Cannot select 'Save as Model'

[Former Member]

Hi all,

I created a very simple model in PA on top of a table in HANA SP5 (AWS box).

The model consists only of a simple HANA R-KMeans algorithm.

While following the user guide, I come to the point of saving the model.  However, this option is greyed out and I do not find a way to enable it.

Is this because I'm using the 30 day trial version?

[Former Member · Accepted solution]

Hello,

There are two types of predictive algorithms - unsupervised and supervised. In case of supervised, it is possible to train the model and create a model and score it using a different set of data. Example of this type of algorithms is Classification, Regression etc. However, in the unsupervised case, it does not make sense to save the model since the algorithm generate different set of results every time. Segmenation (K-Means) falls into this category and hence it is not possible to save the model in case of K-Means. If you want to use "Save the Model" option, you can use one of the supervised type of algorithms. Once you run the analysis, if you right click on the Algorithm component, then you will see the option to save the model.

Hope this addresses your query.

Thanks,

Vishwa

[Former Member · Accepted solution]

Hi Vishwa,

"However, in the unsupervised case, it does not make sense to save the model since the algorithm generate different set of results every time"

That's.... an interesting perspective.

What if somebody wanted to use the output of the clustering model as an input for a supervised problem?

Regards,

Peter

[Former Member · Accepted solution]

Hi Peter,

In case of SAP Predictive Analysis, it is possible to feed the output of clustering analysis to another supervised algorithm like C4.5 in which case you can save the model.

Hope this addresses your question.

Thanks,

Vishwa

[Former Member · Accepted solution]

C4.5 is a totally different algorithm that has completely different fitting method than Kmeans clustering, so simply feeding the data into a different algorithm that you can save really doesn't solve the problem.

It is also not accurate to say that unsupervised algorithms categorically do not need be scorable; there are many business scenarios where you want to score new results against a static model and really no model should really be used unless it is validated against a set of independent data. 

I also noticed the save model button disappeared from the clustering algorithm and it is frustrating, however I expect it is because there is no fitting/predicted function in R for the clustering model. 

When I have needed to score against an existing/saved clustering model, I have done it by exporting the cluster centers table (available either from the printed R output in PA or as an output of the R algorithm) and created a stored proc to run against that table and score records by manually calculating the cluster distances.  I can share my code as an example if you are interested.

It would be nice if this were built into PA.

Hillary Bliss

[Former Member · Accepted solution]

Hi Hillary,

I did not mean to say that C4.5 would be an alternative to K-Means! Just wanted to convey that you can use any other algorithm in case you need to save the model.

Thanks for the feedack on saving the models with K-Means.We will add the capability that you requested to our backlog.

Thanks,

Vishwa

[clarissa_dold]

Hi,

there are no limitations when using the free trial. Can you please provide a screenshot of what's greyed out for you? Is it the checkbox?

There's a tutorial available about saving the model that can be found on the following website:

http://scn.sap.com/docs/DOC-32651

called Save, Export and Import Data Models within the Getting Started section.

Regards,

Clarissa