Spaces:
Sleeping
Sleeping
File size: 21,176 Bytes
4b5c25d | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 | """
LITVISION Recommendation Engine
================================
Complete recommendation pipeline preserving every component from
the original Jupyter Notebook:
β’ Zero-shot genre classification (joeddav/xlm-roberta-large-xnli)
β’ SentenceTransformer embeddings (paraphrase-multilingual-MiniLM-L12-v2)
β’ FAISS IndexFlatIP similarity search
β’ Event-weighted user vector construction
β’ Proportional genre-balanced feed allocation
β’ Cosine-similarity ranking pipeline
NO logic has been simplified, removed, or replaced.
"""
import os
import logging
from datetime import datetime, timedelta
from typing import Dict, List, Optional, Set, Tuple
import numpy as np
import pandas as pd
import torch
import faiss
from transformers import pipeline as hf_pipeline
from sentence_transformers import SentenceTransformer
from utils import get_device, safe_cuda_empty_cache
logger = logging.getLogger("litvision.recommendation")
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# Constants β identical to the notebook
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
GENRES: List[str] = [
"Fantasy", "Romance", "Mystery", "Science Fiction", "Self-Help",
"History", "Business", "Children", "Horror", "Poetry",
]
EVENT_W: Dict[str, float] = {"view": 1.0, "like": 3.0}
TEMPLATES: Dict[str, List[str]] = {
"Fantasy": [
"A young hero discovers a hidden kingdom and must defeat a dark sorcerer.",
"Dragons rise again as an ancient prophecy awakens in the north.",
],
"Romance": [
"Two strangers meet in a small cafe and find love against all odds.",
"A long-distance relationship is tested by secrets and time.",
],
"Mystery": [
"A detective investigates a series of murders in a quiet town.",
"A missing diary reveals clues to an old family crime.",
],
"Science Fiction": [
"A crew travels through a wormhole to save humanity from collapse.",
"An AI gains consciousness and changes the future of Earth.",
],
"Self-Help": [
"A practical guide to build habits and improve focus every day.",
"Learn to manage anxiety with simple routines and mindset shifts.",
],
"History": [
"An account of ancient empires and the wars that shaped the world.",
"A deep dive into the political revolutions of the 20th century.",
],
"Business": [
"How startups scale products and build strong teams.",
"Negotiation tactics and leadership strategies for managers.",
],
"Children": [
"A curious cat explores the city and learns about friendship.",
"A magical school adventure for kids with puzzles and fun.",
],
"Horror": [
"A haunted house whispers at night, luring visitors inside.",
"A village faces a terrifying creature in the woods.",
],
"Poetry": [
"A collection of poems about love, loss, and hope.",
"Minimalist poems inspired by nature and silence.",
],
}
SAMPLE_DATA_DIR = os.path.join(os.path.dirname(__file__), "sample_data")
BOOKS_CSV_PATH = os.path.join(SAMPLE_DATA_DIR, "books.csv")
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# Recommendation Engine
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
class RecommendationEngine:
"""
Production wrapper around the full notebook recommendation pipeline.
Models are loaded lazily on first call or explicitly via ``load_models()``.
"""
def __init__(self) -> None:
self.device: str = "cpu"
self.embed_model: Optional[SentenceTransformer] = None
self.zero_shot = None
self.books_df: Optional[pd.DataFrame] = None
self.book_embeddings: Optional[np.ndarray] = None
self.faiss_index: Optional[faiss.IndexFlatIP] = None
self.bookid_to_idx: Dict[int, int] = {}
self._loaded = False
# Per-user feed state (identical to notebook)
self.user_feed_state: Dict[int, Set[int]] = {}
# ------------------------------------------------------------------
# Model loading
# ------------------------------------------------------------------
def load_models(self) -> None:
"""Load all AI models and build the FAISS index."""
if self._loaded:
return
self.device = get_device()
logger.info("Loading recommendation models β¦")
# 1. Zero-shot classifier β identical model to notebook
logger.info("Loading zero-shot classifier: joeddav/xlm-roberta-large-xnli")
zs_device = 0 if self.device == "cuda" else -1
self.zero_shot = hf_pipeline(
"zero-shot-classification",
model="joeddav/xlm-roberta-large-xnli",
device=zs_device,
)
# FP16 on CUDA for the zero-shot model
if self.device == "cuda":
try:
self.zero_shot.model.half()
logger.info("Zero-shot model converted to FP16")
except Exception as e:
logger.warning(f"Could not convert zero-shot to FP16: {e}")
# 2. SentenceTransformer β identical model to notebook
logger.info(
"Loading SentenceTransformer: "
"sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2"
)
self.embed_model = SentenceTransformer(
"sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2",
device=self.device,
)
if self.device == "cuda":
try:
self.embed_model.half()
logger.info("SentenceTransformer converted to FP16")
except Exception as e:
logger.warning(f"Could not convert embed model to FP16: {e}")
# 3. Load / generate books dataset
self.books_df = self._load_books()
# 4. Classify genres (zero-shot) β identical to notebook
self._classify_genres()
# 5. Build embeddings + FAISS index β identical to notebook
self._build_index()
self._loaded = True
logger.info(
f"Recommendation engine ready β "
f"{len(self.books_df)} books, FAISS dim={self.faiss_index.d}"
)
# ------------------------------------------------------------------
# Books dataset
# ------------------------------------------------------------------
def _load_books(self) -> pd.DataFrame:
"""Load books from CSV or generate the default sample set."""
if os.path.exists(BOOKS_CSV_PATH):
logger.info(f"Loading books from {BOOKS_CSV_PATH}")
df = pd.read_csv(BOOKS_CSV_PATH)
required = {"book_id", "title", "author", "description"}
if not required.issubset(set(df.columns)):
raise ValueError(
f"books.csv must contain columns {required}, "
f"found {set(df.columns)}"
)
return df
logger.info("No books.csv found β generating sample data (seed=42)")
df = self._make_books(200)
os.makedirs(SAMPLE_DATA_DIR, exist_ok=True)
df.to_csv(BOOKS_CSV_PATH, index=False)
logger.info(f"Saved {len(df)} books to {BOOKS_CSV_PATH}")
return df
@staticmethod
def _make_books(n_books: int = 200) -> pd.DataFrame:
"""Generate sample books β identical to notebook ``make_books``."""
np.random.seed(42)
rows = []
for i in range(1, n_books + 1):
g = np.random.choice(GENRES)
desc = np.random.choice(TEMPLATES[g])
title = f"{g} Book {i}"
author = f"Author {np.random.randint(1, 60)}"
rows.append([i, title, author, desc])
return pd.DataFrame(rows, columns=["book_id", "title", "author", "description"])
# ------------------------------------------------------------------
# Genre classification β identical to notebook
# ------------------------------------------------------------------
def classify_genre(self, text: str) -> Tuple[str, float]:
"""Zero-shot genre classification β identical to notebook."""
out = self.zero_shot(text, candidate_labels=GENRES, multi_label=False)
return out["labels"][0], float(out["scores"][0])
def _classify_genres(self) -> None:
"""Classify all books in the dataset β identical to notebook loop."""
if "genre" in self.books_df.columns and "genre_confidence" in self.books_df.columns:
logger.info("Genre columns already present β skipping classification")
return
logger.info("Classifying genres for all books β¦")
genres, scores = [], []
texts = (self.books_df["title"] + " | " + self.books_df["description"]).tolist()
for i, txt in enumerate(texts):
g, s = self.classify_genre(txt)
genres.append(g)
scores.append(s)
if (i + 1) % 50 == 0:
logger.info(f" classified {i + 1}/{len(texts)} books")
self.books_df["genre"] = genres
self.books_df["genre_confidence"] = scores
# Persist updated CSV
self.books_df.to_csv(BOOKS_CSV_PATH, index=False)
logger.info("Genre classification complete")
# ------------------------------------------------------------------
# Embeddings + FAISS β identical to notebook
# ------------------------------------------------------------------
def _build_index(self) -> None:
"""Build SentenceTransformer embeddings and FAISS index."""
self.books_df["text"] = (
"Title: " + self.books_df["title"]
+ " | Author: " + self.books_df["author"]
+ " | Genre: " + self.books_df["genre"]
+ " | Description: " + self.books_df["description"]
)
logger.info("Encoding book embeddings β¦")
self.book_embeddings = self.embed_model.encode(
self.books_df["text"].tolist(),
batch_size=64,
show_progress_bar=True,
convert_to_numpy=True,
normalize_embeddings=True,
).astype("float32")
dim = self.book_embeddings.shape[1]
self.faiss_index = faiss.IndexFlatIP(dim) # cosine (normalized)
self.faiss_index.add(self.book_embeddings)
self.bookid_to_idx = {
int(bid): i
for i, bid in enumerate(self.books_df["book_id"].tolist())
}
logger.info(
f"FAISS index built β embeddings {self.book_embeddings.shape}, "
f"ntotal={self.faiss_index.ntotal}"
)
# ------------------------------------------------------------------
# User interest ratios β identical to notebook
# ------------------------------------------------------------------
def user_interest_ratios(
self,
user_id: int,
interactions_df: pd.DataFrame,
) -> Dict[str, float]:
"""Compute weighted genre interest ratios β identical to notebook."""
u = interactions_df[interactions_df.user_id == user_id].merge(
self.books_df[["book_id", "genre"]], on="book_id", how="left"
)
if u.empty:
return {g: 1 / len(GENRES) for g in GENRES}
u["w"] = u["event_type"].map(EVENT_W).fillna(0.0)
s = u.groupby("genre")["w"].sum().reindex(GENRES, fill_value=0.0)
total = s.sum()
if total == 0:
return {g: 1 / len(GENRES) for g in GENRES}
return (s / total).to_dict()
# ------------------------------------------------------------------
# User vector β identical to notebook
# ------------------------------------------------------------------
def build_user_vector(
self,
user_id: int,
interactions_df: pd.DataFrame,
) -> Tuple[Optional[np.ndarray], Set[int]]:
"""Build weighted user embedding vector β identical to notebook."""
u = interactions_df[interactions_df.user_id == user_id]
if u.empty:
return None, set()
vecs: List[np.ndarray] = []
weights: List[float] = []
seen: Set[int] = set()
for _, row in u.iterrows():
bid = int(row["book_id"])
ev = row["event_type"]
if bid not in self.bookid_to_idx:
continue
w = EVENT_W.get(ev, 0.0)
if w == 0:
continue
vecs.append(self.book_embeddings[self.bookid_to_idx[bid]])
weights.append(w)
seen.add(bid)
if not vecs:
return None, seen
vecs_arr = np.array(vecs)
weights_arr = np.array(weights).reshape(-1, 1)
user_vec = np.sum(vecs_arr * weights_arr, axis=0) / (
np.sum(np.abs(weights_arr)) + 1e-9
)
user_vec = user_vec / (np.linalg.norm(user_vec) + 1e-9)
return user_vec.astype("float32"), seen
# ------------------------------------------------------------------
# Feed allocation β identical to notebook
# ------------------------------------------------------------------
@staticmethod
def allocate_feed(
ratios: Dict[str, float],
unseen_counts: Dict[str, int],
feed_size: int = 50,
) -> Dict[str, int]:
"""Proportional genre allocation β identical to notebook."""
alloc = {g: 0 for g in GENRES}
remaining = feed_size
# Target counts proportional to ratios
target = {
g: int(round(ratios.get(g, 0.0) * feed_size)) for g in GENRES
}
# Cap by availability
for g in GENRES:
alloc[g] = min(target[g], unseen_counts.get(g, 0))
remaining -= alloc[g]
# Distribute leftovers to best-ratio genres that still have items
while remaining > 0:
candidates = [
g for g in GENRES if alloc[g] < unseen_counts.get(g, 0)
]
if not candidates:
break
g = max(candidates, key=lambda x: ratios.get(x, 0.0))
alloc[g] += 1
remaining -= 1
return alloc
# ------------------------------------------------------------------
# Mixed feed builder β identical to notebook
# ------------------------------------------------------------------
def build_mixed_feed(
self,
user_id: int,
interactions_df: pd.DataFrame,
feed_size: int = 50,
random_state: int = 42,
) -> pd.DataFrame:
"""Build a genre-balanced, similarity-ranked feed β identical to notebook."""
ratios = self.user_interest_ratios(user_id, interactions_df)
seen = set(
interactions_df.loc[
interactions_df.user_id == user_id, "book_id"
]
.astype(int)
.tolist()
)
unseen_df = self.books_df[~self.books_df.book_id.isin(seen)].copy()
unseen_counts = (
unseen_df.groupby("genre")["book_id"]
.count()
.reindex(GENRES, fill_value=0)
.to_dict()
)
alloc = self.allocate_feed(ratios, unseen_counts, feed_size=feed_size)
parts: List[pd.DataFrame] = []
for g, k in alloc.items():
if k <= 0:
continue
g_df = unseen_df[unseen_df.genre == g]
if len(g_df) == 0:
continue
parts.append(
g_df.sample(n=min(k, len(g_df)), random_state=random_state)
)
if not parts:
return self.books_df.sample(feed_size, random_state=random_state)[
["book_id", "title", "author", "genre"]
]
feed = pd.concat(parts, ignore_index=True)
# Shuffle / ranking β identical to notebook
feed = feed.sample(frac=1.0, random_state=random_state).reset_index(
drop=True
)
user_vec, _ = self.build_user_vector(user_id, interactions_df)
if user_vec is not None:
idxs = [
self.bookid_to_idx[int(b)] for b in feed["book_id"].tolist()
]
feed_vecs = self.book_embeddings[idxs]
feed["score"] = (feed_vecs @ user_vec).astype(float)
feed = feed.sort_values("score", ascending=False).reset_index(
drop=True
)
cols = ["book_id", "title", "author", "genre"]
if "score" in feed.columns:
cols.append("score")
return feed[cols]
# ------------------------------------------------------------------
# Paginated feed β identical to notebook
# ------------------------------------------------------------------
def get_next_feed_page(
self,
user_id: int,
interactions_df: pd.DataFrame,
page_size: int = 20,
) -> pd.DataFrame:
"""Return the next page of unseen recommendations β identical to notebook."""
shown = self.user_feed_state.get(user_id, set())
# Add temporary "view" interactions for already-shown books
if len(shown) > 0:
temp_rows = pd.DataFrame(
{
"user_id": [user_id] * len(shown),
"book_id": list(shown),
"event_type": ["view"] * len(shown),
"timestamp": [datetime.now().isoformat()] * len(shown),
}
)
temp_interactions = pd.concat(
[interactions_df, temp_rows], ignore_index=True
)
else:
temp_interactions = interactions_df
page = self.build_mixed_feed(
user_id,
temp_interactions,
feed_size=page_size,
random_state=np.random.randint(0, 10_000),
)
# Update shown state
self.user_feed_state[user_id] = shown.union(
set(page["book_id"].astype(int).tolist())
)
return page
def reset_user_feed(self, user_id: int) -> None:
"""Clear pagination state for a user."""
self.user_feed_state.pop(user_id, None)
# ------------------------------------------------------------------
# Generate interactions from request payload
# ------------------------------------------------------------------
@staticmethod
def build_interactions_df(
user_id: int,
interactions: Optional[List[dict]] = None,
viewed_books: Optional[List[int]] = None,
favorite_genres: Optional[List[str]] = None,
) -> pd.DataFrame:
"""
Construct a pandas DataFrame of user interactions from the API
request payload. This merges explicit interaction events,
viewed-book IDs (as implicit views), and favourite genres
(synthesised as likes for books in those genres).
"""
rows: List[dict] = []
now_iso = datetime.now().isoformat()
# 1. Explicit interactions
if interactions:
for inter in interactions:
rows.append(
{
"user_id": user_id,
"book_id": int(inter["book_id"]),
"event_type": inter.get("event_type", "view"),
"timestamp": inter.get("timestamp", now_iso),
}
)
# 2. Viewed books β implicit "view" events
if viewed_books:
for bid in viewed_books:
rows.append(
{
"user_id": user_id,
"book_id": int(bid),
"event_type": "view",
"timestamp": now_iso,
}
)
if not rows:
return pd.DataFrame(
columns=["user_id", "book_id", "event_type", "timestamp"]
)
return pd.DataFrame(rows)
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# Module-level singleton
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
engine = RecommendationEngine()
|