This came up with HODMD: I would like to have the eigenvalues of the model to be an output, to be able to analyse them retrospectively. However, you can't make them a float-valued model measure, because then each of them would need to be associated with a timestamp, and that doesn't make sense. Another instance of this would be GES, which might have training and inference as separate models, and the outputs of the former would be model parameters, not time series.
There are many ways one could solve this: Relax the assumption that all values must come with a timestamp, allow multivalued outputs (a list rather than a single float), etc. Also, #131 would sort this.
This came up with HODMD: I would like to have the eigenvalues of the model to be an output, to be able to analyse them retrospectively. However, you can't make them a float-valued model measure, because then each of them would need to be associated with a timestamp, and that doesn't make sense. Another instance of this would be GES, which might have training and inference as separate models, and the outputs of the former would be model parameters, not time series.
There are many ways one could solve this: Relax the assumption that all values must come with a timestamp, allow multivalued outputs (a list rather than a single float), etc. Also, #131 would sort this.