New Scaling Rollout - Running To Do

[adam-a-a]

New Scaler class approach should be rolled out across all IDAES models. However, we should also be careful that there is little to no loss in useful information; i.e., we should not recklessly replace calculate_scaling_factors routines for models with half-baked Scaler classes. Thus, we should have a way to benchmark current state of scaling vs. any revamped form of scaling through the new Scaler classes. We can take a shot at this by possibly making use of ScalingProfiler, and more likely, making use of IPOPTConvergenceAnalysis. The bare minimum so far would be to check condition numbers between existing scaling and any new scaling introduced.

• Decide on initial, minimal set of scaling/robustness tests to apply to models
• Select an initial model to apply new Scaler class to, along with tests that help benchmark "new" scaling against "old" scaling (calculate_scaling_factors); @dallan-keylogic suggests focusing on something that uses saponification since the property package scaler class already exists.

Coming "soon":
Running checklist of all models in IDAES that do not have Scaler classes yet. We'd then (predominantly) assign codeowners to select models. Right now, there will be too many models to list, and we should probably discuss further before we continue.

[adam-a-a]

One more note: @MarcusHolly has already started rolling this out for some WaterTAP models (currently PRs), and we are still settling on robustness tests that we'd want to consider. Feel free to chime in @MarcusHolly.