187 add equation of state workgraph

[MM0hsin]

Add EOS workgraph

Get Scales > QE and split into test, train and valid file.

[ElliottKasoar]

Looks like a good start!

I think we may want to work with the StructureData a little more directly in general?

Currently, I don't think the input structure isn't actually saved, only the name of the path is, which isn't the most helpful for reproducibility.

I think we want an input to either be SinglefileData, or probably actually the StructureData, similar to our CalcJobs, and the QE CalcJob.

Then, could we not pass the StructureData from the scales_task to the qe_task (it doesn't hurt to have keep an auxiliary intermediate output file, but it's probably cleaner to work with the data more directly where possible)?

[MM0hsin]

Then, could we not pass the StructureData from the scales_task to the qe_task (it doesn't hurt to have keep an auxiliary intermediate output file, but it's probably cleaner to work with the data more directly where possible)?

So you mean storing the outputs from scales_task as variables in a dict, then passing the dict into qe_task?

[ElliottKasoar]

Then, could we not pass the StructureData from the scales_task to the qe_task (it doesn't hurt to have keep an auxiliary intermediate output file, but it's probably cleaner to work with the data more directly where possible)?

So you mean storing the outputs from scales_task as variables in a dict, then passing the dict into qe_task?

Exactly how you store the StructureData probably doesn't matter, as long as the real data is in the graph, but yeah I'd imagine something like that. The main point was just to be able to keep using the structure data directly, rather than writing out the data and then just reading it back in immediately.

[ElliottKasoar]

How does this compare in terms of implementation to similar AiiDA example workflows, e.g. https://aiida-tutorials.readthedocs.io/en/latest/sections/writing_workflows/realworld.html and https://workgraph-collections.readthedocs.io/en/stable/cp2k/eos.html?

[ElliottKasoar]

It would also be nice to actually have the EoS plot here (see examples linked elsewhere). All you need is the energies (from the DFT) and the volumes (which you can get from the structures if this is not saved).

There's probably no time right now, but it would also be cool to then do an eos via janus with the fine-tuned model, and see how that compares.

[MM0hsin]

Couldnt do EOS.plot() as there was no minimum:

but i did change from bar chart to a scatter plot.

[MM0hsin]

@ElliottKasoar

How does this compare in terms of implementation to similar AiiDA example workflows

I was not aware workgraph-collections existed before this. But main difference is we dont use ase EquationOfState directly to calculate the energies, instead we generate the scaled structures and run Quantum Espresso

[MM0hsin]

Added suggested changes