Forecasting inverted time series using Prophet
I am working on a time series problem by making use of Facebook prophet. Prophet surpassed performance compared to other state of the art models as well some deep learning approaches.
I am stuck on a problem where I am trying to train a Prophet model on an inverted time series. Some will wonder why and what’s the use case, let’s say I am covered by NDA at the moment, so can’t speak. Building a Prophet model on forward time series is simple but my question is, can one train it on inverted series to forecast past? Has someone ever tried something like this.
Let’s say I have two series.
Series 1:
ds. y 2018-01-01. 100 2018-01-02. 200 2018-01-03. 80 ... ... 2018-01-31. 100
Series 2:
ds. y 2018-02-28. 150 2018-02-27. 125 2018-02-26. 180 ... ... 2018-02-01. 170
For series 1
, the time is moving in forward direction, hence, that’s good for traditional method. But for series 2
, we know that the time is moving in opposite direction. Is there a way where I could train a model to learn the opposite trend and seasonality using FB-Prophet?
I have been reading how can something like this be implemented using Bi-Directional RNNs? But that did not make much sense because in BRNNs both the direction is in the same series