explicit outer product over common parameters with different parameter values

[kostrzewa]

Imagine you have values A_i parametrised by a set of parameters p^{A_i}_j and another set of value B_k parametrised by a set of parameters p^{B_k}_l.

The two parameter sets share some column names (and in general also some or all parameter values).

The functionality of inner_outer_join is great for producing some set of values C_m which is a combination of A and B, with the inner and outer combinations of the parameter sets taken into account automatically.

What if, however, I would like to combine A and B in a non-diagonal way in the parameters that they share (an outer product, possibly partial, across an inner margin, in some sense).

As an example, let's say A and B each have six parameters src_px src_py src_pz snk_px snk_py snk_pz, but the required combination of these parameters is one where none or just some of the values of src_px src_py src_pz snk_px snk_py snk_pz of A and B are the same. Formally, the new paramval C would thus have 12 parameters A.src_px A.src_py ... A.snk_pz B.src_px B.src_py ... B.snk_pz.

Of course, I can filter A and B "by hand" and create a new paramval with the combination that I would like, inventing the new parameters as required. But had you thought about a better way to do this perhaps?

[kostrzewa]

This is further complicated by the fact that for some of the parameters that A and B carry and have in common, I would in fact like to perform the inner product (stuff like ensemble names, for example).

[kostrzewa]

I guess this can be achieved via convert in the pv_call, but I'm not sure whether that actually can take care of exactly what I need to do. I'll give it a try.

[kostrzewa]

Hmm, going via parameter_to_data is also quite inconvenient, although it at least makes the PVs conformable. I need to think about this...

[martin-ueding]

One way that would not require me to do anything would be just parameter_to_data on each part separately, internally rename the values and then do a combined pv_call. Then you can do the dplyr SQL-like commands on the values.

One could also rename the parameters and then get an outer join with pv_call. In the function one then just returns NA when there is an invalid combination of the parameters. This effectively filters the container. Or one uses inner_outer_join manually and then pv_filter.

Perhaps we figure out a hackish but somewhat concise way to do this and then see whether it can be generalized?

[kostrzewa]

I think at this point I've basically reached the conclusion that it's too much trouble to be worth doing. I'm taking a different approach by using paramval's to represent only certain (very broad) parameter-value dependencies while I use an internal representation with another lookup-table-like construction to resolve fine-grained paramter-value relationships to form the required (partial) outer products.