Datasets:
File size: 7,176 Bytes
e1ced61 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 | """``U_MISSING`` completion-candidate enumeration (``docs/02`` §11.2).
For a hidden node, the state-proof builder (P7) needs a set of *completion*
worlds — the same world with the hidden node set to each of several valid
alternative values — so it can search for a pair that renders identically (same
observed hash) yet yields different unique answers (the ``U_MISSING`` criterion).
This module provides the deterministic enumeration and the
:func:`~explicit_learning.interventions.base.with_value` re-builder; P7 owns the
pair search itself.
Enumeration is seed-determined and yields ``≥8`` distinct admissible values
(mirroring the §7.3 mutation suite's per-leaf count), each producing a fresh
world via :func:`with_value` so the source world stays immutable.
"""
from __future__ import annotations
from collections.abc import Iterator
from dataclasses import dataclass
from fractions import Fraction
from typing import Any
from ..dsl.ast import canonical_number_str, parse_number
from ..sources.base import World
from ..sources.world_ops import (
find_point,
find_row,
find_stating_constraint,
looks_numeric,
node_kind,
)
from .base import InterventionError, with_value
from .substitute import _seeded_offsets
_DEFAULT_COUNT = 8
@dataclass(frozen=True)
class Completion:
"""One alternative completion of a hidden node."""
world: World
node_id: str
value: str # canonical string form of the alternative value
index: int
def valid_values(
world: World,
hidden_node: str,
*,
seed: int,
count: int = _DEFAULT_COUNT,
) -> list[str]:
"""Enumerate ``count`` distinct admissible values for ``hidden_node``.
Numeric sites (a point's ``y``, a row cell, a geometry measure) yield seeded
distinct rationals near the source range; category sites (a label) yield
non-colliding labels. The values are deterministic in ``(seed, node, count)``.
"""
kind = node_kind(world, hidden_node)
if kind is None:
raise InterventionError(f"completion: {hidden_node!r} is not a world node")
if kind == "point":
return _point_values(world, hidden_node, seed, count)
if kind == "row":
return _row_values(world, hidden_node, seed, count)
if kind == "constraint":
return _measure_values(world, hidden_node, seed, count)
if kind in ("series", "entity"):
return _label_values(world, hidden_node, kind, seed, count)
raise InterventionError(f"completion: no sampler for kind {kind!r}")
def completion_candidates(
world: World,
hidden_node: str,
*,
seed: int,
count: int = _DEFAULT_COUNT,
) -> list[Completion]:
"""``count`` completion worlds for ``hidden_node`` (source world immutable).
Each completion is :func:`with_value` applied to a distinct enumerated value;
the source world is never mutated. Raises :class:`InterventionError` if the
node has no mutable value site.
"""
out: list[Completion] = []
for i, value in enumerate(valid_values(world, hidden_node, seed=seed, count=count)):
out.append(
Completion(
world=with_value(world, hidden_node, value),
node_id=hidden_node,
value=value,
index=i,
)
)
return out
def iter_completions(
world: World, hidden_node: str, *, seed: int, count: int = _DEFAULT_COUNT
) -> Iterator[Completion]:
"""Lazy stream of completions (memory-bounded for large enumerations)."""
for i, value in enumerate(valid_values(world, hidden_node, seed=seed, count=count)):
yield Completion(
world=with_value(world, hidden_node, value),
node_id=hidden_node,
value=value,
index=i,
)
# --- value samplers (distinct, seeded) ------------------------------------
def _point_values(world: World, pid: str, seed: int, count: int) -> list[str]:
point = find_point(world, pid)
if point is None:
raise InterventionError(f"point {pid!r} not found")
sidx, _pidx, p = point
siblings = world["series"][sidx].get("points", []) or []
used = {_frac(q.get("y", "0")) for q in siblings}
cur = _frac(p.get("y", "0"))
return _distinct_near(seed, pid, "comp_point", cur, used, count)
def _row_values(world: World, rid: str, seed: int, count: int) -> list[str]:
row = find_row(world, rid)
if row is None:
raise InterventionError(f"row {rid!r} not found")
_ridx, r = row
cells = r.get("cells", {}) or {}
if not cells:
raise InterventionError(f"row {rid!r} has no cells")
key = next(iter(cells))
used = {
_frac(row_.get("cells", {}).get(key))
for row_ in (world.get("rows", []) or [])
if key in (row_.get("cells", {}) or {}) and looks_numeric(row_.get("cells", {}).get(key))
}
cur = _frac(cells[key])
return _distinct_near(seed, f"{rid}:{key}", "comp_cell", cur, used, count)
def _measure_values(world: World, entity: str, seed: int, count: int) -> list[str]:
found = find_stating_constraint(world, entity)
if found is None:
raise InterventionError(f"no stating constraint for {entity!r}")
_cidx, c = found
args = list(c.get("args") or [])
if c.get("predicate") == "MeasureOf" and len(args) >= 2 or looks_numeric(args[1]):
cur = _frac(args[1])
elif looks_numeric(args[0]):
cur = _frac(args[0])
else:
raise InterventionError(f"stating constraint for {entity!r} is not numeric")
return _distinct_near(seed, entity, "comp_measure", cur, set(), count)
def _label_values(world: World, nid: str, kind: str, seed: int, count: int) -> list[str]:
if kind == "series":
existing = {str(t.get("label", "")) for t in world.get("series", []) or []}
else:
existing = {str(e.get("label", e.get("id", ""))) for e in world.get("entities", []) or []}
out: list[str] = []
for n in _seeded_offsets(seed, nid, "comp_label", span=10_000):
cand = f"node_{n}"
if cand not in existing and cand not in out:
out.append(cand)
if len(out) >= count:
break
if len(out) < count:
raise InterventionError(f"could not enumerate {count} labels for {nid!r}")
return out
def _distinct_near(
seed: int, tag: str, kind: str, cur: Fraction, used: set[Fraction], count: int
) -> list[str]:
"""``count`` distinct rationals near ``cur`` not in ``used`` (seeded)."""
out: list[str] = []
seen: set[Fraction] = {cur, *used}
base = int(cur)
for off in _seeded_offsets(seed, tag, kind, span=2 * count * 20 + 10):
cand = Fraction(base + (off % (4 * count + 20)) - (2 * count + 10))
if cand in seen:
continue
seen.add(cand)
out.append(canonical_number_str(cand))
if len(out) >= count:
break
if len(out) < count:
raise InterventionError(f"could not enumerate {count} distinct values for {tag!r}")
return out
def _frac(value: Any) -> Fraction:
return parse_number(value)
__all__ = ["Completion", "completion_candidates", "iter_completions", "valid_values"]
|