決策樹也可以做特征分析啦

那么這個代碼是用于建模初期，你為了大概了解變量的一個基本特征寫的，不是最優(yōu)分組哈，因為這個代碼是將變量最多分為12組，分這么多組的原因也是為了更好的觀察特征而已啦，你要是覺得太多組，你可以改下樹的深度這些調整一下，這里關于變量特征怎么看，我就不說了.....

%macro

zhandapao(data,DVAR,id,dir);

proc datasets lib=work nodetails;

delete

varname_total;

run;

/*建立數值型數據集*/

%let lib=%upcase(%scan(&data.,1,'.'));

%letdname=%upcase(%scan(&data.,2,'.'));

%globalvar_list var_num;

proc sql noprint;

select name,count(*) into :var_list separated by' ',:var_num

from sashelp.VCOLUMN

where left(libname)="&lib."and

left(memname)="&dname."and

type="num"and

lowcase(name)^=lowcase("&DVAR.")

and lowcase(name)^="&id.";

quit;

%put

&var_list.;

/*把數值型變量定義為宏變量*/

%doi=1%to&var_num.;

%letnumvar_name_&i.=%scan(&var_list.,&i.);

%put&numvar_name_1.;

proc split data=&data.splitsize=300

maxbranch=2

MAXDEPTH=5nsurrs=5

assess=lift criterion=gini;

input &&numvar_name_&i./level=interval;

target &DVAR./level=binary;

Score data=&data.out=d_&&numvar_name_&i.;

code file="&dir.treecode_tic_&&numvar_name_&i..sas";

describe file="&dir.treerule_tic_&&numvar_name_&i..txt";

run;

data n_D_&&numvar_name_&i.;

set d_&&numvar_name_&i.;

%include"&dir.treecode_tic_&&numvar_name_&i..sas";

rename p_&DVAR.1=p_&&numvar_name_&i.;

run;

proc sql noprint;

select count(*),max(&&numvar_name_&i.),min(&&numvar_name_&i.)into:total, :max ,:min from n_D_&&numvar_name_&i.;

quit;

data n_D_&&numvar_name_&i.;

set n_D_&&numvar_name_&i.;

if &min.<=&&numvar_name_&i.<=&max.

then flag="no_null";

else flag="null";

run;

proc sql;

select count(*) into:is_null from

n_D_&&numvar_name_&i.;

quit;

%if&is_null.>0%then%do;

proc sql noprint;

select count(*),max(&&numvar_name_&i.),min(&&numvar_name_&i.)into:total,:max ,:min from n_D_&&numvar_name_&i.;

create table total as

select"&&numvar_name_&i."as

varname,

min(&&numvar_name_&i.) as interval_1,

max(&&numvar_name_&i.) as interval_2,

compress(put(min(round(&&numvar_name_&i.,0.0001)),best32.))||'-'||compress(put(max(round(&&numvar_name_&i.,0.0001)),best32.)) as interval,

sum(&DVAR.) as bad_num,

count(*) as total_num,

count(*)/&total.as num_rate,

sum(&DVAR.)/count(*) as bad_rate

from n_D_&&numvar_name_&i.

group by p_&&numvar_name_&i.

union all

select"&&numvar_name_&i."as varname,

-9999as interval_1,

-9999as interval_2,

'null'as interval,

sum(&DVAR.) as bad_num,

count(*) as total_num,

count(*)/&total.as num_rate,

sum(&DVAR.)/count(*) as bad_rate

from n_D_&&numvar_name_&i.(where=(&&numvar_name_&i.=.))

group by p_&&numvar_name_&i.

order by interval_1;

quit;

%end;

%else%do;

proc sql noprint;

select count(*),max(&&numvar_name_&i.),min(&&numvar_name_&i.)into:total,:max ,:min from n_D_&&numvar_name_&i.;

create table total as

select"&&numvar_name_&i."as varname,

min(&&numvar_name_&i.) asninterval_1,

max(&&numvar_name_&i.) as interval_2,

compress(put(min(round(&&numvar_name_&i.,0.0001)),best32.))||'-'||compress(put(max(round(&&numvar_name_&i.,0.0001)),best32.)) as interval,

sum(&DVAR.) as bad_num,

count(*) as total_num,

count(*)/&total.as num_rate,

sum(&DVAR.)/count(*) as bad_rate

from n_D_&&numvar_name_&i.

group by p_&&numvar_name_&i.

order by interval_1;

quit;

%end;

data &&numvar_name_&i.;

set total;

group=_n_;

run;

proc append base=varname_total

data=&&numvar_name_&i.

force;run;

proc datasets lib=work nodetails;

delete total n_: d_:

&&numvar_name_&i.

_namedat;

quit;

%end;

%mend;

解釋一下這個代碼怎么用，這個宏已經是封裝好了的，直接填入參數就可以用了：

zhandapao(data,DVAR,id,dir);

data:填入你的數據集

DVAR：填入你的因變量

id：填入你的數據集的主鍵

dir：這個你需要填一個路徑，是用來放決策樹的規(guī)則的文件下，決策樹的規(guī)則文件你看不懂沒關系，你填個類似“F/DD”的路徑就可以了。

例子：%zhandapao(DD.TEST_DATA,y,CUSTOMER_id,D:test_1);

結果圖就是這樣子：

那么今天的更新就到這里啦