Pipeable Interface

[jaredlander]

Just briefly looked at the README and this looks really nice. One of the many things I like about {ompr} is that it was designed with a pipe in mind. This package uses the model$function() paradigm, and I get why, it's more in line with or-tools, gonum, etc. But that feels less natural in R, at least to me. Would love if it we could use pipes.

[dirkschumacher]

@jaredlander thanks for the feedback. I just haven't figured out a good way to support pipes while still having a mutable solver object. Maybe I provide two ways of working with a solver, once with model$xxx and one with pipes.

[jaredlander]

Admittedly, this is me asking for a whole new interface without offering any help, but the README said you wanted suggestions, so I threw it out there.

I do get the battle between mutable and pipeable. But mutable seems more pythonic and pipeable more Ronic (is that the term?).

I'll say I am excited for the possibility of a tool like {ompr} that can do quadratic programming or arbitrary convex optimization.

Have you ever thought about adding R Bindings to or-tools? It doesn't seem like the Google team will do it (google/or-tools#798), but it seems like the C++ functions could just be wrapped using {Rcpp}. Of course, I'm saying that without taking the burden of doing it, so I apologize for trivializing the effort.

[dirkschumacher]

Thanks a lot for the feedback. That gives one more data point in favor of a pipe interface:)

Re. or-tools: yeah I have thought about it and has been on my list. I wish I had more time for optimization work :) I honestly think that R is a great language for that.

[jaredlander]

R is an amazing language for optimization, I'll dare say it is the successor to MATLAB when it comes to handling matrices.

One thing that or-tools has that so far seems missing in R is interval objects.