0

Objective : forecast product movement of individual product category for each store. customer has around 1000 product categories and over 2000 store.

Approach :
- Use ARIMA with regressors (with lead and lag effect) , TBATS, NNET , other models on training dataset and test the accuracy against test dataset - select best model based on their MAPE and boxtest results. - use best model identified above to generate forecast.

Challenge : The code is taking around 3-5 minutes to run for one product one store. I need to run this code for around 1000 products across 2000.

Question : How do we achieve scalability when we are running forecasting for each store each category ?

Note : All four cores of machine is utilised while running the code. CPU utilisation 80%.

New2Analytics
  • 1
  • what packages are you using? what functions? etc.. – Cyrus Mohammadian Aug 16 '16 at 08:04
  • Solution: 1. buy another machine or cloud computing time; 2. parallelize the calculation -- a reproducible example never hurted anyone ;) http://stackoverflow.com/questions/5963269/how-to-make-a-great-r-reproducible-example – renato vitolo Aug 16 '16 at 08:12

0 Answers0