Easy option for a Time-series forecast

[Sivaji]

I understand in the current version we need to create a time signal using timesince() function, and use the predict function to train the signal that we need over the time signal created. However I was wondering if we can time-series forecast function which internally creates the time signal based on the start datetime of the training window and trains the model.

[Teddy]

Sivaji,

I am not 100% sure if this is what you are looking for, but you can create a time signal from a condition.  This allows you to build a condition with the manual condition tool and apply the following formula to create the time counter.

timesince($condition,1h)

You can then just limit the training window to that condition.  If you need to extend the time counter into the future you may want to apply growend() to the condition in the time since formula above.

Teddy

[Joe Reckamp]

I'll also add that as of R22.0.49, there's new forecast functions like forecastLinear that allows you to create a linear regression out into the future based on the last X number of minutes. You can see a little bit about that at the What's New Knowledge Base here if you are not on that version yet or can check out the formula documentation for more information on those functions.

[Sivaji]

3 hours ago, Joe Reckamp said:

I'll also add that as of R22.0.49, there's new forecast functions like forecastLinear that allows you to create a linear regression out into the future based on the last X number of minutes. You can see a little bit about that at the What's New Knowledge Base here if you are not on that version yet or can check out the formula documentation for more information on those functions.

Thanks Joe, yeah forecastLinear works for linear forecasting. I'm looking for an option in "Model & Predict", for a "Time Series Forecast" which can be similar to the "Prediction" feature, however the only change I'm looking for is there shouldn't any option for "Input Signals", this feature should take the time series as the input signal by default (considering the training window). I am aware that we can create a signal using timesince() function and use this as my "Input Signals" and use the traditional "Prediction" feature. However SEEQ mostly deals with the time-series data, I think having this feature could give benefit to the customers who wants to do some kind of Time Series Forecasting of their signals. Customers might be interested in predicting sinusoidal/seasonal trends also along with linear trends. Again this is a suggestion.