mixle.reason.design module

Experimental design for cross-modal reasoning – which evidence to acquire, in what batch, under a budget.

next_evidence() picks the single most informative item (expected information gain). But real evidence gathering is a budgeted batch over fidelities: acquire the set of (item, fidelity) observations that most sharpens the answer per unit cost, without wasting budget on evidence that is redundant given what you have already chosen. This is cost-aware multi-fidelity experimental design – the discrete-corpus analogue of mixle.doe.multi_fidelity_minimize – and it is adaptive: after each pick it re-scores every candidate against the updated belief, so overlapping evidence is naturally avoided (the near-optimal greedy for submodular information gain).

select_evidence_batch() plans the acquisition; the returned AcquisitionPlan carries the chosen (index, fidelity, gain, cost) trail, the total nats gained, and the assimilated belief.

class AcquisitionPlan(items=<factory>, total_cost=0.0, total_gain=0.0, belief=None)[source]

Bases: object

A budgeted evidence-acquisition plan: the chosen items, the nats they bought, and the final belief.

Parameters:
items: list[tuple[int, str, float, float]]
total_cost: float = 0.0
total_gain: float = 0.0
belief: Any = None
property indices: list[int]
select_evidence_batch(store, belief, *, budget, query=None, fine_cost=1.0, coarse_cost=0.2, fidelities=('coarse', 'fine'), candidates=None, max_items=None, min_gain=1e-9)[source]

Greedily acquire the most-informative-per-cost (item, fidelity) evidence under a total budget.

At each step every remaining candidate is scored – at each allowed fidelity – by the entropy it would remove from the query given the belief so far, divided by its cost; the best affordable one is folded in. Adaptive re-scoring means a batch never double-counts overlapping evidence. Stops when nothing affordable helps.

Parameters:
Return type:

AcquisitionPlan