topic Setting Seed for Cross validation in proc forest in SAS Data Science

Setting Seed for Cross validation in proc forest

NichelleT — Thu, 18 Feb 2021 07:34:51 GMT

Hi everyone,

Is there a way to set a seed for the cross validation statement in proc forest. I have already set a seed for the proc forest model and it states in the log that the cross validation is using the same seed. However I noticed that when I run the same model multiple times, the Cross-Validation Fit Statistics differs for each run.

Re: Setting Seed for Cross validation in proc forest

Rick_SAS — Thu, 18 Feb 2021 21:32:54 GMT

Post your code and the log so we can see what kind of cross-validation you are using.

If you are using the PARTITION statement, you can set the seed there.

Some types of cross-validation do not use randomness. For example, you can put observations 1,6,11,16,... into one fold, observations 2,7,12,17,... into the second fold, etc.

Re: Setting Seed for Cross validation in proc forest

NichelleT — Thu, 25 Feb 2021 08:51:22 GMT

Hi Rick,

This is the code for the proc forest procedure with cross validation statement:

proc forest data=MODELINPUT seed=1854718772
maxdepth=29
ntree=100;
target tgt / level=interval;
input var1 var2 var3 var4 var5 var6 var7 var8 var9 / level=interval;
input var10 / level=nominal;
crossvalidation kfold=5;
run;

Also, I have attached a screenshot of the log.

Thank you!