File size: 16,367 Bytes
db5f09b | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 | """
AETHER Knowledge Graph Engine.
Integrates PyTorch Geometric patterns for relational reasoning:
- RGCN for node classification on knowledge graphs
- ComplEx for link prediction
- Neuro-symbolic bridge: learned attention over symbolic rules
"""
import torch
import torch.nn as nn
import torch.nn.functional as F
from typing import Dict, List, Any, Optional, Tuple
import networkx as nx
import numpy as np
import logging
logger = logging.getLogger("AETHER.Knowledge")
class RGCNLayer(nn.Module):
"""Simplified RGCN layer for knowledge graph reasoning."""
def __init__(self, in_dim: int, out_dim: int, num_relations: int,
num_bases: int = 4):
super().__init__()
self.in_dim = in_dim
self.out_dim = out_dim
self.num_relations = num_relations
self.num_bases = num_bases
self.bases = nn.Parameter(torch.Tensor(num_bases, in_dim, out_dim))
self.comp = nn.Parameter(torch.Tensor(num_relations, num_bases))
self.self_loop = nn.Parameter(torch.Tensor(in_dim, out_dim))
self.bias = nn.Parameter(torch.Tensor(out_dim))
self.reset_parameters()
def reset_parameters(self):
nn.init.xavier_uniform_(self.bases)
nn.init.xavier_uniform_(self.comp)
nn.init.xavier_uniform_(self.self_loop)
nn.init.zeros_(self.bias)
def forward(self, x: Optional[torch.Tensor], edge_index: torch.Tensor,
edge_type: torch.Tensor) -> torch.Tensor:
num_nodes = int(edge_index.max().item()) + 1 if x is None else x.size(0)
if x is None:
x = torch.eye(num_nodes, self.in_dim, device=edge_index.device)
weight = torch.einsum('rb, bio -> rio', self.comp, self.bases)
out = torch.zeros(num_nodes, self.out_dim, device=x.device)
for rel_id in range(self.num_relations):
mask = edge_type == rel_id
if mask.sum() == 0:
continue
rel_edges = edge_index[:, mask]
source = rel_edges[0]
target = rel_edges[1]
messages = torch.mm(x[source], weight[rel_id])
out.index_add_(0, target, messages)
out = out + torch.mm(x, self.self_loop)
out = out + self.bias
return out
class KnowledgeGraphEncoder(nn.Module):
"""Multi-layer RGCN encoder for knowledge graph embeddings."""
def __init__(self, num_nodes: int, hidden_dim: int, num_relations: int,
num_layers: int = 2, num_bases: int = 4):
super().__init__()
self.num_nodes = num_nodes
self.hidden_dim = hidden_dim
self.num_relations = num_relations
self.node_embeddings = nn.Embedding(num_nodes, hidden_dim)
self.layers = nn.ModuleList([
RGCNLayer(
in_dim=hidden_dim if i == 0 else hidden_dim,
out_dim=hidden_dim,
num_relations=num_relations,
num_bases=num_bases,
)
for i in range(num_layers)
])
self.norms = nn.ModuleList([
nn.LayerNorm(hidden_dim)
for _ in range(num_layers)
])
def forward(self, edge_index: torch.Tensor,
edge_type: torch.Tensor) -> torch.Tensor:
num_nodes = int(edge_index.max().item()) + 1
x = self.node_embeddings(torch.arange(num_nodes, device=edge_index.device))
for layer, norm in zip(self.layers, self.norms):
x_new = layer(x, edge_index, edge_type)
x_new = F.relu(norm(x_new))
x = x_new
return x
class ComplExScorer(nn.Module):
"""ComplEx link prediction scorer for knowledge graph completion."""
def __init__(self, num_nodes: int, num_relations: int, hidden_dim: int = 50):
super().__init__()
self.num_nodes = num_nodes
self.num_relations = num_relations
self.hidden_dim = hidden_dim
self.head_real = nn.Embedding(num_nodes, hidden_dim)
self.head_imag = nn.Embedding(num_nodes, hidden_dim)
self.tail_real = nn.Embedding(num_nodes, hidden_dim)
self.tail_imag = nn.Embedding(num_nodes, hidden_dim)
self.rel_real = nn.Embedding(num_relations, hidden_dim)
self.rel_imag = nn.Embedding(num_relations, hidden_dim)
self.reset_parameters()
def reset_parameters(self):
for param in self.parameters():
nn.init.xavier_uniform_(param)
def forward(self, head_idx: torch.Tensor, rel_idx: torch.Tensor,
tail_idx: torch.Tensor) -> torch.Tensor:
hr = self.head_real(head_idx)
hi = self.head_imag(head_idx)
tr = self.tail_real(tail_idx)
ti = self.tail_imag(tail_idx)
rr = self.rel_real(rel_idx)
ri = self.rel_imag(rel_idx)
score = torch.sum(
hr * rr * tr + hr * ri * ti + hi * rr * ti - hi * ri * tr,
dim=-1
)
return score
def loss(self, head_idx: torch.Tensor, rel_idx: torch.Tensor,
tail_idx: torch.Tensor, negative_head: torch.Tensor = None,
negative_tail: torch.Tensor = None) -> torch.Tensor:
pos_score = self.forward(head_idx, rel_idx, tail_idx)
if negative_head is not None:
neg_score = self.forward(negative_head, rel_idx, tail_idx)
elif negative_tail is not None:
neg_score = self.forward(head_idx, rel_idx, negative_tail)
else:
neg_tail = torch.randint(0, self.num_nodes, tail_idx.size(),
device=tail_idx.device)
neg_score = self.forward(head_idx, rel_idx, neg_tail)
pos_loss = F.softplus(-pos_score)
neg_loss = F.softplus(neg_score)
return (pos_loss + neg_loss).mean()
class KnowledgeGraphEngine(nn.Module):
"""
Unified knowledge graph engine combining:
- NetworkX for graph construction and symbolic reasoning
- RGCN for learned embeddings
- ComplEx for link prediction
- Neuro-symbolic bridge for AETHER integration
"""
def __init__(self, embedding_dim: int = 128, num_relations: int = 20,
max_nodes: int = 10000):
super().__init__()
self.embedding_dim = embedding_dim
self.num_relations = num_relations
self.max_nodes = max_nodes
self.graph = nx.DiGraph()
self.node_id_map: Dict[str, int] = {}
self.relation_map: Dict[str, int] = {}
self.next_node_id = 0
self.next_rel_id = 0
self.encoder: Optional[KnowledgeGraphEncoder] = None
self.scorer: Optional[ComplExScorer] = None
self.symbolic_attention = nn.Parameter(torch.ones(num_relations))
self.rules: List[Tuple[str, str, str]] = []
def _get_or_create_node(self, node_name: str) -> int:
if node_name not in self.node_id_map:
self.node_id_map[node_name] = self.next_node_id
self.graph.add_node(self.next_node_id, name=node_name)
self.next_node_id += 1
return self.node_id_map[node_name]
def _get_or_create_relation(self, relation: str) -> int:
if relation not in self.relation_map:
self.relation_map[relation] = self.next_rel_id
self.next_rel_id += 1
return self.relation_map[relation]
def add_fact(self, head: str, relation: str, tail: str,
confidence: float = 1.0):
h_id = self._get_or_create_node(head)
t_id = self._get_or_create_node(tail)
r_id = self._get_or_create_relation(relation)
self.graph.add_edge(h_id, t_id, relation=r_id, name=relation,
confidence=confidence)
self._ensure_model_capacity()
def add_rule(self, premise: Tuple[str, str, str],
conclusion: Tuple[str, str, str]):
self.rules.append((premise, conclusion))
def _ensure_model_capacity(self):
if self.encoder is None and self.next_node_id > 0:
num_nodes = min(self.next_node_id, self.max_nodes)
num_rels = max(self.next_rel_id, self.num_relations)
self.encoder = KnowledgeGraphEncoder(
num_nodes=num_nodes,
hidden_dim=self.embedding_dim,
num_relations=num_rels,
num_layers=2,
)
self.scorer = ComplExScorer(
num_nodes=num_nodes,
num_relations=num_rels,
hidden_dim=self.embedding_dim // 2,
)
logger.info(f"Initialized KG models: {num_nodes} nodes, {num_rels} relations")
def reason_symbolic(self, query_head: str, query_relation: str) -> List[Dict]:
results = []
if query_head not in self.node_id_map:
return results
h_id = self.node_id_map[query_head]
r_name = query_relation
if r_name in self.relation_map:
r_id = self.relation_map[r_name]
for _, target, data in self.graph.out_edges(h_id, data=True):
if data.get('relation') == r_id:
results.append({
"head": query_head,
"relation": r_name,
"tail": self.graph.nodes[target].get('name', str(target)),
"confidence": data.get('confidence', 1.0),
"path": "direct",
})
for premise, conclusion in self.rules:
p_head, p_rel, p_tail = premise
c_head, c_rel, c_tail = conclusion
if p_head == query_head and self._check_fact(premise):
inferred_tail = c_tail
if c_head == "?":
c_head = query_head
results.append({
"head": c_head,
"relation": c_rel,
"tail": inferred_tail,
"confidence": 0.8,
"path": "inferred",
"rule": f"{premise} -> {conclusion}",
})
for neighbor in nx.bfs_tree(self.graph, h_id, depth_limit=2).nodes():
if neighbor != h_id:
for path in nx.all_simple_paths(self.graph, h_id, neighbor, cutoff=2):
if len(path) > 1:
edge_data = self.graph.edges[path[0], path[1]]
results.append({
"head": query_head,
"relation": f"multi-hop via {edge_data.get('name', 'unknown')}",
"tail": self.graph.nodes[neighbor].get('name', str(neighbor)),
"confidence": 0.6 ** (len(path) - 1),
"path": "->".join(str(n) for n in path),
})
return sorted(results, key=lambda x: x["confidence"], reverse=True)
def _check_fact(self, fact: Tuple[str, str, str]) -> bool:
h, r, t = fact
if h not in self.node_id_map or t not in self.node_id_map:
return False
h_id = self.node_id_map[h]
t_id = self.node_id_map[t]
if r not in self.relation_map:
return False
r_id = self.relation_map[r]
return self.graph.has_edge(h_id, t_id) and \
self.graph.edges[h_id, t_id].get('relation') == r_id
def reason_learned(self, query_head: str, query_relation: str,
top_k: int = 5) -> List[Dict]:
if self.scorer is None or query_head not in self.node_id_map:
return []
h_id = self.node_id_map[query_head]
r_id = self.relation_map.get(query_relation)
if r_id is None:
return []
h_tensor = torch.tensor([h_id])
r_tensor = torch.tensor([r_id])
all_tails = torch.arange(self.scorer.num_nodes)
scores = []
batch_size = 1000
for i in range(0, len(all_tails), batch_size):
batch_tails = all_tails[i:i + batch_size]
h_batch = h_tensor.repeat(len(batch_tails))
r_batch = r_tensor.repeat(len(batch_tails))
batch_scores = self.scorer(h_batch, r_batch, batch_tails)
scores.extend(batch_scores.tolist())
scores = torch.tensor(scores)
top_scores, top_indices = torch.topk(scores, min(top_k, len(scores)))
results = []
for idx, score in zip(top_indices, top_scores):
node_name = self.graph.nodes[idx.item()].get('name', str(idx.item()))
results.append({
"head": query_head,
"relation": query_relation,
"tail": node_name,
"confidence": torch.sigmoid(score).item(),
"path": "learned",
})
return results
def query(self, text_query: str, top_k: int = 5) -> Dict[str, Any]:
parts = text_query.lower().split()
if len(parts) >= 2:
head = parts[0].capitalize()
relation = " ".join(parts[1:])
else:
head = text_query.capitalize()
relation = "related_to"
symbolic_results = self.reason_symbolic(head, relation)
learned_results = self.reason_learned(head, relation, top_k)
rel_id = self.relation_map.get(relation, 0)
symbolic_weight = torch.sigmoid(self.symbolic_attention[rel_id % self.num_relations])
learned_weight = 1.0 - symbolic_weight.item()
all_results = []
for r in symbolic_results[:top_k]:
r["source"] = "symbolic"
r["fusion_weight"] = symbolic_weight.item()
all_results.append(r)
for r in learned_results[:top_k]:
r["source"] = "learned"
r["fusion_weight"] = learned_weight
all_results.append(r)
all_results.sort(key=lambda x: x.get("confidence", 0), reverse=True)
return {
"query": text_query,
"results": all_results[:top_k],
"symbolic_weight": symbolic_weight.item(),
"learned_weight": learned_weight,
"num_symbolic": len(symbolic_results),
"num_learned": len(learned_results),
}
def to_pyg_data(self) -> Dict[str, torch.Tensor]:
edges = []
edge_types = []
for u, v, data in self.graph.edges(data=True):
edges.append([u, v])
edge_types.append(data.get('relation', 0))
if not edges:
return {}
edge_index = torch.tensor(edges, dtype=torch.long).t()
edge_type = torch.tensor(edge_types, dtype=torch.long)
return {
"edge_index": edge_index,
"edge_type": edge_type,
"num_nodes": self.next_node_id,
"num_relations": self.next_rel_id,
}
def stats(self) -> Dict[str, Any]:
return {
"num_nodes": self.graph.number_of_nodes(),
"num_edges": self.graph.number_of_edges(),
"num_relations": len(self.relation_map),
"num_rules": len(self.rules),
"node_names": len(self.node_id_map),
}
def export(self) -> Dict[str, Any]:
edges = []
for u, v, data in self.graph.edges(data=True):
edges.append({
"source": u,
"target": v,
"relation_id": data.get('relation'),
"relation_name": data.get('name'),
"confidence": data.get('confidence'),
})
return {
"nodes": {n: self.graph.nodes[n].get('name', str(n))
for n in self.graph.nodes()},
"edges": edges,
"node_id_map": self.node_id_map,
"relation_map": self.relation_map,
"rules": self.rules,
}
|