search

yahoo / egads

A Java package to automatically detect anomalies in large scale time-series data
Other
1.17k stars 331 forks source link

Auto forecast: resistant to model errors #21

Open dareid opened 9 years ago

dareid commented 9 years ago

Using the automatic forecasting option for a small dataset always results in an error from the TripleExponentialSmoothing model which states it needs 2 years of data, which is fair enough.

However, I don't think this should kill the automatic forecasting. Maybe errors from models could be caught, and only the models which successfully built could be assessed?

nlaptev commented 9 years ago

Once again dareid, nice catch. Please submit a pull request for this.