File size: 13,933 Bytes
767a051 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 | """Typed, content-addressed records for TOM Learner 0.1.
The authority boundary is deliberately explicit:
observations -> deterministic split -> candidate hypotheses -> evidence
-> acceptance decision -> optional learned definition -> promotion
The learner never mutates the base world directly. A candidate can become
visible in the authoritative learner overlay only through ``LearnerStore`` and
a parent-bound promotion transaction.
"""
from __future__ import annotations
from dataclasses import dataclass
from typing import Any, Mapping, Sequence
from tom_world03.canonical import require_hash
from tom_world03.rational import Q
CANONICAL_SEED_SHA256 = "d1417a3136772c0cf3eddcd4962ce07d42cbf87616f7b5bae09fc652d9b807b5"
BASE_WORLD_HASH = "sha256:c25c99eeeb728f50b52a06e89b9f669470ccb294fd8f0cc6e4ccb87bde9ff2d9"
BASE_HANDOFF_HASH = "sha256:3d2b46cfd33ba6e5cf0a13697fb59e374a64ad30450fdd3c256c98a04ebc474b"
PROFILE = "TOM-LEARNER-0.1"
OBSERVATION_SCHEMA = "TOM-EXACT-OBSERVATION-0.1"
OBSERVATION_SET_SCHEMA = "TOM-AFFINE-OBSERVATION-SET-0.1"
SPLIT_POLICY_SCHEMA = "TOM-LEARNER-SPLIT-POLICY-0.1"
HYPOTHESIS_FAMILY_SCHEMA = "TOM-AFFINE-HYPOTHESIS-FAMILY-0.1"
ACCEPTANCE_POLICY_SCHEMA = "TOM-LEARNER-ACCEPTANCE-POLICY-0.1"
SPLITS = ("train", "validation", "holdout")
def _require_nonempty(value: Any, label: str) -> str:
text = str(value)
if not text:
raise ValueError(f"{label} must be nonempty")
return text
def _require_hash_text(value: Any, label: str) -> str:
text = str(value)
if not text.startswith("sha256:") or len(text) != 71:
raise ValueError(f"{label} must be a sha256: content hash")
try:
int(text[7:], 16)
except ValueError as exc:
raise ValueError(f"{label} must contain hexadecimal SHA-256") from exc
if text[7:] != text[7:].lower():
raise ValueError(f"{label} must use lowercase hexadecimal")
return text
def _positive_integer(value: Any, label: str, *, minimum: int = 1) -> int:
if isinstance(value, bool) or not isinstance(value, int) or value < minimum:
raise ValueError(f"{label} must be an integer >= {minimum}")
return value
@dataclass(frozen=True, slots=True)
class Observation:
id: str
t: Q
y: Q
provenance: Mapping[str, Any]
content_hash: str
@classmethod
def from_record(cls, record: Mapping[str, Any]) -> "Observation":
require_hash(record, label="observation")
if record.get("schema") != OBSERVATION_SCHEMA:
raise ValueError(f"observation schema must be {OBSERVATION_SCHEMA}")
provenance = record.get("provenance")
if not isinstance(provenance, Mapping):
raise ValueError("observation provenance must be an object")
return cls(
id=_require_nonempty(record.get("id", ""), "observation id"),
t=Q.from_value(record.get("t")),
y=Q.from_value(record.get("y")),
provenance=dict(provenance),
content_hash=_require_hash_text(record.get("content_hash"), "observation content_hash"),
)
def to_record(self) -> dict[str, Any]:
return {
"schema": OBSERVATION_SCHEMA,
"id": self.id,
"t": self.t.to_record(),
"y": self.y.to_record(),
"provenance": dict(self.provenance),
"content_hash": self.content_hash,
}
@dataclass(frozen=True, slots=True)
class SplitPolicy:
strategy: str
ratios: Mapping[str, int]
minimum_counts: Mapping[str, int]
salt: str
content_hash: str
@classmethod
def from_record(cls, record: Mapping[str, Any]) -> "SplitPolicy":
require_hash(record, label="split policy")
if record.get("schema") != SPLIT_POLICY_SCHEMA:
raise ValueError(f"split policy schema must be {SPLIT_POLICY_SCHEMA}")
strategy = str(record.get("strategy", ""))
if strategy != "sha256-id-order-largest-remainder":
raise ValueError("unsupported split strategy")
ratios_raw = record.get("ratios")
minima_raw = record.get("minimum_counts")
if not isinstance(ratios_raw, Mapping) or set(ratios_raw) != set(SPLITS):
raise ValueError("split ratios must name train, validation, and holdout exactly")
if not isinstance(minima_raw, Mapping) or set(minima_raw) != set(SPLITS):
raise ValueError("minimum_counts must name train, validation, and holdout exactly")
ratios = {name: _positive_integer(ratios_raw[name], f"ratio {name}") for name in SPLITS}
minima = {name: _positive_integer(minima_raw[name], f"minimum {name}", minimum=0) for name in SPLITS}
if minima["train"] < 2:
raise ValueError("training minimum must be at least two")
return cls(
strategy=strategy,
ratios=ratios,
minimum_counts=minima,
salt=_require_nonempty(record.get("salt", ""), "split salt"),
content_hash=_require_hash_text(record.get("content_hash"), "split policy hash"),
)
def to_record(self) -> dict[str, Any]:
return {
"schema": SPLIT_POLICY_SCHEMA,
"strategy": self.strategy,
"ratios": {name: int(self.ratios[name]) for name in SPLITS},
"minimum_counts": {name: int(self.minimum_counts[name]) for name in SPLITS},
"salt": self.salt,
"content_hash": self.content_hash,
}
@dataclass(frozen=True, slots=True)
class HypothesisFamily:
model: str
candidate_source: str
max_candidates: int
content_hash: str
@classmethod
def from_record(cls, record: Mapping[str, Any]) -> "HypothesisFamily":
require_hash(record, label="hypothesis family")
if record.get("schema") != HYPOTHESIS_FAMILY_SCHEMA:
raise ValueError(f"hypothesis family schema must be {HYPOTHESIS_FAMILY_SCHEMA}")
if record.get("model") != "y=a*t+b":
raise ValueError("TOM Learner 0.1 supports only y=a*t+b")
source = str(record.get("candidate_source", ""))
if source != "all-distinct-t-unordered-training-pairs":
raise ValueError("unsupported affine candidate source")
return cls(
model="y=a*t+b",
candidate_source=source,
max_candidates=_positive_integer(record.get("max_candidates"), "max_candidates"),
content_hash=_require_hash_text(record.get("content_hash"), "hypothesis family hash"),
)
def to_record(self) -> dict[str, Any]:
return {
"schema": HYPOTHESIS_FAMILY_SCHEMA,
"model": self.model,
"candidate_source": self.candidate_source,
"max_candidates": self.max_candidates,
"content_hash": self.content_hash,
}
@dataclass(frozen=True, slots=True)
class AcceptancePolicy:
require_unique_exact_train: bool
require_exact_validation: bool
require_exact_holdout: bool
reject_contradictions: bool
max_model_complexity: int
content_hash: str
@classmethod
def from_record(cls, record: Mapping[str, Any]) -> "AcceptancePolicy":
require_hash(record, label="acceptance policy")
if record.get("schema") != ACCEPTANCE_POLICY_SCHEMA:
raise ValueError(f"acceptance policy schema must be {ACCEPTANCE_POLICY_SCHEMA}")
bool_fields = (
"require_unique_exact_train",
"require_exact_validation",
"require_exact_holdout",
"reject_contradictions",
)
for field in bool_fields:
if not isinstance(record.get(field), bool):
raise ValueError(f"acceptance policy {field} must be boolean")
return cls(
require_unique_exact_train=bool(record["require_unique_exact_train"]),
require_exact_validation=bool(record["require_exact_validation"]),
require_exact_holdout=bool(record["require_exact_holdout"]),
reject_contradictions=bool(record["reject_contradictions"]),
max_model_complexity=_positive_integer(record.get("max_model_complexity"), "max_model_complexity"),
content_hash=_require_hash_text(record.get("content_hash"), "acceptance policy hash"),
)
def to_record(self) -> dict[str, Any]:
return {
"schema": ACCEPTANCE_POLICY_SCHEMA,
"require_unique_exact_train": self.require_unique_exact_train,
"require_exact_validation": self.require_exact_validation,
"require_exact_holdout": self.require_exact_holdout,
"reject_contradictions": self.reject_contradictions,
"max_model_complexity": self.max_model_complexity,
"content_hash": self.content_hash,
}
@dataclass(frozen=True, slots=True)
class ObservationSet:
id: str
seed_sha256: str
base_world_hash: str
base_handoff_hash: str
input_name: str
output_name: str
observations: tuple[Observation, ...]
split_policy: SplitPolicy
hypothesis_family: HypothesisFamily
acceptance_policy: AcceptancePolicy
provenance: Mapping[str, Any]
content_hash: str
@classmethod
def from_record(cls, record: Mapping[str, Any]) -> "ObservationSet":
require_hash(record, label="observation set")
if record.get("schema") != OBSERVATION_SET_SCHEMA:
raise ValueError(f"observation-set schema must be {OBSERVATION_SET_SCHEMA}")
if record.get("profile") != PROFILE:
raise ValueError(f"observation-set profile must be {PROFILE}")
seed = str(record.get("seed_sha256", ""))
if seed != CANONICAL_SEED_SHA256:
raise ValueError("observation set is not bound to the canonical TOM seed")
base = _require_hash_text(record.get("base_world_hash"), "base_world_hash")
if base != BASE_WORLD_HASH:
raise ValueError("observation set is bound to an unsupported base world")
handoff = _require_hash_text(record.get("base_handoff_hash"), "base_handoff_hash")
if handoff != BASE_HANDOFF_HASH:
raise ValueError("observation set is bound to an unsupported literal handoff")
domain = record.get("domain")
if not isinstance(domain, Mapping) or domain.get("numeric") != "exact-rational":
raise ValueError("observation-set domain must declare exact-rational numeric semantics")
values = record.get("observations")
if not isinstance(values, list) or not values:
raise ValueError("observation set requires a nonempty observations array")
observations = tuple(Observation.from_record(value) for value in values)
ids = [item.id for item in observations]
if len(ids) != len(set(ids)):
raise ValueError("observation IDs must be unique")
if ids != sorted(ids):
raise ValueError("observations must be sorted by ID")
provenance = record.get("provenance")
if not isinstance(provenance, Mapping):
raise ValueError("observation-set provenance must be an object")
split = record.get("split_policy")
family = record.get("hypothesis_family")
policy = record.get("acceptance_policy")
if not all(isinstance(value, Mapping) for value in (split, family, policy)):
raise ValueError("observation-set policies must be objects")
return cls(
id=_require_nonempty(record.get("id", ""), "observation-set id"),
seed_sha256=seed,
base_world_hash=base,
base_handoff_hash=handoff,
input_name=_require_nonempty(domain.get("input", ""), "input field name"),
output_name=_require_nonempty(domain.get("output", ""), "output field name"),
observations=observations,
split_policy=SplitPolicy.from_record(split),
hypothesis_family=HypothesisFamily.from_record(family),
acceptance_policy=AcceptancePolicy.from_record(policy),
provenance=dict(provenance),
content_hash=_require_hash_text(record.get("content_hash"), "observation-set hash"),
)
def observation_map(self) -> dict[str, Observation]:
return {item.id: item for item in self.observations}
def to_record(self) -> dict[str, Any]:
return {
"schema": OBSERVATION_SET_SCHEMA,
"profile": PROFILE,
"id": self.id,
"seed_sha256": self.seed_sha256,
"base_world_hash": self.base_world_hash,
"base_handoff_hash": self.base_handoff_hash,
"domain": {
"input": self.input_name,
"output": self.output_name,
"numeric": "exact-rational",
},
"observations": [item.to_record() for item in self.observations],
"split_policy": self.split_policy.to_record(),
"hypothesis_family": self.hypothesis_family.to_record(),
"acceptance_policy": self.acceptance_policy.to_record(),
"provenance": dict(self.provenance),
"content_hash": self.content_hash,
}
def source_records(self) -> list[Mapping[str, Any]]:
"""Return every literal content-addressed input record in causal order."""
return [
self.split_policy.to_record(),
self.hypothesis_family.to_record(),
self.acceptance_policy.to_record(),
*(item.to_record() for item in self.observations),
self.to_record(),
]
def require_dataset_sequence(values: Any) -> Sequence[Mapping[str, Any]]:
if not isinstance(values, list) or not values:
raise ValueError("benchmark datasets must be a nonempty array")
if not all(isinstance(value, Mapping) for value in values):
raise ValueError("benchmark dataset entries must be objects")
return values # type: ignore[return-value]
|