How/why Factor labels are generated

[dehann]

Point of this Discussion is to make sure we have the right strategy in place to name factors. When we were doing this in Julia only, it worked well, was sensible, and easy. There is some outstanding upgrades on SDKs, but potential need for doing better in factor label generation.

• When adding two separate factors between the same variables :x1, :x2, :c0, what should the name of that factor be:
  • at time of writing, the two factor names are automatically created as :x1x2c0f1 and :x1x2c0f2
• Variable naming has near universal freedom. Variables can be called anything that can be a Julia Symbol (alphanumeric).
• Variables use label as unique identifier (inside a User/Robot/Session)
• Factors must somehow be uniquely identified.
  • Changing how Variables and Factors are uniquely identified will have many implications in JL and elsewhere.

---

See discussion prompting a deeper dive into the Factor naming convention and prior assumptions:

• https://github.com/NavAbility/NavAbilitySDK.jl/issues/77#issuecomment-1079833249

I know that appending f# is standard practice but I have a really hard time with the idea of being forced to do naming the way the framework wants me to. I am a verbose namer and not having choice to call my factor StarttoAprilLandmark1 or something meaningful drives me a bit bonkers. It's a long term concern and has other implications. I also think requiring labels to be unique is a bit silly. Just me. Most databases will store an ID and a display name. We are using label for both.

[dehann]

My feeling is we should let only DFG/IIF decide what the factor names are going to be, since there is no need for the user to have to dream up and keep track of what names do and don't exist. If a users starts naming factors :odometry1, :odometry2, that would create a lot of burden on the user to manage all that. My strong sense is the library should do it and the user should have easy access to what the generated factor names were:

ls(fg, :x1)
[x0x1f1; _x0x1f2;:x1l4f1]

[dehann]

I think what we have is a good solution, barring the challenges with Cloud requiring the full data at this time. I'd like to suggest we get rid of cloud payloads requiring factor labels by some mechanism. I don't think users should have to define and manage factor labels. Either block that field in certain operations, or something else...