linear regression normalization

[Farzaneh_T_]

Hi there,

I am trying to implement Intel DAAL for linear regression. I have noticed that by default it doesn't scale data before doing regression. I can include z-score calculation in the code, but is there any option to be activated in the linear regression algorithm to do normalization before doing the regression?

Many thanks in advance,

Farzaneh

[VictoriyaS_F_Intel · Accepted solution]

Hello Farzaneh,

As you correctly noticed Intel DAAL does not scale the data before doing the regression.

To normalize the data please use one of the normalization algorithms available in the library: https://software.intel.com/sites/products/documentation/doclib/daal/daal-user-and-reference-guides/index.htm#daal_prog_guide/GUID-219BA9AA-5380-429A-8DE1-8194760F1A3E.htm

Best regards,

Victoriya

[Andrey_N_Intel]

Hi Farzaneh,

To double-check that our understanding of your question is correct. Do you suggest to activate z-score in the linear regression algorithm for usability reasons, that is to avoid explicit call to z-score technique in Intel DAAL based application, before training the linear regression model? Or you keep in mind some other reasons/use cases?

Andrey

[Farzaneh_T_]

Hi Andrey,

I need  to get reasonable coefficients from the linear regression model to decide to reduce the number of features that I have. Therefore I need to scale the data before doing linear regression. I wanted to know whether there is any straight-forward way to do that, rather than first calling the z-score, and then start the training the linear regression.

Best,

Farzaneh

[Andrey_N_Intel]

Hi Farzaneh,

As Vika mentioned earlier in this thread, with the present version of Intel DAAL API you need to call z-score first, and then start training the liner regression model. On our side, we will run additional internal analysis of your request.

Thanks, Andrey