File size: 20,707 Bytes
25d2eb7 2827b8a 7a1cd7a f5eb405 2827b8a f5eb405 2827b8a f5eb405 2827b8a f5eb405 2827b8a 892ceeb 2827b8a f5eb405 2827b8a f5eb405 2827b8a f5eb405 2827b8a f5eb405 2827b8a 892ceeb f5eb405 2827b8a f5eb405 2827b8a 25d2eb7 7a1cd7a 892ceeb 7a1cd7a 2827b8a 7a1cd7a 7ed3881 892ceeb 2827b8a f5eb405 25d2eb7 2827b8a 7a1cd7a 7ed3881 7a1cd7a 7ed3881 7a1cd7a 7ed3881 7a1cd7a 7ed3881 7a1cd7a 7ed3881 7a1cd7a 2827b8a 7a1cd7a 2827b8a 7a1cd7a 7ed3881 2827b8a 7ed3881 7a1cd7a 7ed3881 2827b8a 7a1cd7a 2827b8a 7a1cd7a 2827b8a f5eb405 |
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 |
import gradio as gr
from datasets import load_dataset
import numpy as np
from model2vec import StaticModel
from reach import Reach
from difflib import ndiff
import sys
import tqdm
# Load the model at startup
model = StaticModel.from_pretrained("minishlab/M2V_base_output")
# Load the default datasets at startup
default_dataset1_name = "ag_news"
default_dataset1_split = "train"
default_dataset2_name = "ag_news"
default_dataset2_split = "test"
ds_default1 = load_dataset(default_dataset1_name, split=default_dataset1_split)
ds_default2 = load_dataset(default_dataset2_name, split=default_dataset2_split)
def deduplicate(embedding_matrix: np.ndarray, threshold: float, batch_size: int = 1024, progress=None) -> tuple[np.ndarray, dict[int, int]]:
"""
Deduplicate embeddings and return the deduplicated indices and a mapping of removed indices to their corresponding original indices.
"""
reach = Reach(vectors=embedding_matrix, items=[str(i) for i in range(len(embedding_matrix))])
deduplicated_indices = set(range(len(embedding_matrix)))
duplicate_to_original_mapping = {}
results = reach.nearest_neighbor_threshold(
embedding_matrix,
threshold=threshold,
batch_size=batch_size,
show_progressbar=False # Disable internal progress bar
)
# Process duplicates
for i, similar_items in enumerate(progress.tqdm(results, desc="Processing duplicates")):
if i not in deduplicated_indices:
continue
similar_indices = [int(item[0]) for item in similar_items if int(item[0]) != i]
for sim_idx in similar_indices:
if sim_idx in deduplicated_indices:
deduplicated_indices.remove(sim_idx)
duplicate_to_original_mapping[sim_idx] = i
return np.array(list(deduplicated_indices)), duplicate_to_original_mapping
def deduplicate_across_datasets(embedding_matrix_1: np.ndarray, embedding_matrix_2: np.ndarray, threshold: float, batch_size: int = 1024, progress=None) -> tuple[list[int], dict[int, int]]:
"""
Deduplicate embeddings across two datasets and return the indices of duplicates between them.
"""
reach = Reach(vectors=embedding_matrix_1, items=[str(i) for i in range(len(embedding_matrix_1))])
duplicate_indices_in_test = []
duplicate_to_original_mapping = {}
results = reach.nearest_neighbor_threshold(
embedding_matrix_2,
threshold=threshold,
batch_size=batch_size,
show_progressbar=False # Disable internal progress bar
)
# Process duplicates
for i, similar_items in enumerate(progress.tqdm(results, desc="Processing duplicates across datasets")):
similar_indices = [int(item[0]) for item in similar_items if item[1] >= threshold]
if similar_indices:
duplicate_indices_in_test.append(i)
duplicate_to_original_mapping[i] = similar_indices[0]
return duplicate_indices_in_test, duplicate_to_original_mapping
def display_word_differences(x: str, y: str) -> str:
diff = ndiff(x.split(), y.split())
return " ".join([word for word in diff if word.startswith(('+', '-'))])
def perform_deduplication(
deduplication_type,
dataset1_name,
dataset1_split,
dataset1_text_column,
dataset2_name="",
dataset2_split="",
dataset2_text_column="",
threshold=0.8,
progress=gr.Progress(track_tqdm=True)
):
# Monkey-patch tqdm
original_tqdm = tqdm.tqdm
tqdm.tqdm = progress.tqdm
sys.modules['tqdm'].tqdm = progress.tqdm
sys.modules['tqdm.auto'].tqdm = progress.tqdm
try:
# Convert threshold to float
threshold = float(threshold)
if deduplication_type == "Single dataset":
# Check if the dataset is the default one
if dataset1_name == default_dataset1_name and dataset1_split == default_dataset1_split:
ds = ds_default1
else:
ds = load_dataset(dataset1_name, split=dataset1_split)
# Extract texts
texts = [example[dataset1_text_column] for example in ds]
# Compute embeddings
embedding_matrix = model.encode(texts, show_progressbar=False) # Disable internal progress bar
# Show progress bar for embedding computation
embedding_matrix = progress.tqdm(embedding_matrix, desc="Computing embeddings")
# Deduplicate
deduplicated_indices, duplicate_to_original_mapping = deduplicate(embedding_matrix, threshold, progress=progress)
# Prepare the results
num_duplicates = len(duplicate_to_original_mapping)
num_total = len(texts)
num_deduplicated = len(deduplicated_indices)
result_text = f"**Total documents:** {num_total}\n"
result_text += f"**Number of duplicates found:** {num_duplicates}\n"
result_text += f"**Number of unique documents after deduplication:** {num_deduplicated}\n\n"
# Show deduplicated examples
result_text += "**Examples of duplicates found:**\n\n"
num_examples = min(5, num_duplicates)
for duplicate_idx, original_idx in list(duplicate_to_original_mapping.items())[:num_examples]:
original_text = texts[original_idx]
duplicate_text = texts[duplicate_idx]
differences = display_word_differences(original_text, duplicate_text)
result_text += f"**Original text:**\n{original_text}\n\n"
result_text += f"**Duplicate text:**\n{duplicate_text}\n\n"
result_text += f"**Differences:**\n{differences}\n"
result_text += "-" * 50 + "\n\n"
return result_text
elif deduplication_type == "Cross-dataset":
# Dataset 1
if dataset1_name == default_dataset1_name and dataset1_split == default_dataset1_split:
ds1 = ds_default1
else:
ds1 = load_dataset(dataset1_name, split=dataset1_split)
# Dataset 2
if dataset2_name == default_dataset2_name and dataset2_split == default_dataset2_split:
ds2 = ds_default2
else:
ds2 = load_dataset(dataset2_name, split=dataset2_split)
# Extract texts
texts1 = [example[dataset1_text_column] for example in ds1]
texts2 = [example[dataset2_text_column] for example in ds2]
# Compute embeddings
embedding_matrix1 = model.encode(texts1, show_progressbar=False) # Disable internal progress bar
embedding_matrix2 = model.encode(texts2, show_progressbar=False) # Disable internal progress bar
# Show progress bar for embedding computation
embedding_matrix1 = progress.tqdm(embedding_matrix1, desc="Computing embeddings for Dataset 1")
embedding_matrix2 = progress.tqdm(embedding_matrix2, desc="Computing embeddings for Dataset 2")
# Deduplicate across datasets
duplicate_indices_in_ds2, duplicate_to_original_mapping = deduplicate_across_datasets(embedding_matrix1, embedding_matrix2, threshold, progress=progress)
num_duplicates = len(duplicate_indices_in_ds2)
num_total_ds2 = len(texts2)
num_unique_ds2 = num_total_ds2 - num_duplicates
result_text = f"**Total documents in {dataset2_name}/{dataset2_split}:** {num_total_ds2}\n"
result_text += f"**Number of duplicates found in {dataset2_name}/{dataset2_split}:** {num_duplicates}\n"
result_text += f"**Number of unique documents in {dataset2_name}/{dataset2_split} after deduplication:** {num_unique_ds2}\n\n"
# Show deduplicated examples
result_text += "**Examples of duplicates found in Dataset 2:**\n\n"
num_examples = min(5, num_duplicates)
for duplicate_idx in duplicate_indices_in_ds2[:num_examples]:
original_idx = duplicate_to_original_mapping[duplicate_idx]
original_text = texts1[original_idx]
duplicate_text = texts2[duplicate_idx]
differences = display_word_differences(original_text, duplicate_text)
result_text += f"**Original text (Dataset 1):**\n{original_text}\n\n"
result_text += f"**Duplicate text (Dataset 2):**\n{duplicate_text}\n\n"
result_text += f"**Differences:**\n{differences}\n"
result_text += "-" * 50 + "\n\n"
return result_text
finally:
# Restore original tqdm
tqdm.tqdm = original_tqdm
sys.modules['tqdm'].tqdm = original_tqdm
sys.modules['tqdm.auto'].tqdm = original_tqdm
with gr.Blocks() as demo:
gr.Markdown("# Semantic Deduplication")
deduplication_type = gr.Radio(
choices=["Single dataset", "Cross-dataset"],
label="Deduplication Type",
value="Single dataset"
)
with gr.Row():
dataset1_name = gr.Textbox(value="ag_news", label="Dataset 1 Name")
dataset1_split = gr.Textbox(value="train", label="Dataset 1 Split")
dataset1_text_column = gr.Textbox(value="text", label="Text Column Name")
dataset2_inputs = gr.Column(visible=False)
with dataset2_inputs:
gr.Markdown("### Dataset 2")
with gr.Row():
dataset2_name = gr.Textbox(value="ag_news", label="Dataset 2 Name")
dataset2_split = gr.Textbox(value="test", label="Dataset 2 Split")
dataset2_text_column = gr.Textbox(value="text", label="Text Column Name")
threshold = gr.Slider(
minimum=0.0,
maximum=1.0,
value=0.8,
label="Similarity Threshold"
)
compute_button = gr.Button("Compute")
output = gr.Markdown()
# Function to update the visibility of dataset2_inputs
def update_visibility(deduplication_type_value):
if deduplication_type_value == "Cross-dataset":
return gr.update(visible=True)
else:
return gr.update(visible=False)
deduplication_type.change(
update_visibility,
inputs=deduplication_type,
outputs=dataset2_inputs
)
compute_button.click(
fn=perform_deduplication,
inputs=[
deduplication_type,
dataset1_name,
dataset1_split,
dataset1_text_column,
dataset2_name,
dataset2_split,
dataset2_text_column,
threshold
],
outputs=output
)
demo.launch()
# import gradio as gr
# from datasets import load_dataset
# import numpy as np
# from model2vec import StaticModel
# from reach import Reach
# from difflib import ndiff
# def deduplicate(embedding_matrix: np.ndarray, threshold: float, batch_size: int = 1024, progress=gr.Progress(track_tqdm=True)) -> tuple[np.ndarray, dict[int, int]]:
# """
# Deduplicate embeddings and return the deduplicated indices and a mapping of removed indices to their corresponding original indices.
# """
# reach = Reach(vectors=embedding_matrix, items=[str(i) for i in range(len(embedding_matrix))])
# # Use a set for deduplicated indices and keep track of duplicates
# deduplicated_indices = set(range(len(embedding_matrix))) # Start with all indices as deduplicated
# duplicate_to_original_mapping = {}
# results = reach.nearest_neighbor_threshold(
# embedding_matrix,
# threshold=threshold,
# batch_size=batch_size,
# show_progressbar=True
# )
# # Process duplicates
# for i, similar_items in enumerate(progress.tqdm(results, desc="Processing duplicates")):
# if i not in deduplicated_indices:
# continue # Skip already marked duplicates
# # Similar items are returned as (index, score), we are only interested in the index
# similar_indices = [int(item[0]) for item in similar_items if int(item[0]) != i]
# # Mark similar documents as duplicates and map them to the original
# for sim_idx in similar_indices:
# if sim_idx in deduplicated_indices:
# deduplicated_indices.remove(sim_idx)
# duplicate_to_original_mapping[sim_idx] = i # Map duplicate to original
# return np.array(list(deduplicated_indices)), duplicate_to_original_mapping
# def deduplicate_across_datasets(embedding_matrix_1: np.ndarray, embedding_matrix_2: np.ndarray, threshold: float, batch_size: int = 1024, progress=gr.Progress(track_tqdm=True)) -> tuple[list[int], dict[int, int]]:
# """
# Deduplicate embeddings across two datasets and return the indices of duplicates between them.
# """
# reach = Reach(vectors=embedding_matrix_1, items=[str(i) for i in range(len(embedding_matrix_1))])
# # Keep track of duplicates in the second dataset
# duplicate_indices_in_test = []
# duplicate_to_original_mapping = {}
# # Find nearest neighbors from the test set in the train set
# results = reach.nearest_neighbor_threshold(
# embedding_matrix_2,
# threshold=threshold,
# batch_size=batch_size,
# show_progressbar=True
# )
# # Process duplicates
# for i, similar_items in enumerate(progress.tqdm(results, desc="Processing duplicates across datasets")):
# # Similar items are returned as (index, score), we are only interested in the index
# similar_indices = [int(item[0]) for item in similar_items if item[1] >= threshold] # Keep those above the threshold
# # If we find a similar item in the train set, mark it as a duplicate
# if similar_indices:
# duplicate_indices_in_test.append(i)
# duplicate_to_original_mapping[i] = similar_indices[0] # Map duplicate in test to original in train
# return duplicate_indices_in_test, duplicate_to_original_mapping
# def display_word_differences(x: str, y: str) -> str:
# diff = ndiff(x.split(), y.split())
# return " ".join([word for word in diff if word.startswith(('+', '-'))])
# def perform_deduplication(
# deduplication_type,
# dataset1_name,
# dataset1_split,
# dataset1_text_column,
# dataset2_name="",
# dataset2_split="",
# dataset2_text_column="",
# threshold=0.8,
# progress=gr.Progress(track_tqdm=True)
# ):
# # Convert threshold to float
# threshold = float(threshold)
# if deduplication_type == "Single dataset":
# # Load the dataset
# ds = load_dataset(dataset1_name, split=dataset1_split)
# # Extract texts
# texts = [example[dataset1_text_column] for example in ds]
# # Compute embeddings
# model = StaticModel.from_pretrained("minishlab/M2V_base_output")
# embedding_matrix = model.encode(texts, show_progressbar=True)
# # Deduplicate
# deduplicated_indices, duplicate_to_original_mapping = deduplicate(embedding_matrix, threshold, progress=progress)
# # Prepare the results
# num_duplicates = len(duplicate_to_original_mapping)
# num_total = len(texts)
# num_deduplicated = len(deduplicated_indices)
# result_text = f"**Total documents:** {num_total}\n"
# result_text += f"**Number of duplicates found:** {num_duplicates}\n"
# result_text += f"**Number of unique documents after deduplication:** {num_deduplicated}\n\n"
# # Show deduplicated examples
# result_text += "**Examples of duplicates found:**\n\n"
# num_examples = min(5, num_duplicates)
# for duplicate_idx, original_idx in list(duplicate_to_original_mapping.items())[:num_examples]:
# original_text = texts[original_idx]
# duplicate_text = texts[duplicate_idx]
# differences = display_word_differences(original_text, duplicate_text)
# result_text += f"**Original text:**\n{original_text}\n\n"
# result_text += f"**Duplicate text:**\n{duplicate_text}\n\n"
# result_text += f"**Differences:**\n{differences}\n"
# result_text += "-" * 50 + "\n\n"
# return result_text
# elif deduplication_type == "Cross-dataset":
# # Load datasets
# ds1 = load_dataset(dataset1_name, split=dataset1_split)
# ds2 = load_dataset(dataset2_name, split=dataset2_split)
# # Extract texts
# texts1 = [example[dataset1_text_column] for example in ds1]
# texts2 = [example[dataset2_text_column] for example in ds2]
# # Compute embeddings
# model = StaticModel.from_pretrained("minishlab/M2V_base_output")
# embedding_matrix1 = model.encode(texts1, show_progressbar=True)
# embedding_matrix2 = model.encode(texts2, show_progressbar=True)
# # Deduplicate across datasets
# duplicate_indices_in_ds2, duplicate_to_original_mapping = deduplicate_across_datasets(embedding_matrix1, embedding_matrix2, threshold, progress=progress)
# num_duplicates = len(duplicate_indices_in_ds2)
# num_total_ds2 = len(texts2)
# num_unique_ds2 = num_total_ds2 - num_duplicates
# result_text = f"**Total documents in {dataset2_name}/{dataset2_split}:** {num_total_ds2}\n"
# result_text += f"**Number of duplicates found in {dataset2_name}/{dataset2_split}:** {num_duplicates}\n"
# result_text += f"**Number of unique documents in {dataset2_name}/{dataset2_split} after deduplication:** {num_unique_ds2}\n\n"
# # Show deduplicated examples
# result_text += "**Examples of duplicates found in Dataset 2:**\n\n"
# num_examples = min(5, num_duplicates)
# for duplicate_idx in duplicate_indices_in_ds2[:num_examples]:
# original_idx = duplicate_to_original_mapping[duplicate_idx]
# original_text = texts1[original_idx]
# duplicate_text = texts2[duplicate_idx]
# differences = display_word_differences(original_text, duplicate_text)
# result_text += f"**Original text (Dataset 1):**\n{original_text}\n\n"
# result_text += f"**Duplicate text (Dataset 2):**\n{duplicate_text}\n\n"
# result_text += f"**Differences:**\n{differences}\n"
# result_text += "-" * 50 + "\n\n"
# return result_text
# with gr.Blocks() as demo:
# gr.Markdown("# Semantic Deduplication")
# deduplication_type = gr.Radio(
# choices=["Single dataset", "Cross-dataset"],
# label="Deduplication Type",
# value="Single dataset"
# )
# with gr.Row():
# dataset1_name = gr.Textbox(value="ag_news", label="Dataset 1 Name")
# dataset1_split = gr.Textbox(value="train", label="Dataset 1 Split")
# dataset1_text_column = gr.Textbox(value="text", label="Text Column Name")
# dataset2_inputs = gr.Column(visible=False)
# with dataset2_inputs:
# gr.Markdown("### Dataset 2")
# with gr.Row():
# dataset2_name = gr.Textbox(value="ag_news", label="Dataset 2 Name")
# dataset2_split = gr.Textbox(value="test", label="Dataset 2 Split")
# dataset2_text_column = gr.Textbox(value="text", label="Text Column Name")
# threshold = gr.Slider(
# minimum=0.0,
# maximum=1.0,
# value=0.8,
# label="Similarity Threshold"
# )
# compute_button = gr.Button("Compute")
# output = gr.Markdown()
# # Function to update the visibility of dataset2_inputs
# def update_visibility(deduplication_type_value):
# if deduplication_type_value == "Cross-dataset":
# return gr.update(visible=True)
# else:
# return gr.update(visible=False)
# deduplication_type.change(
# update_visibility,
# inputs=deduplication_type,
# outputs=dataset2_inputs
# )
# compute_button.click(
# fn=perform_deduplication,
# inputs=[
# deduplication_type,
# dataset1_name,
# dataset1_split,
# dataset1_text_column,
# dataset2_name,
# dataset2_split,
# dataset2_text_column,
# threshold
# ],
# outputs=output
# )
# demo.launch()
|