Read the adaptive predictors from the BLOB
Hi Team,
We have a requirement to get the data from five tables linked to a single customer class. Once we get the records, we need to process the adaptive model on the customer data.
1. We cannot create one dedicated table with all the predictor columns(coming from 5 tables). As we have around 500+ predictors.
2. We cannot create a data base view to create the dynamic snap shot of the customer data.
Approach(Want to know the design impact): Get all the predictors into one BLOB column. That means I have a table with two columns one with CustomerID*, BLOB. Can we get the predictors from the BLOB column in order to run the adaptive models? If yes, what will the performance impact of having 500+ predictors reading from the BLOB.
Regards,
Nizam
PEG
GB
As far as I'm aware, custom blob are not supported:
https://community.pega.com/support/support-articles/unable-save-attachments-pega-different-table
So for this approach, you would need to leverage the Pega blob. For example, you create a new class with CustomerID (exposed) and a value list or value group property (storing all these predictors, not exposed).
This way these predictor ends up in the Pega blob.
If you read only 1 records, then I expect performance impact won't be huge. After all, a simple case opened by Pega using the Pega blob contains often hundred of data element if we count all the date in pages, page list, page grounp embedded in the case.
If you read many records at the same time, then it will be felt. This is why reports needs to run against exposed properly (and not the blob).
Axiom Technology Group
US
Thanks for responding back to my question. We want to go with pega internal BLOB only but can we retrieve 500+ predictors for adaptive learning? Will it create any new issues.
Regards,
Nizam
PEG
GB
In terms of performance, for simple SQL, queries reading from the blob is 200 times slower than reading from an exposed column.
I wouldn't be able to tell what happens with 500 items in the blob
Pegasystems Inc.
NL
I'm not familiar with the details of your use case, but if you have a couple of tables that are related by a key, why don't you leverage data flow functionality to "left-join" them together (using a Compose shape)? This would result in a Customer class with nested elements for e.g. products, usage, etc. The OOTB DMSample application provides an excellent example of this
In the configuration of your Adaptive model rule, you can use predictors from the embedded pages just like you use predictors from the top level page.
500 predictors is a lot, but ADM should be able to handle this. The law of diminishing returns applies here, and with > ca 700 predictors you can even hit a hard limit (https://community.pega.com/support/support-articles/memory-limit-exceeded-high-number-predictors). It is advised to monitor the models through the various reports and being aware of general guidelines for model predictors (if you have Mesh access, see https://mesh.pega.com/docs/DOC-127801).
Otto
I'm not familiar with the details of your use case, but if you have a couple of tables that are related by a key, why don't you leverage data flow functionality to "left-join" them together (using a Compose shape)? This would result in a Customer class with nested elements for e.g. products, usage, etc. The OOTB DMSample application provides an excellent example of this
In the configuration of your Adaptive model rule, you can use predictors from the embedded pages just like you use predictors from the top level page.
500 predictors is a lot, but ADM should be able to handle this. The law of diminishing returns applies here, and with > ca 700 predictors you can even hit a hard limit (https://community.pega.com/support/support-articles/memory-limit-exceeded-high-number-predictors). It is advised to monitor the models through the various reports and being aware of general guidelines for model predictors (if you have Mesh access, see https://mesh.pega.com/docs/DOC-127801).
Otto
Axiom Technology Group
US
Hi @Otto_Perdeck,
Thanks for sharing your response. We did the same to implement the multiple predictors to the customer table. Its a great idea but what about the association rules? If I create association between two classes, does it save the system performance? Association is a way which pega has suggested to use in one of the documentations.
Regards,
Nizam
Pegasystems Inc.
NL
Hi, I am not sure about the performance of one vs the other, but if you need to combine data from different sources (eg relational DB and DDS), you will need to use a dataflow Compose anyway.