Impact of string names

[sstroemer]

Preface: A lot of this could be irrelevant (or just plain wrong if I misunderstood something) if it changes in v0.7, please just point that out and ignore the respective points.

All questions talking about a "model" are related to the national_scale example model, with subset_time: ['2005-01-01', '2005-07-01'], using calliope_version: 0.6.10.

---

I've seen that the string names of variables/constraints are pretty verbose (and therefore long), which makes sense for readability / debugging. Is there a way to configure this? Is there a benchmark of how much that impacts build time as well as "pass to solver" time? One possibility (if not already present) would be to keep an internal compression dictionary that encodes frequently used string parts and translates between internal names - that can be read - and the ones written to LP files.

[brynpickering]

@sstroemer it would be good to split each of the top-level questions into their own discussion. Will make it easier to respond!

[sstroemer]

Edited this one, and split off each question into separate discussions 👍

[brynpickering]

If you use run.save_logs when running your model (and maybe any time you ask the calliope model to save to LP - I can't remember exactly), then we trigger a Pyomo option to pass the verbose names to the LP generation. By default, there are no strings passed on and everything is just a number corresponding to the position of the object in Pyomo's register of component objects.

In v0.7, I've made this more explicit by allowing you to give everything verbose string names after you have built the backend model, if you like. I found it takes ~10% of the model build time to stringify everything, although I don't know how that impacts LP build time / footprint on disk.