Home
Tilaye Yismaw Alemu edited this page Apr 12, 2018
·
2 revisions
Here are the algorithms used in Forecast-API,
- Random Walk
- Moving Average
- Exponential smoothing
- ARIMA
- Regression
Each algorithm comes with its own set of parameters. Some parameters are common to algorithms, some are unique. Here are the main parameters,
Seasonality
- pre-set seasonality (eg. 24 hours, weekly, yearly)
- automatic seasonality detection
- additive seasonality
- multiplicative seasonality
Trend
- linear
- exponential
Outlier removal
- eg. sudden surge of video views that lasts just for one day and dooesn't occur again
Out of the box, Forecast-API has a set of such pre-configured algorithms. There are currently around 140 variations. A full list of them can be found in canned_set_definitions.txt. On every forecast all of these are competed against each other to pick the one with the least RMSE (Root Mean Squared Error).