File size: 17,742 Bytes
63dd1f4 653b8c1 63dd1f4 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 | """
AbstractReasoner — Vitalis FSI
Reasons about RELATIONSHIPS between concepts.
Not pattern matching. Not retrieval.
Genuine relational reasoning:
- Analogy: A is to B as C is to ?
- Composition: concept_A + concept_B = novel_concept
- Inversion: what is the opposite of this concept?
- Transitivity: if A relates to B and B relates to C, what does A relate to C?
Built entirely on HDC operations. No external models.
"""
import numpy as np
import os
import json
import time
from vitalis_ide.math_core.kernel import VitalisKernel
from src.cognition.abstraction import AbstractionEngine
from src.hippocampus import Hippocampus
class AbstractReasoner:
ANALOGY_THRESHOLD = 0.25
COMPOSITION_DECAY = 0.85
INVERSION_SHIFT = 5000
def __init__(self):
self.kernel = VitalisKernel()
self.abstraction = AbstractionEngine()
self.hippocampus = Hippocampus()
self.path = os.path.expanduser(
"~/.vitalis_workspace/reasoning_log.json"
)
self._log = self._load_log()
def _load_log(self) -> list:
if os.path.exists(self.path):
with open(self.path) as f:
return json.load(f)
return []
def _save_log(self):
os.makedirs(os.path.dirname(self.path), exist_ok=True)
with open(self.path, "w") as f:
json.dump(self._log[-500:], f, indent=2)
# ------------------------------------------------------------------
# Core HDC reasoning operations
# ------------------------------------------------------------------
def _bind(self, a: np.ndarray, b: np.ndarray) -> np.ndarray:
"""Bipolar binding: element-wise multiply."""
return (a.astype(np.int32) * b.astype(np.int32)).astype(np.int8)
def _bundle(self, vecs: list) -> np.ndarray:
"""Bipolar bundling: sum then sign."""
stacked = np.stack(vecs).astype(np.int32).sum(axis=0)
result = np.sign(stacked).astype(np.int8)
result[result == 0] = 1
return result
def _invert(self, vec: np.ndarray) -> np.ndarray:
"""
Semantic inversion: cyclic shift by half the vector length.
Produces a vector maximally dissimilar to the input.
"""
return np.roll(vec, self.INVERSION_SHIFT)
# ------------------------------------------------------------------
# Analogy: A is to B as C is to ?
# ------------------------------------------------------------------
def analogy(
self,
concept_a: str,
concept_b: str,
concept_c: str,
) -> dict:
"""
Solves: A:B :: C:?
HDC method: ? = bind(bind(A, B), C)
Searches abstraction space and hippocampus for closest match.
"""
vec_a = self.kernel.vectorize_tokens(concept_a.split(), positional=False)
vec_b = self.kernel.vectorize_tokens(concept_b.split(), positional=False)
vec_c = self.kernel.vectorize_tokens(concept_c.split(), positional=False)
# ? = B * A^-1 * C (HDC analogy formula)
a_inv = self._bind(vec_a, vec_a) # A bound with itself = identity-like
relation = self._bind(vec_a, vec_b) # encode A→B relationship
answer_vec = self._bind(relation, vec_c) # apply relation to C
# Search for closest concept
candidates = self.abstraction.query_abstractions(answer_vec, top_k=3)
hipp_results = self.hippocampus.similarity_search(answer_vec, top_k=3)
best_match = None
best_score = -1.0
for score, name, _ in candidates:
if score > best_score:
best_score = score
best_match = name
result = {
"type": "analogy",
"query": f"{concept_a}:{concept_b}::{concept_c}:?",
"answer_vec": answer_vec,
"best_match": best_match,
"confidence": round(float(best_score), 4),
"candidates": [(name, round(float(s), 4)) for s, name, _ in candidates],
"timestamp": time.time(),
}
self._log.append({k: v for k, v in result.items() if k != "answer_vec"})
self._save_log()
return result
# ------------------------------------------------------------------
# Composition: merge two concepts into a novel one
# ------------------------------------------------------------------
def compose(self, concept_a: str, concept_b: str) -> dict:
"""
Compose two concepts into a novel concept vector.
The result occupies a position in the space between both inputs.
Weighted by the COMPOSITION_DECAY to prevent drift.
"""
vec_a = self.kernel.vectorize_tokens(concept_a.split(), positional=False)
vec_b = self.kernel.vectorize_tokens(concept_b.split(), positional=False)
# Bundle with decay weighting
composed = self._bundle([vec_a, vec_b])
# Apply composition decay — prevents the result from being
# too close to either parent
noise_mask = np.random.choice(
[-1, 1],
size=self.kernel.dim,
p=[1 - self.COMPOSITION_DECAY, self.COMPOSITION_DECAY]
).astype(np.int8)
composed = self._bind(composed, noise_mask)
# Search for nearest existing concept
candidates = self.abstraction.query_abstractions(composed, top_k=3)
result = {
"type": "composition",
"inputs": [concept_a, concept_b],
"novel_vec": composed,
"nearest": [(name, round(float(s), 4)) for s, name, _ in candidates],
"novelty": round(1.0 - (candidates[0][0] if candidates else 0.0), 4),
"timestamp": time.time(),
}
self._log.append({k: v for k, v in result.items() if k != "novel_vec"})
self._save_log()
return result
# ------------------------------------------------------------------
# Inversion: what is the conceptual opposite?
# ------------------------------------------------------------------
def invert(self, concept: str) -> dict:
"""
Find the conceptual opposite of a concept.
Uses cyclic shift inversion then searches concept space.
"""
vec = self.kernel.vectorize_tokens(concept.split(), positional=False)
inverted = self._invert(vec)
candidates = self.abstraction.query_abstractions(inverted, top_k=3)
hipp_results = self.hippocampus.similarity_search(inverted, top_k=3)
result = {
"type": "inversion",
"concept": concept,
"opposites": [(name, round(float(s), 4)) for s, name, _ in candidates],
"confidence": round(float(candidates[0][0]) if candidates else 0.0, 4),
"timestamp": time.time(),
}
self._log.append(result)
self._save_log()
return result
# ------------------------------------------------------------------
# Transitivity: if A→B and B→C, what is A→C?
# ------------------------------------------------------------------
def transitive_chain(self, concepts: list) -> dict:
"""
Chain reasoning: given [A, B, C, D...],
derive the relationship between A and the last element.
Each step binds the accumulated relationship with the next concept.
"""
if len(concepts) < 2:
return {"error": "Need at least 2 concepts"}
vecs = [
self.kernel.vectorize_tokens(c.split(), positional=False)
for c in concepts
]
# Accumulate relationship via sequential binding
accumulated = vecs[0].copy()
for i in range(1, len(vecs)):
accumulated = self._bind(accumulated, vecs[i])
# Apply position-aware permutation at each step
accumulated = np.roll(accumulated, i * 100)
candidates = self.abstraction.query_abstractions(accumulated, top_k=3)
result = {
"type": "transitivity",
"chain": concepts,
"conclusion": [(name, round(float(s), 4)) for s, name, _ in candidates],
"confidence": round(float(candidates[0][0]) if candidates else 0.0, 4),
"timestamp": time.time(),
}
self._log.append(result)
self._save_log()
return result
def report(self) -> dict:
if not self._log:
return {"status": "No reasoning performed yet"}
type_counts = {}
for entry in self._log:
t = entry.get("type", "unknown")
type_counts[t] = type_counts.get(t, 0) + 1
return {
"total_reasoning_ops": len(self._log),
"by_type": type_counts,
"recent": self._log[-3:],
}
"""
AbstractReasoner — Vitalis FSI
Reasons about RELATIONSHIPS between concepts.
Not pattern matching. Not retrieval.
Genuine relational reasoning:
- Analogy: A is to B as C is to ?
- Composition: concept_A + concept_B = novel_concept
- Inversion: what is the opposite of this concept?
- Transitivity: if A relates to B and B relates to C, what does A relate to C?
Built entirely on HDC operations. No external models.
"""
import numpy as np
import os
import json
import time
from vitalis_ide.math_core.kernel import VitalisKernel
from src.cognition.abstraction import AbstractionEngine
from src.hippocampus import Hippocampus
class AbstractReasoner:
ANALOGY_THRESHOLD = 0.25
COMPOSITION_DECAY = 0.85
INVERSION_SHIFT = 5000
def __init__(self):
self.kernel = VitalisKernel()
self.abstraction = AbstractionEngine()
self.hippocampus = Hippocampus()
self.path = os.path.expanduser(
"~/.vitalis_workspace/reasoning_log.json"
)
self._log = self._load_log()
def _load_log(self) -> list:
if os.path.exists(self.path):
with open(self.path) as f:
return json.load(f)
return []
def _save_log(self):
os.makedirs(os.path.dirname(self.path), exist_ok=True)
with open(self.path, "w") as f:
json.dump(self._log[-500:], f, indent=2)
# ------------------------------------------------------------------
# Core HDC reasoning operations
# ------------------------------------------------------------------
def _bind(self, a: np.ndarray, b: np.ndarray) -> np.ndarray:
"""Bipolar binding: element-wise multiply."""
return (a.astype(np.int32) * b.astype(np.int32)).astype(np.int8)
def _bundle(self, vecs: list) -> np.ndarray:
"""Bipolar bundling: sum then sign."""
stacked = np.stack(vecs).astype(np.int32).sum(axis=0)
result = np.sign(stacked).astype(np.int8)
result[result == 0] = 1
return result
def _invert(self, vec: np.ndarray) -> np.ndarray:
"""
Semantic inversion: cyclic shift by half the vector length.
Produces a vector maximally dissimilar to the input.
"""
return np.roll(vec, self.INVERSION_SHIFT)
# ------------------------------------------------------------------
# Analogy: A is to B as C is to ?
# ------------------------------------------------------------------
def analogy(
self,
concept_a: str,
concept_b: str,
concept_c: str,
) -> dict:
"""
Solves: A:B :: C:?
HDC method: ? = bind(bind(A, B), C)
Searches abstraction space and hippocampus for closest match.
"""
vec_a = self.kernel.vectorize_tokens(concept_a.split(), positional=False)
vec_b = self.kernel.vectorize_tokens(concept_b.split(), positional=False)
vec_c = self.kernel.vectorize_tokens(concept_c.split(), positional=False)
# ? = B * A^-1 * C (HDC analogy formula)
a_inv = self._bind(vec_a, vec_a) # A bound with itself = identity-like
relation = self._bind(vec_a, vec_b) # encode A→B relationship
answer_vec = self._bind(relation, vec_c) # apply relation to C
# Search for closest concept
candidates = self.abstraction.query_abstractions(answer_vec, top_k=3)
hipp_results = self.hippocampus.similarity_search(answer_vec, top_k=3)
best_match = None
best_score = -1.0
for score, name, _ in candidates:
if score > best_score:
best_score = score
best_match = name
result = {
"type": "analogy",
"query": f"{concept_a}:{concept_b}::{concept_c}:?",
"answer_vec": answer_vec,
"best_match": best_match,
"confidence": round(float(best_score), 4),
"candidates": [(name, round(float(s), 4)) for s, name, _ in candidates],
"timestamp": time.time(),
}
self._log.append({k: v for k, v in result.items() if k != "answer_vec"})
self._save_log()
return result
# ------------------------------------------------------------------
# Composition: merge two concepts into a novel one
# ------------------------------------------------------------------
def compose(self, concept_a: str, concept_b: str) -> dict:
"""
Compose two concepts into a novel concept vector.
The result occupies a position in the space between both inputs.
Weighted by the COMPOSITION_DECAY to prevent drift.
"""
vec_a = self.kernel.vectorize_tokens(concept_a.split(), positional=False)
vec_b = self.kernel.vectorize_tokens(concept_b.split(), positional=False)
# Bundle with decay weighting
composed = self._bundle([vec_a, vec_b])
# Apply composition decay — prevents the result from being
# too close to either parent
noise_mask = np.random.choice(
[-1, 1],
size=self.kernel.dim,
p=[1 - self.COMPOSITION_DECAY, self.COMPOSITION_DECAY]
).astype(np.int8)
composed = self._bind(composed, noise_mask)
# Search for nearest existing concept
candidates = self.abstraction.query_abstractions(composed, top_k=3)
result = {
"type": "composition",
"inputs": [concept_a, concept_b],
"novel_vec": composed,
"nearest": [(name, round(float(s), 4)) for s, name, _ in candidates],
"novelty": round(1.0 - (candidates[0][0] if candidates else 0.0), 4),
"timestamp": time.time(),
}
self._log.append({k: v for k, v in result.items() if k != "novel_vec"})
self._save_log()
return result
# ------------------------------------------------------------------
# Inversion: what is the conceptual opposite?
# ------------------------------------------------------------------
def invert(self, concept: str) -> dict:
"""
Find the conceptual opposite of a concept.
Uses cyclic shift inversion then searches concept space.
"""
vec = self.kernel.vectorize_tokens(concept.split(), positional=False)
inverted = self._invert(vec)
candidates = self.abstraction.query_abstractions(inverted, top_k=3)
hipp_results = self.hippocampus.similarity_search(inverted, top_k=3)
result = {
"type": "inversion",
"concept": concept,
"opposites": [(name, round(float(s), 4)) for s, name, _ in candidates],
"confidence": round(float(candidates[0][0]) if candidates else 0.0, 4),
"timestamp": time.time(),
}
self._log.append(result)
self._save_log()
return result
# ------------------------------------------------------------------
# Transitivity: if A→B and B→C, what is A→C?
# ------------------------------------------------------------------
def transitive_chain(self, concepts: list) -> dict:
"""
Chain reasoning: given [A, B, C, D...],
derive the relationship between A and the last element.
Each step binds the accumulated relationship with the next concept.
"""
if len(concepts) < 2:
return {"error": "Need at least 2 concepts"}
vecs = [
self.kernel.vectorize_tokens(c.split(), positional=False)
for c in concepts
]
# Accumulate relationship via sequential binding
accumulated = vecs[0].copy()
for i in range(1, len(vecs)):
accumulated = self._bind(accumulated, vecs[i])
# Apply position-aware permutation at each step
accumulated = np.roll(accumulated, i * 100)
candidates = self.abstraction.query_abstractions(accumulated, top_k=3)
result = {
"type": "transitivity",
"chain": concepts,
"conclusion": [(name, round(float(s), 4)) for s, name, _ in candidates],
"confidence": round(float(candidates[0][0]) if candidates else 0.0, 4),
"timestamp": time.time(),
}
self._log.append(result)
self._save_log()
return result
def report(self) -> dict:
if not self._log:
return {"status": "No reasoning performed yet"}
type_counts = {}
for entry in self._log:
t = entry.get("type", "unknown")
type_counts[t] = type_counts.get(t, 0) + 1
return {
"total_reasoning_ops": len(self._log),
"by_type": type_counts,
"recent": self._log[-3:],
}
|