EDA 常用的統計方法

先整理一下 有空再詳細寫
1.基本統計量 mean/std/min/max/percentile/
2.distribution verify and display
3.testing hypotheses
4.DOE
5.regression
6.SPC

進階分析
1.MVA(PCA/FDA)
2.Data mining

問題分類
1.Process control
2.Root cause analysis
3.Yield impact
4.Yield prediction
5.Wafer map analysis

R package

The advantage of using R packages. How to create your own R packages and a little about S3 and S4 class.

http://epub.ub.uni-muenchen.de/6175/1/tr036.pdf

我的31歲

30歲那年的年尾 離開了生平第一個工作的公司
31歲那年 為自己的理想打拼 為自己的人生努力

那年 讓我學看到很多 學到很多
沒有當時的歷練 就不會有現在的自己

現在雖然離開了當時努力的公司 仍是感謝有過那段經驗

How many software packages is too much?

看了這篇
How many software packages is too much後的想法 一套DM package就夠了
重要的是對DM method的瞭解 知道什麼方法可以用在什麼樣的問題以及它的限制 用不同software的不同方法會有不同結果通常是因為一些預設值不同造成的 所以在用DM解決問題時 還是要回到最基本的功夫上

mixed type sort

It's a common problem in semiconductor EDA analysis which plot by lot+wafer number as axis ticks. But the usual sort cannot get the right order for character embed numbers.
The package gtools has the "mixedsort" method for solving this problem.

	>library(gtools)
	>x <- paste("AA_",1:24, sep="")
	>(y <- sort(x))
	[1] "AA_1" "AA_10" "AA_11" "AA_12" "AA_13" "AA_14" "AA_15" "AA_16" "AA_17" "AA_18" AA_19"
	[12] "AA_2" "AA_20" "AA_21" "AA_22" "AA_23" "AA_24" "AA_3" "AA_4" "AA_5" "AA_6" "AA_7"
	[23] "AA_8" "AA_9"
	>(y1 <- mixedsort(x))
	[1] "AA_1" "AA_2" "AA_3" "AA_4" "AA_5" "AA_6" "AA_7" "AA_8" "AA_9" "AA_10" "AA_11"
	[12]"AA_12" "AA_13" "AA_14" "AA_15" "AA_16" "AA_17" "AA_18" "AA_19" "AA_20" "AA_21" "AA_22"
	[23]"AA_23" "AA_24"
	#you can use mixedorder to get the order
	> (y1 <- order(x))
	[1] 1 10 11 12 13 14 15 16 17 18 19 2 20 21 22 23 24 3 4 5 6 7 8 9
	> (y2 <- mixedorder(x))
	[1] 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

view raw mixedsort hosted with ❤ by GitHub

Data Modeling

好文
如何建構與評估經濟理論模型

"1.經濟理論模型必須要有實證動機;2. 經濟理論模型必須能夠透過估計檢定或是模擬校準(calibra-
tion) 等方式, 評估該理論模型的良莠"


同樣的對於統計模型 DM modeling 也是吧

Axis scaling algorithm

A good look axes scale usually increase by 10^(x) or 0.5*10^(x) according to the data's range. Dorothy E. Pugh provide an algorithm (and SAS Macro) for generated suitable axis scale, I modified the SAS code into R code

	getAxisInterval <- function(x){
	threshold <- c(3.162, 1.25, 1)
	y <- range(x)
	z <- format(diff(y), digits = 7, scientific = T)
	tmp <- unlist(strsplit(z, "e"))
	exppart <- as.numeric(tmp[2])
	mantissa <- round(abs(as.numeric(tmp[1])), digit = 3)
	intby <- switch(min(which(mantissa>=threshold, arr.ind = TRUE)), 10^exppart,
	0.5*10^exppart, 10^(expart-1), 10^exppart)
	#original limits
	#tmp <- (0.5+0.01)*intby
	#axislimit <- round(range(x) + c(-tmp, tmp), digit = -(exppart-1))
	axislimit <- c(floor(y[1]/intby)*intby, ceiling(y[2]/intby)*intby)
	return(list(axislimit = axislimit, stepBy = intby))
	}
	#for specified ticks location you can use
	#ats <- seq(axislimit[1], axislimit[2], intby)

view raw axisScaling hosted with ❤ by GitHub

reference:Dorothy E. Pugh(SUGI25), "A Robust Generalized Axis-scaling Macro"