File size: 21,825 Bytes
fc4c82e |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 |
# Import dependencies
from langchain.prompts import PromptTemplate
from langchain.chains import LLMChain
from pyvis.network import Network
from pprint import pprint
import networkx as nx
import gradio as gr
import re
import datasets
from huggingface_hub import login, HfApi
from datasets import Dataset, load_dataset
from rapidfuzz import fuzz, process
import math
import pandas as pd
import gspread
import torch
import json
from typing import Callable, Optional
from dataclasses import dataclass
from datasets import load_dataset
from transformers import (
AutoModelForSequenceClassification,
TrainingArguments,
Trainer,
AutoModelForCausalLM,
AutoTokenizer,
BitsAndBytesConfig,
pipeline
)
from peft import PeftModel, LoraConfig, get_peft_model, TaskType
REPO_ID_NEAR_FIELD_RAW = "milistu/AMAZON-Products-2023"
REPO_ID_NEAR_FIELD = "aslan-ng/amazon_products_2023"
REPO_ID_FAR_FIELD = "aslan-ng/amazon_products_2025"
REPO_ID_LORA_GREEN_PATENTS = "aslan-ng/lora-green-patents"
def product_quality_score(average_rating: float, rating_number: int):
"""
Bayesian Average (Amazon-style)
Args:
avg_rating: product's average rating
rating_number: number of reviews
"""
m = 1 # Minimum number of reviews required (tunable)
C = 3.5 # Global average rating (baseline)
if rating_number <= 0 or average_rating is None:
return C # fallback to global mean
return (rating_number / (rating_number + m)) * average_rating + (m / (rating_number + m)) * C
def load_near_field_raw_from_huggingface():
"""
Load the raw near-field dataset from HuggingFace.
"""
ds = datasets.load_dataset(REPO_ID_NEAR_FIELD_RAW, split="train")
print("Initial size: ", len(ds))
# Drop the extra categories
main_categories_to_remove = ["meta_Books", "meta_CDs_and_Vinyl", "meta_Digital_Music", "meta_Gift_Cards", "meta_Grocery_and_Gourmet_Food",
"meta_Magazine_Subscriptions", "meta_Software", "meta_Video_Games"]
ds = ds.filter(lambda row: row["filename"] not in main_categories_to_remove) ###
# Keep only the columns we care about
cols_to_keep = ["title", "description", "main_category", "average_rating", "rating_number"]
ds = ds.remove_columns([c for c in ds.column_names if c not in cols_to_keep])
# Add product quality score column
def add_quality_score(batch):
return {
"product_quality_score": [
product_quality_score(r, n)
for r, n in zip(batch["average_rating"], batch["rating_number"])
]
}
ds = ds.map(add_quality_score, batched=True)
# Only keep rows with valid values
def is_valid(v):
"""
Must have valid values in the row. Will be used for filtering.
"""
if v is None:
return False
if isinstance(v, str):
if v.strip() == "":
return False
return True
def keep_row(row):
"""
Keep only the columns with valid data
"""
if is_valid(row.get("title")) and \
is_valid(row.get("description")) and \
is_valid(row.get("main_category")) and \
is_valid(row.get("average_rating")) and \
is_valid(row.get("rating_number")):
return True
return False
ds = ds.filter(keep_row)
return ds.to_pandas()
def load_near_field_from_huggingface():
"""
Load the near-field dataset from HuggingFace.
"""
ds = load_dataset(REPO_ID_NEAR_FIELD, split="train")
return ds.to_pandas()
def save_near_field_to_huggingface():
"""
Save the near-field dataset from HuggingFace.
"""
df = load_near_field_raw_from_huggingface()
ds = Dataset.from_pandas(df)
ds.push_to_hub(REPO_ID_NEAR_FIELD)
print(f"✅ Pushed {len(ds)} rows to {REPO_ID_NEAR_FIELD}")
#save_near_field_to_huggingface() # Run it once
dataset_near_field = load_near_field_from_huggingface()
def load_far_field_from_sheet():
"""
Load the far-field dataset from Google Sheets.
"""
auth.authenticate_user()
from google.auth import default
COLS = ["title", "description", "average_rating", "rating_number"]
categories = ["Home & Kitchen", "Beauty & Personal Care", "Sports & Outdoors", "Clothing, Shoes & Jewelry", "Industrial & Scientific",
"Appliances", "Arts, Crafts & Sewing", "Electronics"]
sh = gspread.authorize(default()[0]).open_by_key(SHEET_ID_FAR_FIELD)
frames = []
for ws in sh.worksheets(): # iterate ALL sheets/tabs
rows = ws.get_all_records()
if not rows:
continue
df = pd.DataFrame(rows)
# Keep only the exact columns you want
df = df[COLS].copy()
# Add the tab name as main_category
df["main_category"] = ws.title
frames.append(df)
df = pd.concat(frames, ignore_index=True) if frames else pd.DataFrame(columns=COLS + ["main_category"])
# Add product quality score column
def _safe_pqs(row):
ar, n = row["average_rating"], row["rating_number"]
if pd.notna(ar) and pd.notna(n):
return product_quality_score(ar, n)
return float("nan")
df["product_quality_score"] = df.apply(_safe_pqs, axis=1)
return df
def load_far_field_from_huggingface():
"""
Load the far-field dataset from HuggingFace.
"""
ds = load_dataset(REPO_ID_FAR_FIELD, split="train")
return ds.to_pandas()
def save_far_field_to_huggingface():
"""
Save the far-field dataset from HuggingFace.
"""
df = load_far_field_from_sheet()
ds = Dataset.from_pandas(df)
ds.push_to_hub(REPO_ID_FAR_FIELD)
print(f"✅ Pushed {len(ds)} rows to {REPO_ID_FAR_FIELD}")
#save_far_field_to_huggingface() # Run it once
dataset_far_field = load_far_field_from_huggingface()
def product_score(product_quality_score: float, fuzzy_score: float):
"""
Combine product score and fuzzy score into a single score.
"""
return math.sqrt(product_quality_score * fuzzy_score)
def query_near_field(input: str, top_k: int=1):
"""
Return top_k fuzzy matches for query against dataset titles as a pandas DataFrame.
Always returns exactly top_k rows (if available).
"""
if top_k <= 0:
raise ValueError
n = len(dataset_near_field)
if top_k > n:
print(f"Warning: top_k ({top_k}) is greater than the number of examples in the near-field dataset ({n}). Returning all examples.")
return dataset_near_field.reset_index(drop=True)
matches = process.extract(
input,
dataset_near_field["title"].fillna("").astype(str).tolist(),
scorer=fuzz.token_set_ratio,
limit=n
)
rows = []
for _text, fuzzy_score, idx in matches:
row = dataset_near_field.iloc[idx].to_dict() # pandas way
row["data_source"] = "near_field"
row["fuzzy_score"] = fuzzy_score
product_quality_score = row.get("product_quality_score")
row["score"] = product_score(product_quality_score, fuzzy_score)
rows.append(row)
return (
pd.DataFrame(rows)
.sort_values("score", ascending=False)
.head(top_k)
.reset_index(drop=True)
)
def query_far_field(input: str, top_k: int):
"""
Return top_k random elements from the far_field dataset as a pandas DataFrame.
The input string is ignored.
"""
if top_k < 0:
raise ValueError
n = len(dataset_far_field)
if top_k > n:
print(f"Warning: top_k ({top_k}) is greater than the number of examples in the far-field dataset ({n}). Returning all examples.")
return dataset_far_field.reset_index(drop=True)
# Sample random rows without replacement
sampled = dataset_far_field.sample(n=top_k, random_state=None).reset_index(drop=True)
# Add the rest
sampled["fuzzy_score"] = [
fuzz.token_set_ratio(str(t) if pd.notna(t) else "", input)
for t in sampled.get("title", "")
]
product_quality_scores = sampled.get("product_quality_score")
fuzzy_scores = sampled["fuzzy_score"]
sampled["score"] = [product_score(a, b) for a, b in zip(product_quality_scores, fuzzy_scores)]
sampled["data_source"] = "far_field"
return sampled
def split_near_and_far_fields(total_examples: int, near_far_ratio: float = 0.5):
"""
Split the examples between near and far field.
The ratio represents the examples that will be in the near field to total (near + far).
"""
ratio = near_far_ratio
# Validate ratio
if ratio < 0 or ratio > 1:
raise ValueError("Ratio must be between 0 and 1")
if total_examples < 2:
raise ValueError("Total examples must be at least 2")
near_field_examples = int(total_examples * ratio)
far_field_examples = total_examples - near_field_examples
return near_field_examples, far_field_examples
def query(input: str, total_examples: int, near_far_ratio: float = 0.5):
near_field_examples, far_field_examples = split_near_and_far_fields(total_examples, near_far_ratio)
far_field_result = query_far_field(input, far_field_examples)
#print(far_field_result.head())
near_field_result = query_near_field(input, near_field_examples)
#print(near_field_result.head())
result = pd.concat([near_field_result, far_field_result], ignore_index=True)
return result
def lora_load():
model_name = "distilbert-base-uncased" # same base you trained on
tokenizer = AutoTokenizer.from_pretrained(REPO_ID_LORA_GREEN_PATENTS) # , token=token)
base_model = AutoModelForSequenceClassification.from_pretrained(model_name, num_labels=2) # , token=token)
model = PeftModel.from_pretrained(base_model, REPO_ID_LORA_GREEN_PATENTS) # , token=token)
clf = pipeline("text-classification", model=model, tokenizer=tokenizer)
return clf
clf = lora_load()
def sustainability_filter(input: str, total_examples: int, near_far_ratio: float = 0.5):
initial_products = query(input, total_examples, near_far_ratio)
initial_products_list = initial_products['description'].tolist()
filtered_products = clf(initial_products_list) # 1 for green patents, 0 otherwise
# Step 1. Extract the labels from your analysis
labels = [item["label"] for item in filtered_products]
# Step 2. Add them to the dataframe temporarily
initial_products["label"] = labels
# Step 3. Keep only rows where label != 'LABEL_0'
filtered_df = initial_products[initial_products["label"] != "LABEL_0"].copy()
# (Optional) Drop the helper column if you don’t need it anymore
filtered_df.drop(columns="label", inplace=True)
return filtered_df#sustainable_products
SYSTEM_PROMPT = """
You are a product analyst. You'll receive product description as input, and extract some product functionality and some product values. Each functionality and value should be 1-5 keywords only.
Product functionality refers to what the product does: its features, technical capabilities, and performance characteristics. It answers the question: “What can this product do?”
Product value refers to the benefit the customer gains from using the product: how it improves their life, solves their problem, or helps them achieve goals. It answers the question: “Why does this matter to the customer?”
Do **not** duplicate an item in both lists. Keep **functionalities** as concrete features. Keep **values** as clear user benefits. Keep short_title of the product shorter than 5 words.
Your Output is a dictionary. Here is the format:
# Your Input:
<product_description>
# Your Output:
{
"short_title": <short_title>,
"values": [
<value1>,
<value2>,
...
],
"functionalities": [
<function1>,
<function2>,
...
]
}
Select and return only the 5 most relevant values and only the 5 most relevant functionalities for each product.
Don't return anything out of the output format.
"""
@dataclass
class LLMConfig:
model_id: str # e.g. "Qwen/Qwen2.5-1.5B-Instruct" or "Qwen/Qwen2.5-3B-Instruct"
system_prompt: str = "" # optional system prompt
max_new_tokens: int = 256
temperature: float = 0.2
top_p: float = 0.9
repetition_penalty: float = 1.05
use_4bit: bool = True # good default for Colab VRAM
def create_llm(
*,
model_id: str,
max_new_tokens: int = 256,
temperature: float = 0.2,
top_p: float = 0.9,
repetition_penalty: float = 1.05,
use_4bit: bool = True
) -> Callable[[str], str]:
"""
Load an off-the-shelf chat LLM and return a callable llm(prompt) -> str.
Pass ONLY the model parameters you want. No size mapping. No llama_cpp.
"""
cfg = LLMConfig(
model_id=model_id,
system_prompt=SYSTEM_PROMPT,
max_new_tokens=max_new_tokens,
temperature=temperature,
top_p=top_p,
repetition_penalty=repetition_penalty,
use_4bit=use_4bit,
)
has_cuda = torch.cuda.is_available()
qconfig: Optional[BitsAndBytesConfig] = None
if has_cuda and cfg.use_4bit:
qconfig = BitsAndBytesConfig(load_in_4bit=True, bnb_4bit_compute_dtype=torch.bfloat16)
tokenizer = AutoTokenizer.from_pretrained(cfg.model_id, use_fast=True)
model = AutoModelForCausalLM.from_pretrained(
cfg.model_id,
device_map="auto",
torch_dtype=torch.bfloat16 if has_cuda else torch.float32,
quantization_config=qconfig,
).eval()
def _format_messages(user_text: str) -> str:
msgs = []
if cfg.system_prompt:
msgs.append({"role": "system", "content": cfg.system_prompt})
msgs.append({"role": "user", "content": user_text})
if hasattr(tokenizer, "apply_chat_template") and tokenizer.chat_template is not None:
return tokenizer.apply_chat_template(msgs, tokenize=False, add_generation_prompt=True)
# Fallback if no chat template is present
sys = f"System: {cfg.system_prompt}\n\n" if cfg.system_prompt else ""
return f"{sys}User: {user_text}\nAssistant:"
@torch.inference_mode()
def llm(prompt: str,
max_new_tokens: int = None,
temperature: float = None,
top_p: float = None,
repetition_penalty: float = None) -> str:
text = _format_messages(prompt)
inputs = tokenizer(text, return_tensors="pt").to(model.device)
out = model.generate(
**inputs,
max_new_tokens=max_new_tokens or cfg.max_new_tokens,
do_sample=(temperature or cfg.temperature) > 0.0,
temperature=temperature or cfg.temperature,
top_p=top_p or cfg.top_p,
repetition_penalty=repetition_penalty or cfg.repetition_penalty,
pad_token_id=tokenizer.eos_token_id,
eos_token_id=tokenizer.eos_token_id,
)
gen = out[0][inputs["input_ids"].shape[-1]:]
return tokenizer.decode(gen, skip_special_tokens=True).strip()
print(f"Loaded: {cfg.model_id} | 4-bit: {bool(qconfig)} | Device: {model.device}")
return llm
def response_to_triplets(response: str):
data = json.loads(response)
product_title = data["short_title"]
triples_list = []
for value in data.get("values", []):
triples_list.append([product_title, "HAS_VALUE", value])
for func in data.get("functionalities", []):
triples_list.append([product_title, "HAS_FUNCTIONALITY", func])
return triples_list
llm = create_llm(
model_id="Qwen/Qwen2.5-3B-Instruct",
max_new_tokens=200,
temperature=0.2,
top_p=0.9,
repetition_penalty=1.05,
use_4bit=True, # set False if you have lots of VRAM
)
def main(input: str, total_examples: int, near_far_ratio: float = 0.5):
all_triplets_list = []
sustainable_results = sustainability_filter(input, total_examples=total_examples, near_far_ratio=near_far_ratio)
#print(type(sustainable_results))
for i, product in sustainable_results.iterrows():
product_title = product["title"]
product_full_description = product_title + " " + product["description"]
#print(type(product_title))
response = llm(product_full_description)
#print(response)
triplets_list = response_to_triplets(response)
print(triplets_list)
for triplet in triplets_list:
all_triplets_list.append(triplet)
return all_triplets_list
def create_graph_from_triplets(triplets):
G = nx.DiGraph()
for triplet in triplets:
line = str(triplet).strip()
if not line:
continue
# Try comma-delimited with max 2 splits
parts = [p.strip(" ()") for p in line.split(",", 2)]
if len(parts) != 3:
# Fallback: pipe-delimited
parts = [p.strip(" ()") for p in line.split("|")]
if len(parts) != 3:
continue # malformed, skip
subject, predicate, obj = parts
if subject and predicate and obj:
G.add_edge(subject, obj, label=predicate)
return G
def nx_to_pyvis(networkx_graph):
from pyvis.network import Network
pyvis_graph = Network(notebook=True, cdn_resources="remote")
# 🎨 Color scheme
color_product_bg = "#FFAA00" # product nodes
color_value_bg = "#4CAF50" # value nodes
color_func_bg = "#2196F3" # functionality nodes
color_default_bg = "#CCCCCC"
color_value_edge = "#81C784"
color_func_edge = "#64B5F6"
color_default_edge = "#999999"
# Accept both correct and misspelled labels
FUNC_LABELS = {"HAS_FUNCTIONALITY", "HAS_FUNTIONALITY"}
VALUE_LABELS = {"HAS_VALUE"}
# 🧭 Collect roles from edges
subjects = set()
value_nodes = set()
func_nodes = set()
for u, v, data in networkx_graph.edges(data=True):
lbl = str(data.get("label", "")).strip()
subjects.add(u)
if lbl in VALUE_LABELS:
value_nodes.add(v)
elif lbl in FUNC_LABELS:
func_nodes.add(v)
# 🧩 Add nodes with role-based styling FIRST
for node in networkx_graph.nodes():
if node in value_nodes:
bg = color_value_bg
shape = "ellipse"
elif node in func_nodes:
bg = color_func_bg
shape = "diamond"
elif node in subjects:
bg = color_product_bg
shape = "box"
else:
bg = color_default_bg
shape = "dot"
pyvis_graph.add_node(
node,
label=str(node),
color={"background": bg, "border": "#333333"},
shape=shape
)
# 🔗 Add edges with colors AFTER nodes
for u, v, data in networkx_graph.edges(data=True):
lbl = str(data.get("label", "")).strip()
if lbl in VALUE_LABELS:
edge_color = color_value_edge
elif lbl in FUNC_LABELS:
edge_color = color_func_edge
else:
edge_color = color_default_edge
pyvis_graph.add_edge(u, v, label=lbl, title=lbl, color=edge_color)
return pyvis_graph
def generateGraph(triples_list):
triplets = [t.strip() for t in triples_list if t.strip()]
graph = create_graph_from_triplets(triplets)
pyvis_network = nx_to_pyvis(graph)
pyvis_network.toggle_hide_edges_on_drag(True)
pyvis_network.toggle_physics(False)
pyvis_network.set_edge_smooth('discrete')
html = pyvis_network.generate_html()
html = html.replace("'", "\"")
return f"""<iframe style="width: 100%; height: 600px;margin:0 auto" name="result" allow="midi; geolocation; microphone; camera;
display-capture; encrypted-media;" sandbox="allow-modals allow-forms
allow-scripts allow-same-origin allow-popups
allow-top-navigation-by-user-activation allow-downloads" allowfullscreen=""
allowpaymentrequest="" frameborder="0" srcdoc='{html}'></iframe>"""
def pipeline_fn(user_text: str, total_examples: int, near_far_ratio: float):
try:
if not user_text.strip():
return "<div style='padding:12px;color:#b00;'>Please enter some text.</div>"
# ✅ Call main with both parameters
triples = main(
user_text,
total_examples=total_examples,
near_far_ratio=near_far_ratio
) or []
# Convert triplets into readable strings
triples_list = []
for t in triples:
if isinstance(t, (tuple, list)) and len(t) == 3:
triples_list.append(f"{t[0]}, {t[1]}, {t[2]}")
else:
triples_list.append(str(t))
return generateGraph(triples_list)
except Exception:
import traceback
return "<pre style='white-space: pre-wrap; font-size:12px; color:#b00;'>" + traceback.format_exc() + "</pre>"
demo = gr.Interface(
fn=pipeline_fn,
inputs=[
gr.Textbox(label="Enter your query / text", value="", lines=6),
gr.Number(label="Number of examples", value=6, precision=0),
gr.Slider(
minimum=0.0,
maximum=1.0,
value=0.8,
step=0.05,
label="Near/Far Ratio"
),
],
outputs=gr.HTML(),
title="Knowledge Graph",
allow_flagging="never",
live=False,
css="""
#component-0, #component-1, #component-2, #component-3, #component-4 {
display: flex;
justify-content: center;
align-items: center;
flex-direction: column;
}
.gradio-container {
justify-content: center !important;
align-items: center !important;
text-align: center;
}
textarea, iframe {
margin: 0 auto;
display: block;
}
"""
)
if __name__ == "__main__":
demo.launch(quiet=True) |