topic Compute Mean Square Error on validation data set in SAS Data Science

Compute Mean Square Error on validation data set

John4 — Mon, 20 Mar 2017 09:48:31 GMT

I split my data into 70% training 20% validation and 10% test sets.

and i run this code :

proc genmod data=valid;
class sexe country job;
model price = country sexe job / dist=gamma link=log;
run;

how can I compute Mean Square Error on validation data set?

Re: Compute Mean Square Error on validation data set

Ksharp — Mon, 20 Mar 2017 10:31:14 GMT

Use CODE statement to get VALID table 's predict value.

Once you got predict value , it is easy to get MSE.

proc genmod data=train;
class sex country job;
model price=country sex job/dist=gamma link=log;
code file='c:\mycode.sas';
run;
data Score;
set vallid(rename=(price=old_price)) end=last;
%include 'c:\mycode.sas';

sum+(price-old_price)**2;
if last then mse=sum/_n_;
run;

Re: Compute Mean Square Error on validation data set

John4 — Mon, 20 Mar 2017 11:11:30 GMT

why it's

proc genmod data=train;

and not :

proc genmod data=valid;

I change into "valid" but I got zero as MSE. Is there any function to compute MSE on SAS ?

Re: Compute Mean Square Error on validation data set

Ksharp — Mon, 20 Mar 2017 11:21:56 GMT

use TRAIN to get parameter estimator , so you can use it to score VALID data.

Yes.there is function MSE to compute, check documentation and better use IML code to get MSE.

Re: Compute Mean Square Error on validation data set

Reeza — Mon, 20 Mar 2017 11:32:03 GMT

1. Use STORE statement in PROC GENMOD to get model

2. Use PROC PLM with FITSTATISTICS to score dataset and generate stats.