topic Re: Bootstrapping SAS EG vs SAS Viya in SAS Data Science

Bootstrapping SAS EG vs SAS Viya

smackerz1988 — Sun, 09 Jun 2024 20:52:40 GMT

Hi,

I'm currently waiting to get access for sas viya but I have a test case scenario to compare between sas eg and sas viya and just want to verify the sas viya code. For bootstrapping in sas eg I have this code

proc surveyselect data=data

out=BootSamples noprint

seed=25 

reps=2000   

method=urs

samprate=1

outhits;

run;

Am I right that this would be the equivalent code in sas viya or am i missing something?

/* Start CAS session and load data into CAS */
cas mysess sessopts=(caslib='casuser');
libname mylib cas sessref=mysess;

/* Load example data */
proc casutil;
    load data=data casout="sample" replace;
run;

/* Perform bootstrap resampling using the sampling action set */
proc cas;
    action sampling.srs result=r /
        table={caslib='casuser', name='sample'}
        output={casout={caslib='casuser', name='BootSamples', replace=true}}
        samppct=100 /* Sampling rate of 100% for bootstrap */
        method='URS' /* Unrestricted random sampling with replacement */
        seed=25 /* Seed for reproducibility */
        reps=2000; /* Number of bootstrap replicates */
        selection={name='Freq', includeFreq=true}; /* Include frequency counts in the output */
quit;

/* Fetch and display some of the bootstrap samples (Optional) */
proc cas;
    table.fetch / table={caslib='casuser', name='BootSamples'} to=10;
quit;

/* End CAS session */
cas mysess terminate;

Re: Bootstrapping SAS EG vs SAS Viya

sbxkoenk — Sun, 09 Jun 2024 22:03:28 GMT

Hello,

Method='URS' in sampling.srs action is not an existing parameter. Where did you get that from?

See :

Home > Analytics > SAS Data Science > CAS Bootstrapping
https://communities.sas.com/t5/SAS-Data-Science/CAS-Bootstrapping/m-p/931229#M10828

Koen

Re: Bootstrapping SAS EG vs SAS Viya

smackerz1988 — Sun, 09 Jun 2024 23:00:36 GMT

Thanks for the link @sbxkoenk .So if I understand correctly if I'm using sas viya 4 I should use method = SRS instead?

Re: Bootstrapping SAS EG vs SAS Viya

smackerz1988 — Mon, 10 Jun 2024 12:24:55 GMT

@sbxkoenk is this code more appropriate?

/* Perform bootstrap resampling using the sampling action set */
proc cas;
    action loadActionSet / actionSet='sampling';
quit;

proc cas;
    action sampling.srs result=r /
        table={caslib='casuser', name='sample'}
        output={casOut={caslib='casuser', name='BootSamples', replace=true}}
        samppct=100 /* Sampling rate of 100% for bootstrap */
        replace=True /* Sampling with replacement */
        seed=25 /* Seed for reproducibility */
        reps=2000; /* Number of bootstrap replicates */
        selection={name='Freq', includeFreq=true}; /* Include frequency counts in the output */
quit;

I am unable to test it on sas viya currently but would really appreciate if I can verify it is a more accurate reflection of sas viya 4 functionality so I can compare and contrast with sas 9.4

Re: Bootstrapping SAS EG vs SAS Viya

sbxkoenk — Mon, 10 Jun 2024 12:45:40 GMT

The code is NOT correct.

The below sampling.srs parameters do not exist:

replace=True /* Sampling with replacement */ reps=2000; /* Number of bootstrap replicates */

Here are a few links to get you started.

https://github.com/statmike/Resampling-Methods-in-SAS-Viya : Fast, easy resampling methods using the SAS Viya CAS Engine (bootstrap, double-Bootstrap, jackknife)
Bootstrap Resampling At Scale: Part 1 (of 3) dd. 03 March 2020
https://statmike.com/blog/sgf2020p1
Bootstrap Resampling At Scale: Part 2 (of 3) dd. 04 March 2020
https://statmike.com/blog/sgf2020p2
Bootstrap Resampling At Scale: Part 3 (of 3) dd. 05 March 2020
https://statmike.com/blog/sgf2020p3

Please learn about the APPLYROWORDER procedure-option (e.g. PROC PARTITION) and the ADDROWID data set option, as you will need these to ensure reproducibility.

Last but not least , PROC SURVEYSELECT runs on the compute server (SPRE) in SAS Viya 4 ... you can still use this SAS/STAT procedure (it's just not CAS-enabled).
You can turn to the CAS engine (distributed computing) for the remainder of your calculations.

Koen

Re: Bootstrapping SAS EG vs SAS Viya

smackerz1988 — Tue, 11 Jun 2024 09:29:39 GMT

This is basically the code that I need. Thanks for your help

/* Initialize CAS Session */
cas mysess sessopts=(caslib='casuser');
libname mylib cas sessref=mysess;

/* Load Dataset into CAS */
proc casutil;
    load data=sashelp.cars casout="sample" replace;
quit;

/* Load and Define Action Set */
proc cas;
    builtins.actionSetFromTable / table={caslib="Public" name="resampleActionSet.sashdat"} name="resample";
quit;

/* Bootstrap in CAS with Timing */
%let _time1=%sysfunc(time());

proc cas;
    resample.bootstrap / 
        intable='sample' 
        B=200 
        seed=12345 
        Bpct=1 
        case='ID' 
        strata='none' 
        strata_table='none';
    datastep.runcode result=t / code='data sample_bs; set sample_bs; host=_hostname_; threadid=_threadid_; run;';
    simple.crossTab / table={name="sample_bs" where="bag=1"} row="bsid" col="host" aggregator="N";
    simple.crossTab / table={name="sample_bs" where="bag=1"} row="bsid" col="threadid" aggregator="N";
run;

%let _time2=%sysfunc(time());
%let elapsed = %sysevalf(&_time2 - &_time1);
%let minutes = %sysfunc(floor(%sysevalf(&elapsed/60)));
%let seconds = %sysevalf(&elapsed - (&minutes*60));
%put CAS Bootstrap Time: &minutes minutes &seconds seconds;

/* Clear CAS Session */
*cas mysess clear;

Re: Bootstrapping SAS EG vs SAS Viya

sbxkoenk — Wed, 12 Jun 2024 11:12:45 GMT

Indeed.

Just a reminder ... so you (and others) are aware.

resample.bootstrap is not a SAS-supplied action. It's a custom action programmed by the owner of the GitHub - repository.

So, you also need the code to define the action in your SAS Viya environment.

BR, Koen

Re: Bootstrapping SAS EG vs SAS Viya

smackerz1988 — Sat, 15 Jun 2024 15:14:23 GMT

@sbxkoenk Just to clarify the resampleActionSet action was already available in the public cas library file without having to download or clone directly from the github repository?. Is that correct?. I guess what I need to understand how is sas viya able to use the .sashdat file to load from github repository? Hope I'm making sense. I know when I run this code

/* Load and Define Action Set */
proc cas;
    builtins.actionSetFromTable / table={caslib="Public" name="resampleActionSet.sashdat"} name="resample";
quit;

/* List Available Action Sets */
proc cas;
    action builtins.actionSetInfo;
run;

I am able to see resample and that it is user defined

Re: Bootstrapping SAS EG vs SAS Viya

sbxkoenk — Sat, 15 Jun 2024 21:56:04 GMT

@smackerz1988 wrote:

@sbxkoenk Just to clarify the resampleActionSet action was already available in the public cas library file without having to download or clone directly from the github repository?. Is that correct?.

Not correct.

You have to make the custom action set -- that you defined -- available in the PUBLIC CAS library (as a *.sashdat file) and then load from there.

(No need to clone repository ... just download the *.sas program and run or copy/paste and run)

Here's how to proceed (see also the README of the GitHub repository):

Setting up the actions in your environment

Run the code in resample - defineActionSet.sas. Some lines that may need changing:

line 1: connects to a CAS session
To Save the actions for future sessions and use by other users:
- line 174: create an in-memory table of the action set
- line 175: persist the in-memory table in .sashdat file. Here it is pointed as caslib="Public".
If you need to remove the action set then uncomment and use:
- line 177: removes the persisted in-memory table

Actions Instructions

To use the actions you will need to load the user defined actions with:

builtins.actionSetFromTable / table={caslib="Public" name="resampleActionSet.sashdat"} name="resample";

BR,
Koen

Re: Bootstrapping SAS EG vs SAS Viya

smackerz1988 — Sun, 16 Jun 2024 14:09:35 GMT

Ah I understand now! Thanks for clarifying that makes sense!

Re: Bootstrapping SAS EG vs SAS Viya

RyanCarr — Fri, 30 Aug 2024 15:39:55 GMT

Hi Koen,

I'm running on Viya V.04.00M0P051324 and the sampling action set does not seem to honor the replace=True option for sampling with replacement AND I can't seem to get the GitHub link for the resample action set to work.

Is there a bootstrap action I should be using for this version or do I need to figure out why I can't find the .sas program to create the resample action set ?

thanks,

Ryan

Re: Bootstrapping SAS EG vs SAS Viya

sbxkoenk — Fri, 30 Aug 2024 16:34:42 GMT

Hello,

I also get a 404-error ("This is not the web page you are looking for.") when following the GitHub link. Don't know why.
There's no bootstrap action in the SAS supplied actions!
The replace=TRUE options that you are seeing in the sampling.srs documentation ...(https://go.documentation.sas.com/doc/en/vdmmlcdc/8.1/casactstat/casactstat_srs_syntax.htm)
have nothing to do with "sampling with replacement". It's to replace output tables if they already exist.
I have no alternative for you. 😞 You can always use PROC SURVEYSELECT with the URS method (NOT SRS), but PROC SURVEYSELECT is not CAS-enabled and it will run -- single threaded -- on the compute server (NOT on the CAS server).
You can do the Unrestricted Random Sample (URS) with a datastep as well, but the datastep uses statements (like retain) that will not work on the CAS engine. See Implement five sampling methods in the SAS DATA step - The DO Loop.
There's probably a way to build reproducible bootstrap (re-)samples using the IML action set in PROC CAS, but I would have to write that code for you (I don't have it readily available).

Good luck,

Koen

Re: Bootstrapping SAS EG vs SAS Viya

FreelanceReinh — Sat, 31 Aug 2024 13:15:05 GMT

@sbxkoenk wrote:

You can do the Unrestricted Random Sample (URS) with a datastep as well, but the datastep uses statements (like retain) that will not work on the CAS engine. See Implement five sampling methods in the SAS DATA step - The DO Loop.

If the RETAIN statement was an obstacle (although it is used in CAS example code), it could be omitted, as it appears to be totally redundant in that URS sampling program.