Skip to Content
0

Input table for TIme Series Forecast

Mar 06, 2017 at 07:58 AM

139

avatar image

Hi Experts,

I am very new to HANA PAL.

I want to forecast sales of each store for next 12 months

The source table has 9 fields. The structure of my input table is mentioned below

"StoreId" INTEGER

"DayOfWeek" NVARCHAR(3),

"Date" DATE ,

"Sales" DOUBLE ,

"Customers" DOUBLE ,

"Open" INTEGER ,

"Promo" INTEGER ,

"StateHoliday" NVARCHAR(3),

"SchoolHoliday" INTEGER .

I have one year data month wise sales and I want to forecast the next 12 months data.

My horizon is 12.

I am using AUTOARIMA to predict my model.

but when I passing the my above table as input to the procedure generator wrapper function I am getting signature error.

"SYS"."AFLLANG_WRAPPER_PROCEDURE_CREATE": line 156 col 5 (at pos 5073): [423] (range 3) AFL error exception: AFL error: Registration of AFLLANG wrapper procedure "schemaname"."<procedurename>" failed with error 'inconsistent parameter description

My question is

1) Do I need to create another table with Store ID, Date and Sales values and pass this table as source or Do I need to change the data type of my table so that it will match the signature

2) I want to know how the influence of Open,Promo,Sateholiday and Schoolday values on sales and forecast my sales. For this do I need to use multiple linear regression model

Thanks

Pawan Akella

10 |10000 characters needed characters left characters exceeded
* Please Login or Register to Answer, Follow or Comment.

3 Answers

Best Answer
Hasan Rafiq Mar 13, 2017 at 10:32 AM
0

Hi Pawan,

As per your data it seems that you would actually be using Auto ARIMA to come out with an ARIMAX model. Also as per the documentation you can see that none of the ARIMA models can handle only INT , DOUBLE till date.

You might consider to remove them, for example: day of week can be split to 7 columns like MON, TUE... SUN each with a value 0 or 1( called as hot encoding technique ).

After this your wrapper code will be created without any issues. The results that you get from AUTO Arima will have to be further passed to ARIMAXFORECAST in your case to forecast newer values.

However, also consider using SAP BO PA as suggested by Antoine.

Thanks,

Hasan

Share
10 |10000 characters needed characters left characters exceeded
Antoine CHABERT
Mar 06, 2017 at 08:06 AM
1

Hi Pawan, I will try to get someone to answer your PAL-related questions. In the meantime, what don't you use our SAP BusinessObjects Predictive Analytics trial version and give a try to our Automated Time Series data mining function? It might be helpful. KInd regards Antoine

Show 1 Share
10 |10000 characters needed characters left characters exceeded

Hi Antonie, Waiting for the help. Regards Pawan Akella

0
Meilin Bai
Mar 10, 2017 at 03:39 AM
0

Hi Pawan,

1. Currently, AUTOARIMA does not support categorical variables. To handle categorical regressors, therefore, one should manually transform them to ordered ones before carrying out AUTOARIMA. We were discussing whether to support categorical variables in the future version.

2. Multiple linear regression, of course, accepts categorical variables. However, it neglects the historical series as well as historical noises. So, I cannot assure that if it can yield a reasonable evaluation of the variable importances.

Hope it helps.

Best regards,

Meilin

Share
10 |10000 characters needed characters left characters exceeded