and Linear Model Selection |
---|---|
作 者 | 張世宗; 史玉山; | 書刊名 | 中國統計學報 |
卷 期 | 40:1 2002.03[民91.03] |
頁 次 | 頁77-95 |
分類號 | 319.711 |
關鍵詞 | 字串編碼; 最短敘述長度; 模型選取; 線性模式; Coding; Linear model; Minimum description length; Model selection; |
語 文 | 中文(Chinese) |
中文摘要 | 在資料編碼的問題上,常要考慮編碼所須花費的成本與其有效性。利用此一原則所推導出來的最短敘述長度方法,不但具有統計相關性質,且適用於模型選取。本文乃針對由最短敘述長度方法所建構的gMDL準則,藉由摸摸實驗來探討其在小樣本下線性模型選取的表現,並與AIC及BIC準則做比較。我們發現在有限樣本下,一般而言gMDL準則在線性模型選取土的表現優於AIC或BIC準則。但若非零的迴歸參數值中有較小的值出現時,gMDL準則比起 其它兩種準則容易忽略該參數,以致降低了選中正確的模型的準確率。而此情形可透過樣本數的增加而獲得改善。 |
英文摘要 | In this paper, we briefly survey the connection between minimum description length principle derived from Information science and statistical model selection. A model selection criterion, gMDL, was introduced based on this principle and Bayesian analysis. The gMDL criterion is compared with the AIC and the BIC criteria in some small sample settings. In the simulation studies, We find gMDL generally tends to select the correct model more frequently. Some results also reveal its weakness which could be corrected by increasing sample size. |
本系統中英文摘要資訊取自各篇刊載內容。