查詢結果分析
來源資料
頁籤選單縮合
題名 | 資料探勘--超大型資料庫基本統計量的計算=Mining the Descriptive Statistics of a Large Data Bases |
---|---|
作者 | 馬瀰嘉; 蘇佩芳; 林共進; | 書刊名 | 中國統計學報 |
卷期 | 41:1 2003.03[民92.03] |
頁次 | 頁97-116 |
分類號 | 319.5 |
關鍵詞 | 中位數; 平均數; 百分位數; 相關係數; 變異數; Correlation coefficient; Mean; Median; Quantile; Variance; |
語文 | 中文(Chinese) |
中文摘要 | 針對一組與時俱增的龐大資料,日後若欲計算其基本統計量,如:平均數、變異數、中位數、百分位數或相關係數等,就必需一直儲存這些資料,但面對如此龐大的資料,在計算中位數或百分位數等需先將資料排序的統計量時,由於計算機能力的限制,如資料超過允許最大的陣列數,或記憶體不足的問題,造成電腦不能計算。因此如何找出一種逐步遞迴的方法:每隔一段時間後,丟棄原始的資料,僅保留幾個具代表性的統計量,使得日後欲計算先前所有資料的基本統計量時,可由計算這些各區段統計量獲得。本文提出二階段組合中心趨勢的統計量,如:平均數、中位數、值域中點和利用Han & Kamber (2000) 書中近似中位數公式來求得前k段資料的近似中位數和近似百分位數,並利用統計模擬來比較不同近似方法的好壞。 |
英文摘要 | The fast-growing and tremendous amount of data are collected and stored in computers. When the descriptive statistics like mean, variance, median, quantile and correlation coefficient are computed for a large data set, the computer is limited in its memory spare. If the raw data is discarded at fixed periods, one wants to find a recursive method by using some representative statistics of subgroup data to compute the descriptive statistics of accumulative and numerous data bases. This paper proposes a two-stages method of approximate median (or quantile) by taking median (or quantile) of subgroup data in the first stage and taking mean, median and midrange of the previous medians (or quantiles) in the second stage. The approximate median (or quantile) formula of Han and Kamber (2000) is also considered. Furthermore, the statistical properties of some procedures are discussed. Finally, these methods are compared by the moving mean squared error in simulation. |
本系統之摘要資訊系依該期刊論文摘要之資訊為主。