I have N time series of shape (30*36) (time step, feature). For each time series 500 parameters that can be seen as the history of the time series have to be added. This parameters are different for each time series.
For now, the way I found to feed a ML algorithm with the time series plus the parameters is to create 500 new constants features per time serie correponding to this parameter.
The final N time series would have the shape (30*536), where the 500 last features are constant in time.
However this approach have to main issues :
- Important features (the parameters) are constants
- I am increasing by 15 the size of my data (dimension wise)
Any ideas of how I could deal with this parameters ?
