Dehann Fourie

when `ccw.params` gets created on an `approxConv`, and if is `.partial` factor, then only the partial dimensions of interest should be used on the construction of ccw before numerical calculations....

would it be sensible to have a function that just returns the partial dimensions instead? E.g. ```julia plr = PartialLinearRelative(..., (p)->p[2]) ``` and let the factor residual functions handle the...

> Your two examples - horizontal range and heading-only constraints - would normally not be implemented as generic partials, right? Correct, the way I have it at this time is...

Yes, let's convert one or two factors to Manopt.jl and work it incrementally from there. Shall we transition the newest abstract to Manopt away from Optim to do this? https://github.com/JuliaRobotics/DistributedFactorGraphs.jl/blob/bfa8a2ae72cdabb352d193deef86bf7ab2877ba9/src/entities/DFGFactor.jl#L13...

Note that when using `getSolverParams(fg).dbg=true` each of the major stages of the CSM clique inference subgraphs are stored to the `getLogPath(fg)` destination -- including: - `fg_build`, as copied from main...

Hi @Affie , yeah so the best I have been able to do so far is as follows: First thing is to find where the logs are going to land:...

Actually there is one more you can do with `hist::Dict` (automatically fetched for all CSM if `async=false`), use `printCliqHistorySequential` to see output like this: ```julia using RoME fg = generateCanonicalFG_Hexagonal()...

The most pressing developer tool here is to better manage how CSM terminates on error. The two options I have come up with are either (sure there are more options...

reminder that the revamped `getSolverParams(fg).useMsgLikelihoods = true` is default false at this time and will be made default true in future.

Another trick that doesn't come up too often but none the less: `getSolverParams(fg).multiproc = false` -- is default true. Similarly, you can skip up or down stages by `getSolverParams(fg).dodownsolve =...