Not able to delegate model training to SAP HANA through SAP PA desktop

[Former Member]

Hi All,

For last couple of days, I have been trying to delegate my model training to SAP HANA through SAP PA Desktop 2.5 version but it does not seem to take place. I have already done all the pre-requisite as mentioned below.

1. Have got APL 2.5 installed on HANA server.

2. Have assigned AFL_PM_CREATOR_ERASER_EXECUTE AND AFL__SYS_AFL_APL_AREA_EXECUTE rights to the right user.

3. Have executed required sql scripts as mentioned in APL 2.5 reference document under section 2.7

4. Have ensured that APL is installed correctly by using sql under section 2.6 as well us by using PING function.

5. have also ensured that script server instance is running in HANA.

6. Have created the required ODBC connection to HANA database through SAP PA desktop version.

7. Have ensured that required option to delegate the training is checked under preferences in SAP PA desktop version.

8. Have ensured that I am not use decision tree option while generating the model.

9. Have ensured that HANA database revision is correct.

10. Have checked that APL version matches with the SAP PA desktop version.

I am trying to do a classification/regression model. Can  anybody help me find out what else I might be missing or should do? My objective is to boost the performance because it takes lot of time to create a model on local machine as the size of the input data is very large.

Any help would be highly appreciated.

Thanks

Rajiv kumar

[former_member381272 · Accepted solution]

Hi Rajiv,

Would you mind to grant the 'sap.pa.apl.base.roles::APL_EXECUTE' to your user please ?

Regards,

[achab]

Hi Rajiv,

can you please share the file that you used for your tests with me?

Thanks in advance,

Antoine

[achab]

Hi Rajiv, can you please give a try to Jean-Raymond's suggestion and see if that solves your problem? Thanks & regards Antoine

[Former Member]

Hi Antoine,

I was able to delegate the training process after assigning APL_EXECUTE rights. However, I don't see expected results. Whole objective of this exercise was to gain performance as documented by SAP.

When I create a connection with HANA and delegate the training process to HANA it takes about 7 times more time than it would take through a CSV file. My Hana Box is 128 GB and my file has about 20K records with 754 columns. Do you know what might be causing this delay?

Thanks

Raijv kumar

[achab]

Hi Rajiv,

I would recommend to compare things that can be compared, for instance the learning time when delegating the creation of the predictive model to SAP HANA or not delegating it.

For accurate & "clean" comparison, system has to be warmed up, as the first call will generate specific artifacts in SAP HANA. The second learn should be faster.

Last point, your data volume is not that much (approx. 15 million data cells) so the difference between delegated and non-delegated mode might not be super significant.

Can you please kindly flag the post as "Answered" and credit Jean-Raymond with the proper answer?

Do not hesitate to create new posts if you have new questions.

I hope this helps,

Thanks & regards

Antoine

[achab]

To add to my previous answer, the main purposes of using the delegation mode to APL are to:

• avoid data transfer from SAP HANA to the Predictive Analytics server
• avoid network overloading due to data transfer
• limit security issues.

Regarding pure training performance both modes (delegated to HANA vs non-delegated to HANA) should be roughly on par.

As a side-note, if this is an option for you to share with me the file that you used for comparing HANA access and CSV access, our engineering department is keen on investigating your specific issue.

You can email the file to me directly if you want it to be kept confidential.

Thanks & regards

Antoine

[achab]

Hi Rajiv,

Just in case, are you using a custom cutting strategy?

See https://cp.hana.ondemand.com/dps/d/preview/13284cb50f744bf39a5f27bcd6c150da/2.5/en-US/index.html#0d9...

Thanks & regards

Antoine

[abdel_dadouche]

Hi Rjiv,

Would be able to share the HANA full version? As you may have noticed we have 2 version of the APL in 2.5, where one is for SPS 9 and the other for SPS 11.

Also one of the log file (%TEMP%\kxenwizard_SQL_log_%USERNAME%.txt) would definitely tell us what is happening behind the scene), so if you can share it?

Regards

@bdel

[achab]

Hi,

In fact with SAP 2.5 delegation is supported only with HANA SPS10, revisions 102.03 and above.

See slide 17 of https://support.sap.com/content/dam/library/ssp/infopages/pam-essentials/Pred_Ana_20.pdf

Thanks & regards

Antoine

[marc_daniau]

Assuming you have the right version and are not in a multi-tenant environment, one additional check you could do is: using HANA Studio logged as the same HANA user used in the ODBC connection, run the sample script claims-apl_Create_Train_ex.sql under:

...\apl-2.5.0.0-hanasp10-linux_x64\samples\sql\procedure\apl_insurance_fraud

[Former Member]

Hi,

I am not using a custom cutting strategy.

Thanks

[Former Member]

Hi antonie,

I am using SAP HANA 1.00.110.00.1447753075 version

Thanks

[Former Member]

Hi,

I am using SAP HANA 1.00.110.00.1447753075 version. Attached is the log file.

Thanks

kxenwizard_log_ab56122.txt.zip

[achab]

It means you are using HANA SPS11 and Automated Analytics 2.5 does not yet support it (although APL does). Stay tuned as our next release is planned to support HANA SPS11 for Automated Analytics.

[Former Member]

Do you have an anticipated date for the next release?

Thanks

[achab]

No we do not communicate on that date. Sooner than later 🙂

[Former Member]

Can we also expect support for SPS 9(revision 97) with new version?

Thanks

Rajiv kumar

[achab]

I would rather recommend to look after HANA SPS10 or HANA SPS11. As I said before it is too early to share more details. For SAP 2.5 you have to look after HANA SPS10, revisions 102.03 and above.

[abdel_dadouche]

Hi Rajiv,

The zip only contains the standard log (kxenwizard_log_ab56122.txt not the SQL one: kxenwizard_SQL_log_ab56122.txt)

Also it seems that you process generate plenty of logs therefore can you also send the kxenwizard_SQL_log_ab56122.txt.1, kxenwizard_SQL_log_ab56122.txt.2, kxenwizard_SQL_log_ab56122.txt.3 & kxenwizard_SQL_log_ab56122.txt.4 if any?

Best Regards

@bdel