Full life cycle PA Implementation

[naveen_kumar29091984]

Hi All,

Is there any sample to show complete work flow (full life cycle PA Implementation) which explains ,

1. Data  pre-processing (using correlation matrix, cleanup method)

2. Building the model (Decision Tree or any classification model)

3. Measure performance of model (PSME,  AUC-Area under curve and Confusion Matrix)

Regards,

Naveen

[achab · Accepted solution]

Hi Naveen, could you please let us know what is missing in our answers? In case nothing is missing, could you please set the thread as "Answered"? Thanks in advance, Antoine

[naveen_kumar29091984 · Accepted solution]

Hi ,

I am using sensor data in expert analytics,(Using HANA studio I pull this data to Expert Analytics)

I build the model using DT (Decision Tree) and NB(Naive Bayes) algorithm but when I try to compare the model,I am getting below error:

Regards,

NK

Message was edited by: Henry Banks

[Former Member · Accepted solution]

Hi,

The model compare and statistics node supports two class problems for classification.

If you target is multiclass then these components will not work.

Thanks

Paul

[Former Member · Accepted solution]

Hi Naveen,

Adding to Paul's comment... that error you see is on the partition node and is very likely due to unsupported types in your dataset. Please refer section "The following are the supported CSTYPE" in release restriciton note: https://launchpad.support.sap.com/#/notes/0002295127 for details.

Also, I am not sure if I understand what system date change are you referring to there and its objective.

Regards,

Jayant

[achab · Accepted solution]

Hi Naveen,

Please refer to Jayant's latest answer - I think it is the correct root cause.

In general please do not use one post for many questions as this is not the recommended SCN practice.

Thanks for your understanding,

Best regards,

Antoine

[naveen_kumar29091984 · Accepted solution]

Many Thanks to All.

Regards,

NK

[achab]

Hi Naveen,

To date the closest we have to this is the magnificent set of tutorials built by Andreas here: .

To further clarify, does your question apply to the Automated mode, to the Expert mode or to both?

Thanks & regards

Antoine

[naveen_kumar29091984]

Thanks Antoine

Both.

Mainly into Expert Analytics Mode.

Is there any Custom component availabe(.spar) for AUC(area under Curve) and PSME

Regards,

Naveen

[achab]

Adding to this.

Few additional things:

- Jayant covered an end to end scenario in his blog here:

-  This will also be presented in an ASUG webinar planned for October 12: ASUG.com - Events

- it's a webinar series where we cover a scenario end to end.

- in Jayant's blog you can notice we display AUC indicators, alongside with the Prediction Confidence (KR) and the Predictive Power (KI)

Thanks & regards,

Antoine

[Former Member]

Hi Naveen,

You can use the Model Statistics node in Expert mode to evaluate any two-class classification model (AUC, S(KS), Gain%, Lift%, etc.) or regression model (R2, L1, L2, Linf, Error Mean, Error Std Dev, etc.). The component computes different KPIs (as listed in brackerts) from the results of training that can be fed into Model Compare node to automatically pick the best model in a parallel chain.

Blog post referred by Antoine has more details.

Regards,

Jayant

PS: You can have a look at 's blog for extensions in R for use in Expert mode.