SAS ROUTINE FOR PREDICTORS RANKING BY MAXIMIZED CHI-SQUARE

shiyiming · 发表于 2010-10-22 13:24:02

From Wensui Liu's blog

******************************************; * A SAS ROUTINE FOR PREDICTORS RANKING *; * BY MAXIMIZED CHI-SQUARE BASED UPON THE *; * BINARY SPLIT IMPLEMENTED IN DMSPLIT *; * PROCEDURE IN ENTERPRISE MINER. *; * -------------------------------------- *; * author: <a href="mailto:wensliu@paypal.com">wensliu@paypal.com</a> *; ******************************************; libname data 'D:\projects\woe\data'; options mprint mlogic; %let varlist = x2 x3 x4 x5 x10 x11 x12 x13 x14 x15; %macro dmsplit(data = , y = , x = &varlist); %let i = 1; %local i; data _tmp1(keep = &y &varlist); set &data; where &y in (0, 1); run; proc sql; create table _out ( variable char(32), type char(1), chi_sq num ); quit; %do %while (%scan(&varlist, &i) ne %str()); %let var = %scan(&varlist, &i); data _tmp2(keep = &y &var); set _tmp1; if _n_ = 1 then do; call symput('vtype', vtype(&var)); end; run; proc dmdb data = _tmp2 out = _db1 dmdbcat = _ct1; %if &vtype = C %then %do; class &y &var; %end; %else %if &vtype = N %then %do; class &y; var &var; %end; run; proc dmsplit data = _db1 dmdbcat = _ct1 outvars = _tmp3 noprint passes = 5; var &var; target &y; run; %if %sysfunc(exist(_tmp3)) %then %do; proc sql noprint; select count(*) into :nobs from _tmp3; quit; %if &nobs > 0 %then %do; proc sql; insert into _out select upcase(_split_), "&vtype", round(_chisqu_, 0.01) from _tmp3 where _parent_ = 0; quit; %end; %else %do; options obs = max nosyntaxcheck; proc sql; insert into _out values("%upcase(&var)", "&vtype", .);; quit; %end; %end; proc datasets library = work nolist; delete _tmp2 _tmp3 / memtype = data; run; quit; %let i = %eval(&i + 1); %end; proc format; picture chi_fmt . = 'N / A'; run; proc sort data = _out; by descending chi_sq; run; proc report data = _out box spacing = 1 split = "*"; column("Predictors Ranking by*Maximized Chi-Square Based on Binary Cut" variable type chi_sq); define variable / "Predictor" width = 20 center; define type / "Type" width = 10 center; define chi_sq / "ChiSQ" width = 15 center format = chi_fmt.; run; %mend dmsplit; %dmsplit(data = data.credit, y = y, x = &varlist); +-----------------------------------------------+ | Predictors Ranking by | | Maximized Chi-Square Based on Binary Cut | | Predictor Type ChiSQ | |-----------------------------------------------| | X11 | C | 91.51 | |--------------------+----------+---------------| | X3 | N | 34.26 | |--------------------+----------+---------------| | X10 | C | 19.99 | |--------------------+----------+---------------| | X2 | N | 14.36 | |--------------------+----------+---------------| | X4 | N | 7.39 | |--------------------+----------+---------------| | X12 | C | 6.17 | |--------------------+----------+---------------| | X5 | N | 3.95 | |--------------------+----------+---------------| | X14 | C | 2.06 | |--------------------+----------+---------------| | X13 | C | 1.79 | |--------------------+----------+---------------| | X15 | C | N / A | +-----------------------------------------------+

		自动登录	找回密码
密码			立即注册