Big Data and Business Analytics

206  •  Big Data and Business Analytics

package to regenerate hundreds of thousands of historical forecasts in a

matter of hours: 100,000 forecasts would take less than two hours on 1,000

computers. Also, the output of forecast mapper contains millions of data.

frame rows, which makes the aggregation step very slow using standard R

data manipulation. For 1,000 time series in the datastore and 100 dierent

training end dates, the output of forecast mapper would have more than

20 million rows. Using the R Dremel package, we can perform basic aggre-

gations over this 20-million row data set in seconds. For example, we can

easily compute the MAPE fo ...