Using Sim selections about mdsynthesis HOT 3 CLOSED

You have to give selections a name in order to generate them later; the usage should look like:

ag = 'bynum 0:10'

S.selections.add('my favorite atoms', ag)

Selections are stored as (lists of) selection strings, and they are applied on call to give back an AtomGroup from the active universe. I think it should be possible to feed AtomGroups in directly instead of selection strings, storing a list of atom indices instead of strings for the selection definition.

BTW, you can avoid the (admittedly awkward) add syntax by using setitem syntax:

S.selections['my favorite atoms'] = 'bynum 1:10'

and get the corresponding AtomGroup from the selection definition with:

ag = S.selections['my favorite atoms']

I don't know if there's a clean way to handle += as an operation, especially if some selections are defined as lists of strings while others are defined by lists of atoms. I think maintaining a predictable atom order would get very tricky real fast.

from mdsynthesis.

Yeah I don't mind hacking in adding AtomGroups directly myself, I hate working in terms of selection strings myself.

For the second point, there's the python method __iadd__ (and friends), which lets you handle augmented assignment in special ways.

https://docs.python.org/2/library/operator.html#operator.iadd

I forgot they're all key: value pairs, so possible things you could do

S.selections += {'something': 'sel 1', 'somethign else': 'sel 2'}

S.selections += 'key', 'value'

Food for thought at least

from mdsynthesis.

I don't mind adding in these kind of enhancements so long as they can be unambiguously implemented at both the interface and persistence levels. I would suggest the following:

1. add a new table definition to persistence.SimFile for _SelectionAtoms, which stores integers in a single column representing the atom indices of the selection
2. add logic to persistence.SimFile.add_selection to create a _SelectionAtoms table for integers given as selection, as well as the usual _Selection table for selection strings given as selection (current behavior)
3. add logic to aggregators.Selections add and __setitem__ (or better, change __setitem__ to call add, and change only add) to handle AtomGroups. That is, to pull out their atom indexes and feed them to self._backend.add_selection

I prefer to keep MDAnalysis calls out of persistence as much as possible to limit the complexity of the File classes; this separation preserves that.

from mdsynthesis.