The check for vanishing dimensions in the construction of GMWOPResult is imprecise because it uses the dimensions from all outputs and we might have related outputs which are producing multiple different independent dimensions.
See JuliaDataCubes/YAXArrays.jl#608 for an example from YAXArrays.
It is unclear whether we could remove the singleton dimension here directly or whether they would be needed to clarify the compute graph down the line if the results of such an operation is used in another computation and combinging into the added dimension.
We might want to think about replacing the singleton dimensions with metadata information to keep track of the dimension graph without having to show this to regular users.
The check for vanishing dimensions in the construction of GMWOPResult is imprecise because it uses the dimensions from all outputs and we might have related outputs which are producing multiple different independent dimensions.
See JuliaDataCubes/YAXArrays.jl#608 for an example from YAXArrays.
It is unclear whether we could remove the singleton dimension here directly or whether they would be needed to clarify the compute graph down the line if the results of such an operation is used in another computation and combinging into the added dimension.
We might want to think about replacing the singleton dimensions with metadata information to keep track of the dimension graph without having to show this to regular users.