Business Context
An e-commerce platform needs to refresh product recommendations for its entire user base overnight. This involves scoring a very large table of user-product pairs against a recommendation model. The primary success criterion is the ability to process millions of records reliably and efficiently within the batch window.
About the Set : neuralNet
Training of classical artificial neural networks.
Discover all actions of neuralNet Data Preparation
Generates a large table `user_product_pairs` with 1 million rows to simulate a realistic batch scoring workload. A pre-trained `recommendation_model` is assumed to exist.
Copied!
| 1 | DATA mycas.user_product_pairs(promote=yes); |
| 2 | DO UserID = 1 to 10000; |
| 3 | DO ProductID = 1 to 100; |
| 4 | UserAge = 18 + floor(rand('uniform')*50); |
| 5 | ProductCategory = ceil(rand('uniform')*10); |
| 6 | OUTPUT; |
| 7 | END; |
| 8 | END; |
| 9 | RUN; |
Étapes de réalisation
1
Execute the `annScore` action on the large-scale dataset. No complex parameters are used; the focus is on raw throughput and stability.
Copied!
| 1 | PROC CAS; |
| 2 | neuralNet.annScore / |
| 3 | TABLE={name='user_product_pairs'}, |
| 4 | modelTable={name='recommendation_model'}, |
| 5 | casOut={name='recommendation_scores', replace=true}; |
| 6 | RUN; |
| 7 | QUIT; |
2
Perform a quick verification by checking the row count of the output table to ensure it matches the input table's row count.
Copied!
| 1 | PROC CAS; |
| 2 | TABLE.tableInfo / TABLE='recommendation_scores'; |
| 3 | RUN; |
| 4 | QUIT; |
Expected Result
The action should complete successfully without timeout or memory errors. The resulting table `mycas.recommendation_scores` must contain exactly 1,000,000 rows, corresponding to the input data. The log should indicate a successful run, and the output table will contain the original data plus a new column (e.g., `_NN_Pred_`) with the recommendation score for each user-product pair.