Timeout Errors

You are probably reading this page after clicking a link in the sidebar about a timeout error.

The Forecast Forge server has a time limit on how long it will allow a forecast to run for. This limit is in place to stop one broken forecast (e.g. if it is stuck in a neverending loop) blocking server resources for other users.

Approximately 90% of forecasts complete within 30 seconds and, at the moment the time limit is set to 10 minutes. So if you are running into the timeout error then your forecast is taking much longer to run than most people’s.

This can be caused by two things:

1. You are using a very long training period (e.g. 10 years of daily data) and a lot of regregressor columns
2. The data you are trying to forecast is not a good fit for the Forecast Forge model

Of these two, the latter is by far the most common but you should know if option 1 applies to you.

Things you can try

• Reduce the amount of training data you are using. Keep at least three years, but ask yourself “what does data from 10 years ago tell us about what will happen next month?”
• Plot your data to check if you can see weekly and annual patterns. Forecast Forge expects to find these and it will slow down the model fitting if it does not
• Add extra regressor columns to attempt to remove noise from the data. This will work particularly well if the regressors you add can help explain the missing seasonal patterns described above
• Send your data to Richard who will be able to advise on any potential issues