透過您的圖書館登入
IP:18.191.180.252
  • 學位論文

貝氏迴歸分析在伯氏多項式上對S型之研究

Bayesian Regression with Sigmodial Random Bernstein Polynomials

指導教授 : 吳裕振

摘要


這篇論文主要是利用貝氏方法來研究 S 型的迴歸曲線,就是以隨機伯氏多項式為事前分佈,用馬可夫鏈蒙地卡羅 (M.C.M.C.) 之方法來計算事後分佈. 因為我們發現伯氏多項式可以很容易來描述幾何圖形,所以我們也找到了對伯氏多項式之係數加以限制,即可得到我們所要研究的迴歸曲線是 S 型. 而且我們也證明了所有的 S 型連續函數,都可以用我們找到限制條件下的伯氏多項式來逼近,因此我們就可用這篇論文主要是利用貝氏方法來研究 S 型的迴歸曲線,就是以隨機伯氏多項式為事前分佈,用馬可夫鏈蒙地卡羅 (M.C.M.C.) 之方法來計算事後分佈. 因為我們發現伯氏多項式可以很容易來描述幾何圖形,所以我們也對伯氏多項式之係數加以限制,即可得到我們所要研究的迴歸曲線是 S 型. 而且我們也證明了所有的 S 型連續函數,都可以用我們找到限制條件下的伯氏多項式來逼近,因此我們就可用它來當作我們要估計迴歸曲線的模型. 在我們拿到資料後,根據分析,確定圖形為 S 型,我們就可用本篇論文的方法來估計. 本論文的貝氏估計方法是用馬可夫鏈蒙地卡羅 (M.C.M.C.) 之方法來計算事後分佈,我們寫的演算法是容易被寫成程式(利用軟體 Matlab),所以容易計算, 而且所得到的估計也相當的不錯,這將呈現在論文的最後部份.因為在 M.L.E. 的部分還未有此估計,所以無法與 M.L.E. 做比較,但我們也可以當做未來研究的主題.

並列摘要


This paper is about the regression curve with sigmodial using the Bayesian method. Use random Bernstein polynomials as the prior, and calculate the posterior by using Markov Chain Monte Carlo (M.C.M.C.) method. Because we discovered that Bernstein polynomials may describe the geometric graph easily, therefore we also controled the coefficient of Bernstein polynomials. Then obtained the regression curve with sigmodial which we want to study. We had also proven that all continuous functions with sigmodial can be approached by contrained Bernstein polynomials. Therefore we can use it to be the model of regression curve which is going to be estimated. According to analyze the data which we got, we can confirm that graph is sigmodial. Hence we can use the method of this paper to estimate it. The method of Bayesian estimation of this paper is using Markov Chain Monte Carlo (M.C.M.C.) method to calculate the posterior. Our algorithm is easy to be written into the program (using the package software Matlab), whence it is easy to calculate. In fact the estimation which we obtain is quite good, present in the last of this paper. Because there is still no one estimating it by M.L.E., we can not compare with it. But it can be a good subject for the research in the future.

參考文獻


of signs revisited. Am. Math. Mon. 105 447-451.
Bayesian survival analysis using Bernstein polynomials. Scand.
J. of Statist. 32, 447-446.
Green, P. G. (1995). Reversible jump Markov chain Monte Carlo
computation and Bayesian model determination. Biometrika 71

延伸閱讀