SelectionCache#
- class braincell.filter.SelectionCache(_masks=dict(), _morpho=None, _revision=-1)[source]#
Scratch space threaded through a whole selection expression tree.
Memoization keyed by expression value is the whole of it:
evaluated()lets a composite reuse an operand’s mask instead of walking the morphology again, so(A | B) & (A | C)evaluatesAonce rather than twice. Consumers that derive something from a mask can memoize that too, by giving it its own cache instance rather than sharing the keyspace holding the masks – see_RegionCacheinbraincell._discretization.mechanism.Notes
Memoized entries are valid only for the morphology and structural revision they were produced from. Handing the same cache to a second morphology, or reusing it after
Morphology.attach(), drops the stored masks rather than returning a stale answer.This class previously also carried
tree_distance_to_root,euclidean_distance_to_root, andbranch_radius_summary, reserved for the region types that still raiseNotImplementedError. Nothing ever wrote them. Whoever implements one of those types should add the field it actually needs rather than inherit three guesses – note that per-node tree distance is already computed bybraincell.morph._spatial.MorphologySpatialGeometry.- evaluated(expr, morpho, evaluate)[source]#
Return
evaluate(), reusing an earlier result forexpr.- Parameters:
expr (
object) – The expression being evaluated. Used as the memo key, so the result is shared between operands that compare equal – and, just as importantly, not shared between two that do not.morpho (
object) – Morphology the result belongs to.evaluate (
Callable[[],object]) – Zero-argument thunk producing the value when there is no hit.
- Returns:
The freshly computed or previously memoized value – a mask for the sub-expression cache, whatever the thunk returns otherwise.
- Return type:
Notes
Expressions carrying an unhashable payload (a list of branch indices, an array of bounds) simply do not memoize; they fall through to
evaluate()unchanged.