[DEV] Overview of issues related to parametric and consolation

[dehann]

State of the nation:

Framework

Working on the parametric solution #468, I found it difficult in a few areas and tried to come up with a framework that the current implementation can be consolidated into. The framework implemented in PR #468 should be seen as my proposal on:

1. Belief messages Consolidate belief messages and message channels  #459
2. Message channels Consolidate belief messages and message channels  #459
3. The BayesTree structure - Could only be partially implemented
   • Towards breaking tree dependence on Graphs.jl Towards breaking tree dependence on Graphs.jl #525
   • WIP Linear parametric solution WIP Linear parametric solution #468
   • TreeClique index field and underlining graphs structure indexes TreeClique index field and underlining graphs structure indexes #540
   • Also Clique label field and label attribute is a bit confusing with getter/setter.  #551

Related: Potential improvements to manifolds and storage, but nothing done there #465

Parametric Solution

WIP Linear parametric solution #468 merged without:

1. Calculate covariances for parametric tree solve Calculate covariances for parametric tree solve #539
2. Parametric tree initialization [Feature] Parametric tree initialization #541
3. Manifold and non-linear
   • Wrapper for parametric factor evaluation Consolidate / Simplify factor evaluation (CalcFactor) #467

Related: build linear Gaussian chain rule utility function #466

Dehann's Ticklist

The quick answer is I'm working on none of them.

• ❌ not going to do
• ✔️ done
• Currently working on

Hi @Affie, I need some help understanding what you are doing at the moment please (a cheeky request) -- could you tick all that apply... We need to stay a little synchronized here, you are moving quick with so many things! :-)

• Writing issues (DF nice!)
• ✔️ Paramatric proof of concept 2D linear?
  • POC done with outstanding functions as listed above
• ✔️ Parametric PoC manifold (non-linear) improvement?
  • Used optim project/retract methods
• Parametric integration with existing factor definitions -- (Consolidate / Simplify factor evaluation (CalcFactor) #467)
• Parametric integration or improvements to current CSMC data containers?
• ✔️ Parametric PoC using existing BayesTree?
  • POC done with outstanding functions as listed above
• ✔️ Preparation on current Bayes tree so that we can slowly move away from Graphs.jl for the BayesTree specifically?
  • See above
• Building a new MetaBayesTree (BayesTree alternative to Graphs.jl #609)? Blocking on TreeClique index field and underlining graphs structure indexes #540
• Standardizing message channels on the existing Bayes tree? Blocking on Consolidate belief messages and message channels  #459
• Building a new message channels on the existing Bayes tree? Blocking on Consolidate belief messages and message channels  #459
• ❌ Building separate message channels on a new Bayes tree that is not at all integrated with current Bayes tree?
  • I hope that is not what I did
• ✔️ Standardizing or incorporating new fields in existing Belief message types? Blocking on Consolidate belief messages and message channels  #459
• ✔️ Building new message types that work on existing Bayes tree? Blocking on Consolidate belief messages and message channels  #459
• ❌ Building new messages that only work on a new MetaBayesTree and not at all consolidated or integrated with existing code?
  • Again, I hope not

Please list any other things below?

• [BUGFIX] Sub-graph regression #556

PS, if any of these items above are not consolidated, how and where are we going to consolidate them?

PSS, Please edit with sub-bullet if any of these items are already finished?

PSSS, we may need to sync this a group a little and rather finish some items before starting new stuff.

[dehann]

As I understand the two most importing things you are blocking on me on are:

• How to do covariance calculation for Gaussian belief prop,
• Consolidation of the tree message Channels in Bayes tree.

[dehann]

Haha, thanks enjoyed the "state of the nation" theme! Thanks this actually really helps a lot with scoping work on my side. Lets work to clear this list of stuff before we start anything new?

[dehann]

CSM and ParametricCSM should be interchangeable, so that users can float between various Chapman-Kolmogorov solution strategies in development and use.

[Affie]

I'm closing this as the only remaining issue is captured in #467

[dehann]

Thanks @Affie, that was an major pile of work. Thanks for all the effort!