Business Context
A marketing analyst needs to run a complex customer segmentation analysis (using the `fastknn` action) which takes a considerable amount of time. They want to start the job, disconnect their client session to work on other tasks, and retrieve the results later once the analysis is complete. This ensures that their local resources are not tied up and improves productivity.
Data Preparation
Create a sample customer table with demographic and behavioral data for segmentation. The table contains 1 million records to simulate a moderately long-running analysis.
Copied!
| 1 | DATA casuser.customer_data; |
| 2 | call streaminit(123); |
| 3 | DO i = 1 to 1000000; |
| 4 | customer_id = i; |
| 5 | age = 20 + floor(rand('UNIFORM') * 50); |
| 6 | income = 30000 + floor(rand('UNIFORM') * 100000); |
| 7 | recency = floor(rand('UNIFORM')*365); |
| 8 | frequency = 1 + floor(rand('UNIFORM')*50); |
| 9 | monetary = 10 + rand('UNIFORM')*500; |
| 10 | OUTPUT; |
| 11 | END; |
| 12 | RUN; |
Étapes de réalisation
1
Start a named CAS session 'analyst_session', load the required action set, and run the data preparation step.
Copied!
| 1 | cas analyst_session name='analyst_session'; |
| 2 | PROC CAS; |
| 3 | SESSION analyst_session; |
| 4 | ACTION BUILTINS.LOADACTIONSET / actionSet='kClus'; |
| 5 | /* Assume data_prep code has been run here */ |
| 6 | ACTION TABLE.tableInfo / caslib='casuser' TABLE='customer_data'; |
| 7 | RUN; |
2
In 'analyst_session', start the 'fastknn' analysis asynchronously and print the session's UUID.
Copied!
| 1 | PROC CAS; |
| 2 | SESSION analyst_session; |
| 3 | ACTION kClus.fastknn RESULT=knn_job / |
| 4 | TABLE={name='customer_data'}, |
| 5 | inputs={{name='age'}, {name='income'}, {name='recency'}}, |
| 6 | k=5, |
| 7 | OUTPUT={casout={name='customer_segments', replace=true}, copyvars='ALL'}, |
| 8 | async='knn_analysis_job'; |
| 9 | PRINT 'Analyst Session UUID: ' || SESSION.sessionId(); |
| 10 | RUN; |
3
In a separate session ('admin_session'), use the UUID from the previous step to switch the running job to batch mode.
Copied!
| 1 | cas admin_session name='admin_session'; |
| 2 | PROC CAS; |
| 3 | SESSION admin_session; |
| 4 | /* Replace 'uuid-from-analyst-session' with the actual UUID */ |
| 5 | ACTION SESSION.batchresults / uuid='uuid-from-analyst-session'; |
| 6 | RUN; |
4
Later, from any session, check the status of the job until it is completed.
Copied!
| 1 | PROC CAS; |
| 2 | /* This step might be run multiple times */ |
| 3 | ACTION SESSION.actionstatus / name='knn_analysis_job'; |
| 4 | RUN; |
5
Once the job is complete, fetch the results and view the output table.
Copied!
| 1 | PROC CAS; |
| 2 | ACTION SESSION.fetchresult / name='knn_analysis_job'; |
| 3 | ACTION TABLE.fetch / TABLE={name='customer_segments', caslib='casuser'}, to=10; |
| 4 | RUN; |
Expected Result
The `batchresults` action should successfully detach the 'knn_analysis_job'. The 'analyst_session' client is immediately freed. The `actionstatus` call will initially show the job as 'running' and later as 'completed'. The `fetchresult` call will retrieve the full results dictionary of the `fastknn` action, and the 'customer_segments' output table will be created and populated in the 'casuser' caslib.