File size: 14,423 Bytes
f9bc0c1 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 | """Ablation-plan audits for research-grade OCR training branches."""
from __future__ import annotations
import json
from dataclasses import dataclass, field
from pathlib import Path
from typing import Any
from .errors import DataValidationError
from .manifest import ManifestEntry, load_manifest
@dataclass(slots=True)
class AugmentationAblationArmAudit:
name: str
profiles: list[str]
train_manifest: str
eval_manifest: str
audit_path: str
train_count: int | None
eval_count: int | None
train_manifest_count: int | None = None
eval_manifest_count: int | None = None
train_claim_eligible_count: int = 0
eval_claim_eligible_count: int = 0
passed: bool = False
issues: list[str] = field(default_factory=list)
warnings: list[str] = field(default_factory=list)
def to_dict(self) -> dict[str, Any]:
return {
"name": self.name,
"profiles": self.profiles,
"train_manifest": self.train_manifest,
"eval_manifest": self.eval_manifest,
"audit_path": self.audit_path,
"train_count": self.train_count,
"eval_count": self.eval_count,
"train_manifest_count": self.train_manifest_count,
"eval_manifest_count": self.eval_manifest_count,
"train_claim_eligible_count": self.train_claim_eligible_count,
"eval_claim_eligible_count": self.eval_claim_eligible_count,
"passed": self.passed,
"issues": self.issues,
"warnings": self.warnings,
}
@dataclass(slots=True)
class AugmentationAblationAudit:
config_path: str
output_json_path: str
output_md_path: str
ablation_id: str
passed: bool
promotion_ready: bool
arm_count: int
real_review_audits: list[str]
real_evidence_count: int
verified_reference_count: int
arms: list[AugmentationAblationArmAudit]
issues: list[str] = field(default_factory=list)
warnings: list[str] = field(default_factory=list)
def to_dict(self) -> dict[str, Any]:
return {
"config_path": self.config_path,
"output_json_path": self.output_json_path,
"output_md_path": self.output_md_path,
"ablation_id": self.ablation_id,
"passed": self.passed,
"promotion_ready": self.promotion_ready,
"arm_count": self.arm_count,
"real_review_audits": self.real_review_audits,
"real_evidence_count": self.real_evidence_count,
"verified_reference_count": self.verified_reference_count,
"arms": [arm.to_dict() for arm in self.arms],
"issues": self.issues,
"warnings": self.warnings,
}
def audit_augmentation_ablation(
config_path: str | Path,
output_dir: str | Path,
*,
real_review_audits: list[str | Path] | None = None,
) -> AugmentationAblationAudit:
config_file = Path(config_path)
config = _read_json_object(config_file, "augmentation ablation config")
base_dir = config_file.parent
issues: list[str] = []
warnings: list[str] = []
ablation_id = _clean_str(config.get("ablation_id")) or config_file.stem
claim_policy = config.get("claim_policy")
if not isinstance(claim_policy, dict):
issues.append("claim_policy must be an object")
elif claim_policy.get("claim_evidence_eligible") is not False:
issues.append("claim_policy.claim_evidence_eligible must be false for synthetic ablation configs")
source_bundle = config.get("source_bundle")
if not isinstance(source_bundle, dict):
issues.append("source_bundle must be an object")
else:
_audit_optional_passed_json(
source_bundle.get("full_split_audit"),
base_dir,
"source_bundle.full_split_audit",
issues,
warnings,
require_zero_issues=True,
)
font_readiness = source_bundle.get("font_readiness_report")
if font_readiness:
_audit_optional_passed_json(font_readiness, base_dir, "source_bundle.font_readiness_report", issues, warnings)
arms_payload = config.get("arms")
arms: list[AugmentationAblationArmAudit] = []
if not isinstance(arms_payload, list) or not arms_payload:
issues.append("arms must be a non-empty list")
else:
for index, payload in enumerate(arms_payload, start=1):
if not isinstance(payload, dict):
arm = AugmentationAblationArmAudit(
name=f"arm-{index}",
profiles=[],
train_manifest="",
eval_manifest="",
audit_path="",
train_count=None,
eval_count=None,
issues=[f"arms[{index}] must be an object"],
)
arms.append(arm)
continue
arms.append(_audit_ablation_arm(payload, base_dir, index))
real_paths = [Path(path) for path in (real_review_audits or [])]
real_issues, real_warnings, real_evidence_count, verified_reference_count = _audit_real_review_evidence(real_paths)
issues.extend(real_issues)
warnings.extend(real_warnings)
for arm in arms:
issues.extend(f"{arm.name}: {issue}" for issue in arm.issues)
warnings.extend(f"{arm.name}: {warning}" for warning in arm.warnings)
promotion_ready = not issues and real_evidence_count > 0 and verified_reference_count > 0 and all(arm.passed for arm in arms)
out = Path(output_dir)
out.mkdir(parents=True, exist_ok=True)
report = AugmentationAblationAudit(
config_path=str(config_file),
output_json_path=str(out / "augmentation-ablation-audit.json"),
output_md_path=str(out / "augmentation-ablation-audit.md"),
ablation_id=ablation_id,
passed=promotion_ready,
promotion_ready=promotion_ready,
arm_count=len(arms),
real_review_audits=[str(path) for path in real_paths],
real_evidence_count=real_evidence_count,
verified_reference_count=verified_reference_count,
arms=arms,
issues=issues,
warnings=warnings,
)
_write_augmentation_ablation_audit(report)
return report
def _audit_ablation_arm(payload: dict[str, Any], base_dir: Path, index: int) -> AugmentationAblationArmAudit:
issues: list[str] = []
warnings: list[str] = []
name = _clean_str(payload.get("name")) or f"arm-{index}"
profiles = [str(item) for item in payload.get("profiles") or [] if str(item)]
if not profiles:
issues.append("profiles must be a non-empty list")
train_manifest = _clean_str(payload.get("train_manifest"))
eval_manifest = _clean_str(payload.get("eval_manifest"))
audit_path = _clean_str(payload.get("audit"))
train_count = _optional_int(payload.get("train_count"))
eval_count = _optional_int(payload.get("eval_count"))
train_manifest_count, train_claims = _audit_arm_manifest(train_manifest, base_dir, "train_manifest", train_count, issues)
eval_manifest_count, eval_claims = _audit_arm_manifest(eval_manifest, base_dir, "eval_manifest", eval_count, issues)
_audit_optional_passed_json(audit_path, base_dir, "audit", issues, warnings, require_zero_issues=True)
if train_claims:
issues.append(f"train_manifest has {train_claims} claim_evidence_eligible synthetic row(s)")
if eval_claims:
issues.append(f"eval_manifest has {eval_claims} claim_evidence_eligible synthetic row(s)")
return AugmentationAblationArmAudit(
name=name,
profiles=profiles,
train_manifest=train_manifest,
eval_manifest=eval_manifest,
audit_path=audit_path,
train_count=train_count,
eval_count=eval_count,
train_manifest_count=train_manifest_count,
eval_manifest_count=eval_manifest_count,
train_claim_eligible_count=train_claims,
eval_claim_eligible_count=eval_claims,
passed=not issues,
issues=issues,
warnings=warnings,
)
def _audit_arm_manifest(
path_value: str,
base_dir: Path,
field_name: str,
expected_count: int | None,
issues: list[str],
) -> tuple[int | None, int]:
if not path_value:
issues.append(f"{field_name} is required")
return None, 0
path = _resolve_config_path(path_value, base_dir)
if not path.is_file():
issues.append(f"{field_name} does not exist: {path}")
return None, 0
try:
entries = load_manifest(path)
except Exception as exc:
issues.append(f"{field_name} failed to load: {path}: {exc}")
return None, 0
if expected_count is not None and len(entries) != expected_count:
issues.append(f"{field_name} count mismatch: expected {expected_count}, found {len(entries)}")
claim_count = sum(1 for entry in entries if _claim_evidence_eligible(entry))
return len(entries), claim_count
def _audit_real_review_evidence(paths: list[Path]) -> tuple[list[str], list[str], int, int]:
issues: list[str] = []
warnings: list[str] = []
accepted_total = 0
verified_total = 0
if not paths:
issues.append("at least one real Gorkhapatra review audit is required before ablation promotion")
return issues, warnings, accepted_total, verified_total
for path in paths:
try:
payload = _read_json_object(path, "Gorkhapatra review audit")
except DataValidationError as exc:
issues.append(str(exc))
continue
if payload.get("passed") is not True:
issues.append(f"real review audit did not pass: {path}")
if payload.get("require_verified_references") is not True:
issues.append(f"real review audit was not run with --require-verified-references: {path}")
accepted = _optional_int(payload.get("accepted_count")) or 0
verified = _optional_int(payload.get("verified_reference_count")) or 0
if accepted < 1:
issues.append(f"real review audit has no accepted rows: {path}")
if verified < accepted:
issues.append(f"real review audit verified_reference_count is below accepted_count: {path}")
accepted_total += accepted
verified_total += verified
for warning in payload.get("warnings") or []:
warnings.append(f"{path}: {warning}")
return issues, warnings, accepted_total, verified_total
def _audit_optional_passed_json(
path_value: Any,
base_dir: Path,
field_name: str,
issues: list[str],
warnings: list[str],
*,
require_zero_issues: bool = False,
) -> None:
path_string = _clean_str(path_value)
if not path_string:
issues.append(f"{field_name} is required")
return
path = _resolve_config_path(path_string, base_dir)
try:
payload = _read_json_object(path, field_name)
except DataValidationError as exc:
issues.append(str(exc))
return
if payload.get("passed") is not True:
issues.append(f"{field_name} did not pass: {path}")
if require_zero_issues and payload.get("issues"):
issues.append(f"{field_name} has issues: {path}")
for warning in payload.get("warnings") or []:
warnings.append(f"{field_name}: {path}: {warning}")
def _read_json_object(path: Path, label: str) -> dict[str, Any]:
if not path.is_file():
raise DataValidationError(f"{label} does not exist: {path}")
try:
payload = json.loads(path.read_text(encoding="utf-8"))
except json.JSONDecodeError as exc:
raise DataValidationError(f"{label} is invalid JSON: {path}: {exc}") from exc
if not isinstance(payload, dict):
raise DataValidationError(f"{label} must be a JSON object: {path}")
return payload
def _resolve_config_path(path_value: str, base_dir: Path) -> Path:
path = Path(path_value)
if path.is_absolute():
return path
if path.exists():
return path
return base_dir / path
def _clean_str(value: Any) -> str:
return value.strip() if isinstance(value, str) else ""
def _optional_int(value: Any) -> int | None:
if isinstance(value, bool):
return None
if isinstance(value, int):
return value
return None
def _claim_evidence_eligible(entry: ManifestEntry) -> bool:
return (entry.metadata or {}).get("claim_evidence_eligible") is not False
def _write_augmentation_ablation_audit(report: AugmentationAblationAudit) -> None:
json_path = Path(report.output_json_path)
md_path = Path(report.output_md_path)
json_path.write_text(json.dumps(report.to_dict(), ensure_ascii=False, indent=2) + "\n", encoding="utf-8")
lines = [
"# Augmentation Ablation Audit",
"",
f"- ablation: `{report.ablation_id}`",
f"- passed: `{report.passed}`",
f"- promotion ready: `{report.promotion_ready}`",
f"- arms: `{report.arm_count}`",
f"- real review audits: `{len(report.real_review_audits)}`",
f"- real accepted rows: `{report.real_evidence_count}`",
f"- verified references: `{report.verified_reference_count}`",
"",
"## Arms",
"",
]
for arm in report.arms:
lines.extend(
[
f"### {arm.name}",
"",
f"- passed: `{arm.passed}`",
f"- profiles: `{', '.join(arm.profiles)}`",
f"- train rows: `{arm.train_manifest_count if arm.train_manifest_count is not None else 'unknown'}`",
f"- eval rows: `{arm.eval_manifest_count if arm.eval_manifest_count is not None else 'unknown'}`",
f"- train claim eligible rows: `{arm.train_claim_eligible_count}`",
f"- eval claim eligible rows: `{arm.eval_claim_eligible_count}`",
"",
]
)
if arm.issues:
lines.extend(["Issues:", ""])
lines.extend(f"- {issue}" for issue in arm.issues)
lines.append("")
if report.issues:
lines.extend(["## Issues", ""])
lines.extend(f"- {issue}" for issue in report.issues)
if report.warnings:
lines.extend(["", "## Warnings", ""])
lines.extend(f"- {warning}" for warning in report.warnings)
md_path.write_text("\n".join(lines) + "\n", encoding="utf-8")
|