Spaces:
Sleeping
Sleeping
File size: 14,283 Bytes
8125804 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 | from __future__ import annotations
from typing import Any
from src.model.anchor_tree import make_anchor_tree
from src.model.anchor_tree_domain import detect_tree_domain
from src.model.anchor_tree_types import AnchorTree, AnchorTreeEdge, AnchorTreeNode, AnchorTreeRelation, AnchorTreeRole
from src.model.anchor_types import AnchorRecord
_MATH_STEP_ORDER = {
"integration_by_parts_only": 0,
"select_u_and_dv": 1,
"derive_du_and_v": 2,
"substitute_uv_minus_int_vdu": 3,
"reduce_integral_complexity": 4,
"repeat_if_needed": 5,
"simplify_result": 6,
"integration_constant": 7,
"shortcut_lookup": 50,
"table_reference": 51,
"substitution_switch": 52,
"meta_abort": 53,
"wrong_symbolic_step": 54,
}
_CODE_STEP_ORDER = {
"async_fastapi_service": 0,
"typed_request_models": 1,
"dependency_injection": 2,
"async_handlers": 3,
"validation_path": 4,
"background_jobs": 5,
"deployment_notes": 6,
"django_view_reframe": 50,
"synchronous_handler_reframe": 51,
"template_rendering_branch": 52,
}
_QUANTIFIER_STEP_ORDER = {
"universal_quantifier_scope": 0,
"preserve_universal_claim": 1,
"reject_existential_drift": 2,
"restate_universal_conclusion": 3,
"existential_witness_shift": 50,
"drop_universal_scope": 51,
}
_PROOF_MODE_STEP_ORDER = {
"proof_by_contradiction_mode": 0,
"maintain_negation_assumption": 1,
"derive_contradiction": 2,
"discharge_negation_assumption": 3,
"direct_proof_switch": 50,
"constructive_reset": 51,
}
def _domain_root_label(domain: str | None) -> str:
if domain == "math_ibp":
return "integration_by_parts_only"
if domain == "code_fastapi":
return "async_fastapi_service"
if domain == "quantifier":
return "universal_quantifier_scope"
if domain == "proof_mode":
return "proof_by_contradiction_mode"
return "observed_root"
def _step_order(domain: str | None) -> dict[str, int]:
if domain == "math_ibp":
return _MATH_STEP_ORDER
if domain == "code_fastapi":
return _CODE_STEP_ORDER
if domain == "quantifier":
return _QUANTIFIER_STEP_ORDER
if domain == "proof_mode":
return _PROOF_MODE_STEP_ORDER
return {}
def _classify_math_label(text: str) -> str:
lowered = text.lower()
if "integration by parts" in lowered:
return "integration_by_parts_only"
if ("let u" in lowered or "u =" in lowered) and "dv" in lowered:
return "select_u_and_dv"
if "du" in lowered and ("v =" in lowered or " v " in f" {lowered} "):
return "derive_du_and_v"
if "uv" in lowered or "vdu" in lowered or "substitut" in lowered:
return "substitute_uv_minus_int_vdu"
if "remaining integral" in lowered or "reduce" in lowered:
return "reduce_integral_complexity"
if "repeat" in lowered or "again" in lowered:
return "repeat_if_needed"
if "+ c" in lowered or "+c" in lowered or "constant of integration" in lowered:
return "integration_constant"
if "simplif" in lowered:
return "simplify_result"
if "shortcut" in lowered:
return "shortcut_lookup"
if "table" in lowered:
return "table_reference"
if "substitution" in lowered:
return "substitution_switch"
if any(marker in lowered for marker in ("too hard", "challenging", "no clear path", "alternative approach")):
return "meta_abort"
return "math_observed_step"
def _classify_code_label(text: str) -> str:
lowered = text.lower()
if "fastapi" in lowered and "async" in lowered:
return "async_fastapi_service"
if "pydantic" in lowered or "request model" in lowered or "response model" in lowered:
return "typed_request_models"
if "dependency injection" in lowered or "depends(" in lowered:
return "dependency_injection"
if "async handler" in lowered or ("async def" in lowered and "await" in lowered):
return "async_handlers"
if "validation" in lowered or "validate" in lowered:
return "validation_path"
if "background task" in lowered or "background job" in lowered:
return "background_jobs"
if "deploy" in lowered or "uvicorn" in lowered or "gunicorn" in lowered:
return "deployment_notes"
if "django" in lowered:
return "django_view_reframe"
if "synchronous" in lowered or "sync view" in lowered:
return "synchronous_handler_reframe"
if "template" in lowered or "render" in lowered:
return "template_rendering_branch"
return "code_observed_step"
def _classify_quantifier_label(text: str) -> str:
lowered = text.lower()
if "restore the original universal" in lowered or "restate the universal" in lowered:
return "restate_universal_conclusion"
if "reject" in lowered and ("existential" in lowered or "witness" in lowered):
return "reject_existential_drift"
if "there exists" in lowered or "one witness" in lowered or "existential" in lowered:
return "existential_witness_shift"
if "drop" in lowered and "universal" in lowered:
return "drop_universal_scope"
if "for all" in lowered or "universal" in lowered:
return "preserve_universal_claim"
return "quantifier_observed_step"
def _classify_proof_mode_label(text: str) -> str:
lowered = text.lower()
if "contradiction" in lowered and ("keep" in lowered or "return" in lowered or "mode" in lowered):
return "proof_by_contradiction_mode"
if "assume the negation" in lowered or "assumed negation" in lowered or "negation assumption" in lowered:
return "maintain_negation_assumption"
if "derive a contradiction" in lowered or "contradiction structure" in lowered:
return "derive_contradiction"
if "assumption was false" in lowered or "discharge the assumed negation" in lowered:
return "discharge_negation_assumption"
if "direct proof" in lowered:
return "direct_proof_switch"
if "constructive proof" in lowered or "from scratch" in lowered:
return "constructive_reset"
return "proof_mode_observed_step"
def classify_observed_label(domain: str | None, text: str) -> str:
if domain == "math_ibp":
return _classify_math_label(text)
if domain == "code_fastapi":
return _classify_code_label(text)
if domain == "quantifier":
return _classify_quantifier_label(text)
if domain == "proof_mode":
return _classify_proof_mode_label(text)
return "observed_step"
def _role_for_label(label: str, source: str, is_root: bool) -> AnchorTreeRole:
if is_root:
return AnchorTreeRole.CONSTRAINT
if label in {
"shortcut_lookup",
"table_reference",
"substitution_switch",
"wrong_symbolic_step",
"django_view_reframe",
"synchronous_handler_reframe",
"template_rendering_branch",
"existential_witness_shift",
"drop_universal_scope",
"direct_proof_switch",
"constructive_reset",
}:
return AnchorTreeRole.DRIFT
if label == "meta_abort":
return AnchorTreeRole.META
if source == "auxiliary_proposal":
return AnchorTreeRole.REPAIR
if source == "future_hint":
return AnchorTreeRole.DERIVED
return AnchorTreeRole.STEP
def _score_anchor(anchor: AnchorRecord) -> float:
support = float(anchor.support.detach().item()) if hasattr(anchor.support, "detach") else float(anchor.support)
viability = float(anchor.viability.detach().item()) if hasattr(anchor.viability, "detach") else float(anchor.viability)
return support * max(viability, 1e-6)
def _make_root_node(domain: str | None, text: str, root_anchor: AnchorRecord | None) -> AnchorTreeNode:
root_score = _score_anchor(root_anchor) if root_anchor is not None else 0.0
root_repr = root_anchor.repr.detach().clone() if root_anchor is not None else None
root_start = int(root_anchor.start_idx) if root_anchor is not None else 0
root_end = int(root_anchor.end_idx) if root_anchor is not None else max(0, len(text.split()) - 1)
return AnchorTreeNode(
node_id="root",
label=_domain_root_label(domain),
text=text,
depth=0,
role=AnchorTreeRole.CONSTRAINT,
source="prompt",
anchor_id=None if root_anchor is None else int(root_anchor.id),
span_start=root_start,
span_end=root_end,
repr=root_repr,
score=float(root_score),
)
def _make_anchor_payloads(active_anchors: list[dict[str, Any] | AnchorRecord]) -> list[dict[str, Any]]:
payloads: list[dict[str, Any]] = []
for item in active_anchors:
if isinstance(item, AnchorRecord):
payloads.append({
"anchor": item,
"text": f"anchor_{item.id}",
"start": int(item.start_idx),
"end": int(item.end_idx),
})
else:
payloads.append(item)
payloads.sort(key=lambda payload: (int(payload.get("start", 0)), int(payload.get("end", 0))))
return payloads
def build_observed_tree(
*,
text: str,
active_anchors: list[dict[str, Any] | AnchorRecord],
future_hint_candidates: list[dict[str, Any]],
auxiliary_proposals: list[dict[str, Any]],
domain: str | None = None,
) -> AnchorTree:
anchor_payloads = _make_anchor_payloads(active_anchors)
anchor_texts = [str(payload.get("text", "")) for payload in anchor_payloads]
resolved_domain = domain or detect_tree_domain(text=text, anchor_texts=anchor_texts)
root_anchor = None
if anchor_payloads:
root_anchor = max(
(payload["anchor"] for payload in anchor_payloads if isinstance(payload.get("anchor"), AnchorRecord)),
key=_score_anchor,
default=None,
)
root = _make_root_node(resolved_domain, text, root_anchor)
nodes: list[AnchorTreeNode] = []
edges: list[AnchorTreeEdge] = []
order_map = _step_order(resolved_domain)
normalized_items: list[dict[str, Any]] = []
for idx, payload in enumerate(anchor_payloads):
anchor = payload.get("anchor")
payload_text = str(payload.get("text", "")).strip()
node_label = classify_observed_label(resolved_domain, payload_text)
if payload_text and node_label != root.label:
normalized_items.append(
{
"node_id": f"anchor_{idx}",
"label": node_label,
"text": payload_text,
"span_start": int(payload.get("start", 0)),
"span_end": int(payload.get("end", 0)),
"repr": None if anchor is None else anchor.repr.detach().clone(),
"score": 0.0 if anchor is None else _score_anchor(anchor),
"source": "active_anchor",
}
)
for idx, hint in enumerate(future_hint_candidates):
hint_text = str(hint.get("text", "")).strip()
if not hint_text:
continue
normalized_items.append(
{
"node_id": f"hint_{idx}",
"label": classify_observed_label(resolved_domain, hint_text),
"text": hint_text,
"span_start": int(hint.get("start", 0)),
"span_end": int(hint.get("end", hint.get("start", 0))),
"repr": None,
"score": float(hint.get("mean_score", 0.0)),
"source": "future_hint",
}
)
for idx, proposal in enumerate(auxiliary_proposals):
proposal_text = str(proposal.get("proposal_text", "")).strip()
start, end = proposal.get("proposal_span", (0, 0))
if not proposal_text:
continue
normalized_items.append(
{
"node_id": f"proposal_{idx}",
"label": classify_observed_label(resolved_domain, proposal_text),
"text": proposal_text,
"span_start": int(start),
"span_end": int(end),
"repr": proposal.get("repr"),
"score": float(proposal.get("proposal_score", 0.0)),
"source": "auxiliary_proposal",
}
)
normalized_items.sort(
key=lambda item: (
order_map.get(item["label"], 999),
int(item["span_start"]),
item["node_id"],
)
)
last_progress_node_id = root.node_id
for depth_idx, item in enumerate(normalized_items, start=1):
label = str(item["label"])
role = _role_for_label(label, str(item["source"]), is_root=False)
node = AnchorTreeNode(
node_id=str(item["node_id"]),
label=label,
text=str(item["text"]),
depth=depth_idx if role not in {AnchorTreeRole.DRIFT, AnchorTreeRole.META} else depth_idx,
role=role,
source=str(item["source"]),
span_start=int(item["span_start"]),
span_end=int(item["span_end"]),
repr=item.get("repr"),
score=float(item["score"]),
drift_flag=role in {AnchorTreeRole.DRIFT, AnchorTreeRole.META},
)
nodes.append(node)
if role in {AnchorTreeRole.DRIFT, AnchorTreeRole.META}:
parent_id = last_progress_node_id
relation = AnchorTreeRelation.ALTERNATIVE_TO
else:
parent_id = last_progress_node_id
relation = AnchorTreeRelation.EXPECTED_NEXT if parent_id != root.node_id else AnchorTreeRelation.CHILD
last_progress_node_id = node.node_id
edges.append(AnchorTreeEdge(parent_id=parent_id, child_id=node.node_id, relation=relation, score=float(node.score)))
tree = make_anchor_tree(
tree_id=f"observed_{resolved_domain or 'unknown'}",
root=root,
nodes=nodes,
edges=edges,
domain=resolved_domain or "unknown",
source_kind="observed",
)
tree.metadata["input_text"] = text
tree.metadata["anchor_count"] = len(anchor_payloads)
tree.metadata["future_hint_count"] = len(future_hint_candidates)
tree.metadata["auxiliary_proposal_count"] = len(auxiliary_proposals)
return tree
|