Extend Visualization support for Lotka-Volterra and Ensembles

[JosephCottam]

New visualizations needed:

Dynamical/ODE system models (like Lotka-Voltera, SIR, etc.):

Example notebook.

Particularly interested in distribution visualizations so we can compare the different distributions, source data and the impact of interventions.

Desire is to compare:

• Data: Pre-intervention, post-intervention
• Predictions: pre-calibration, post-calibration, post-intervention
  • CALIBRATION can occur on pre- or post-intervention data (but usually pre-intervention)
• If working with intervention data, there is at least one intervention reference-point (possibly more).

Cell 48 has a good example of pre-intervention data, with pre- and post-intervention predictions. Cell 51 has predictions based on a model calibrated on pre-intervention data.

Ensembles (more than one model):

Example notebook

Comparison of various distributions, data and sample trajectories (cell 45 has a decently complete example).

• Reference data (training and testing subsets)
• Predictions of raw model distribution results over time (not ensembles)
• Prediction of ensemble distribution (before and after calibration)
• Small-number of reference trajectories
  • New sample from the posterior or prior distributions
  • "Representative" of samples used to build the prior or posterior
  • Closest match to the training or testing data

Visualization of the linear combination of models and how that combination varies over time (cell 51).

• Build out with new ideas and combining what is there in various ways
• Principle problem with current state is that the time-dependency between the rows is not clear).

[SamWitty]

@JosephCottam will resolve this issue by piping a few exemplar AMRs through the visualization pipeline. Will need @sabinala and/or @djinnome to help determine which AMRs.