amachine.am_cohesion

 1from abc import ABC, abstractmethod
 2import warnings
 3from dataclasses import dataclass
 4
 5class CohesionKernel(ABC):
 6	"""
 7	Defines the substitution weight distributions over target symbols
 8	given a source symbol.
 9	"""
10
11	def _normalize(self, source : str, dist: dict[str, float]) -> dict[str, float]:
12		wsum = sum(w for w in dist.values())
13		if wsum <= 1e-15:
14			warnings.warn( "Weight function sums to 0, implicitly mapping source with probability 1" )
15			res = dist.copy()
16			res[ source ] = 1.0
17			return res
18		return {s: w / wsum for s, w in dist.items()}
19
20	@abstractmethod
21	def cohesion_scores( self, source: str, symbols: set[str] ) -> dict[str, float]:
22		"""
23		Maps a source symbol to a distribution of cohesion scores over other symbols.
24		"""
25
26@dataclass
27class Uniform(CohesionKernel):
28	def cohesion_scores(self, source: str, symbols: set[str]) -> dict[str, float]:
29		return { s: 1.0 for s in symbols }
30
31@dataclass
32class Marginal(CohesionKernel):
33	"""Per-symbol weight distribution."""
34	scores : dict[str, float]
35
36	def cohesion_scores(self, source: str, symbols: set[str]) -> dict[str, float]:
37		return {s: self.scores[s] for s in symbols }
38
39@dataclass
40class Symmetric(CohesionKernel):
41	"""w(a->b) == w(b->a)"""
42	scores : dict[frozenset, float]
43	def cohesion_scores(self, source: str, symbols: set[str]) -> dict[str, float]:
44		return {s: self.scores[frozenset({source, s})] for s in symbols }
45
46@dataclass
47class Asymmetric(CohesionKernel):
48	"""Directed pairwise scores."""
49	scores : dict[tuple[str,str], float]
50
51	def cohesion_scores(self, source: str, symbols: set[str]) -> dict[str, float]:
52		return {s: self.scores[(source, s)] for s in symbols }