Updates
Browse files
app.py
CHANGED
@@ -20,8 +20,17 @@ def deduplicate_embeddings(
|
|
20 |
threshold: float = 0.9,
|
21 |
batch_size: int = 1024,
|
22 |
progress=None
|
23 |
-
):
|
24 |
-
"""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
25 |
if embeddings_b is None:
|
26 |
reach = Reach(vectors=embeddings_a, items=[str(i) for i in range(len(embeddings_a))])
|
27 |
duplicate_to_original = {}
|
@@ -49,26 +58,53 @@ def deduplicate_embeddings(
|
|
49 |
return duplicate_indices_in_b, duplicate_to_original
|
50 |
|
51 |
def display_word_differences(x: str, y: str) -> str:
|
52 |
-
"""
|
|
|
|
|
|
|
|
|
|
|
|
|
53 |
diff = ndiff(x.split(), y.split())
|
54 |
return " ".join(word for word in diff if word.startswith(("+", "-")))
|
55 |
|
56 |
-
def load_dataset_texts(dataset_name, dataset_split, text_column):
|
57 |
-
"""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
58 |
ds = load_dataset(dataset_name, split=dataset_split)
|
59 |
return [example[text_column] for example in ds]
|
60 |
|
61 |
def perform_deduplication(
|
62 |
-
deduplication_type,
|
63 |
-
dataset1_name,
|
64 |
-
dataset1_split,
|
65 |
-
dataset1_text_column,
|
66 |
-
dataset2_name="",
|
67 |
-
dataset2_split="",
|
68 |
-
dataset2_text_column="",
|
69 |
-
threshold=default_threshold,
|
70 |
-
progress=gr.Progress(track_tqdm=True)
|
71 |
):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
72 |
try:
|
73 |
threshold = float(threshold)
|
74 |
|
@@ -76,8 +112,8 @@ def perform_deduplication(
|
|
76 |
yield "Loading Dataset 1...", ""
|
77 |
texts1 = load_dataset_texts(dataset1_name, dataset1_split, dataset1_text_column)
|
78 |
yield "Computing embeddings for Dataset 1...", ""
|
79 |
-
#embeddings1 = compute_embeddings(texts1, batch_size=64, progress=progress, desc="Dataset 1 embeddings")
|
80 |
embeddings1 = model.encode(texts1, show_progressbar=True)
|
|
|
81 |
if deduplication_type == "Single dataset":
|
82 |
# Deduplicate within Dataset 1
|
83 |
yield "Deduplicating within Dataset 1...", ""
|
@@ -114,8 +150,8 @@ def perform_deduplication(
|
|
114 |
yield "Loading Dataset 2...", ""
|
115 |
texts2 = load_dataset_texts(dataset2_name, dataset2_split, dataset2_text_column)
|
116 |
yield "Computing embeddings for Dataset 2...", ""
|
117 |
-
#embeddings2 = compute_embeddings(texts2, batch_size=64, progress=progress, desc="Dataset 2 embeddings")
|
118 |
embeddings2 = model.encode(texts2, show_progressbar=True)
|
|
|
119 |
# Deduplicate Dataset 2 against Dataset 1
|
120 |
yield "Deduplicating Dataset 2 against Dataset 1...", ""
|
121 |
duplicate_indices, duplicate_mapping = deduplicate_embeddings(
|
@@ -152,6 +188,12 @@ def perform_deduplication(
|
|
152 |
|
153 |
with gr.Blocks(css="#status_output { height: 50px; overflow: auto; }") as demo:
|
154 |
gr.Markdown("# Semantic Deduplication")
|
|
|
|
|
|
|
|
|
|
|
|
|
155 |
|
156 |
deduplication_type = gr.Radio(
|
157 |
choices=["Single dataset", "Cross-dataset"],
|
@@ -177,7 +219,7 @@ with gr.Blocks(css="#status_output { height: 50px; overflow: auto; }") as demo:
|
|
177 |
status_output = gr.Markdown(elem_id="status_output")
|
178 |
result_output = gr.Markdown()
|
179 |
|
180 |
-
def update_visibility(choice):
|
181 |
return gr.update(visible=choice == "Cross-dataset")
|
182 |
|
183 |
deduplication_type.change(update_visibility, inputs=deduplication_type, outputs=dataset2_inputs)
|
@@ -198,3 +240,205 @@ with gr.Blocks(css="#status_output { height: 50px; overflow: auto; }") as demo:
|
|
198 |
)
|
199 |
|
200 |
demo.launch()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
20 |
threshold: float = 0.9,
|
21 |
batch_size: int = 1024,
|
22 |
progress=None
|
23 |
+
) -> tuple[np.ndarray, dict[int, int]]:
|
24 |
+
"""
|
25 |
+
Deduplicate embeddings within one dataset or across two datasets.
|
26 |
+
|
27 |
+
:param embeddings_a: Embeddings of Dataset 1.
|
28 |
+
:param embeddings_b: Optional, embeddings of Dataset 2.
|
29 |
+
:param threshold: Similarity threshold for deduplication.
|
30 |
+
:param batch_size: Batch size for similarity computation.
|
31 |
+
:param progress: Gradio progress tracker for feedback.
|
32 |
+
:return: Deduplicated indices and a mapping of removed indices to their original counterparts.
|
33 |
+
"""
|
34 |
if embeddings_b is None:
|
35 |
reach = Reach(vectors=embeddings_a, items=[str(i) for i in range(len(embeddings_a))])
|
36 |
duplicate_to_original = {}
|
|
|
58 |
return duplicate_indices_in_b, duplicate_to_original
|
59 |
|
60 |
def display_word_differences(x: str, y: str) -> str:
|
61 |
+
"""
|
62 |
+
Display the word-level differences between two texts.
|
63 |
+
|
64 |
+
:param x: First text.
|
65 |
+
:param y: Second text.
|
66 |
+
:return: A string showing word-level differences.
|
67 |
+
"""
|
68 |
diff = ndiff(x.split(), y.split())
|
69 |
return " ".join(word for word in diff if word.startswith(("+", "-")))
|
70 |
|
71 |
+
def load_dataset_texts(dataset_name: str, dataset_split: str, text_column: str) -> list[str]:
|
72 |
+
"""
|
73 |
+
Load texts from a specified dataset and split.
|
74 |
+
|
75 |
+
:param dataset_name: Name of the dataset.
|
76 |
+
:param dataset_split: Split of the dataset (e.g., 'train', 'validation').
|
77 |
+
:param text_column: Name of the text column.
|
78 |
+
:return: A list of texts from the dataset.
|
79 |
+
"""
|
80 |
ds = load_dataset(dataset_name, split=dataset_split)
|
81 |
return [example[text_column] for example in ds]
|
82 |
|
83 |
def perform_deduplication(
|
84 |
+
deduplication_type: str,
|
85 |
+
dataset1_name: str,
|
86 |
+
dataset1_split: str,
|
87 |
+
dataset1_text_column: str,
|
88 |
+
dataset2_name: str = "",
|
89 |
+
dataset2_split: str = "",
|
90 |
+
dataset2_text_column: str = "",
|
91 |
+
threshold: float = default_threshold,
|
92 |
+
progress: gr.Progress = gr.Progress(track_tqdm=True)
|
93 |
):
|
94 |
+
"""
|
95 |
+
Perform deduplication on one or two datasets based on the deduplication type.
|
96 |
+
|
97 |
+
:param deduplication_type: 'Single dataset' or 'Cross-dataset'.
|
98 |
+
:param dataset1_name: Name of the first dataset.
|
99 |
+
:param dataset1_split: Split of the first dataset.
|
100 |
+
:param dataset1_text_column: Text column of the first dataset.
|
101 |
+
:param dataset2_name: Optional, name of the second dataset (for cross-dataset deduplication).
|
102 |
+
:param dataset2_split: Optional, split of the second dataset.
|
103 |
+
:param dataset2_text_column: Optional, text column of the second dataset.
|
104 |
+
:param threshold: Similarity threshold for deduplication.
|
105 |
+
:param progress: Gradio progress tracker.
|
106 |
+
:return: Status updates and result text for the Gradio interface.
|
107 |
+
"""
|
108 |
try:
|
109 |
threshold = float(threshold)
|
110 |
|
|
|
112 |
yield "Loading Dataset 1...", ""
|
113 |
texts1 = load_dataset_texts(dataset1_name, dataset1_split, dataset1_text_column)
|
114 |
yield "Computing embeddings for Dataset 1...", ""
|
|
|
115 |
embeddings1 = model.encode(texts1, show_progressbar=True)
|
116 |
+
|
117 |
if deduplication_type == "Single dataset":
|
118 |
# Deduplicate within Dataset 1
|
119 |
yield "Deduplicating within Dataset 1...", ""
|
|
|
150 |
yield "Loading Dataset 2...", ""
|
151 |
texts2 = load_dataset_texts(dataset2_name, dataset2_split, dataset2_text_column)
|
152 |
yield "Computing embeddings for Dataset 2...", ""
|
|
|
153 |
embeddings2 = model.encode(texts2, show_progressbar=True)
|
154 |
+
|
155 |
# Deduplicate Dataset 2 against Dataset 1
|
156 |
yield "Deduplicating Dataset 2 against Dataset 1...", ""
|
157 |
duplicate_indices, duplicate_mapping = deduplicate_embeddings(
|
|
|
188 |
|
189 |
with gr.Blocks(css="#status_output { height: 50px; overflow: auto; }") as demo:
|
190 |
gr.Markdown("# Semantic Deduplication")
|
191 |
+
gr.Markdown("""
|
192 |
+
This demo showcases semantic deduplication using Model2Vec.
|
193 |
+
It can be used to identify duplicate texts within a single dataset or across two datasets.
|
194 |
+
You can adjust the similarity threshold to control the strictness of the deduplication.
|
195 |
+
NOTE: this demo runs on a free CPU backend, so it may be slow for large datasets. For faster results, please run the code locally.
|
196 |
+
""")
|
197 |
|
198 |
deduplication_type = gr.Radio(
|
199 |
choices=["Single dataset", "Cross-dataset"],
|
|
|
219 |
status_output = gr.Markdown(elem_id="status_output")
|
220 |
result_output = gr.Markdown()
|
221 |
|
222 |
+
def update_visibility(choice: str):
|
223 |
return gr.update(visible=choice == "Cross-dataset")
|
224 |
|
225 |
deduplication_type.change(update_visibility, inputs=deduplication_type, outputs=dataset2_inputs)
|
|
|
240 |
)
|
241 |
|
242 |
demo.launch()
|
243 |
+
|
244 |
+
|
245 |
+
# import gradio as gr
|
246 |
+
# from datasets import load_dataset
|
247 |
+
# import numpy as np
|
248 |
+
# from model2vec import StaticModel
|
249 |
+
# from reach import Reach
|
250 |
+
# from difflib import ndiff
|
251 |
+
|
252 |
+
# # Load the model
|
253 |
+
# model = StaticModel.from_pretrained("minishlab/M2V_base_output")
|
254 |
+
|
255 |
+
# # Default parameters
|
256 |
+
# default_dataset_name = "sst2"
|
257 |
+
# default_dataset_split = "train"
|
258 |
+
# default_text_column = "sentence"
|
259 |
+
# default_threshold = 0.9
|
260 |
+
|
261 |
+
# def deduplicate_embeddings(
|
262 |
+
# embeddings_a: np.ndarray,
|
263 |
+
# embeddings_b: np.ndarray = None,
|
264 |
+
# threshold: float = 0.9,
|
265 |
+
# batch_size: int = 1024,
|
266 |
+
# progress=None
|
267 |
+
# ):
|
268 |
+
# """Deduplicate within one dataset or across two datasets."""
|
269 |
+
# if embeddings_b is None:
|
270 |
+
# reach = Reach(vectors=embeddings_a, items=[str(i) for i in range(len(embeddings_a))])
|
271 |
+
# duplicate_to_original = {}
|
272 |
+
# results = reach.nearest_neighbor_threshold(
|
273 |
+
# embeddings_a, threshold=threshold, batch_size=batch_size, show_progressbar=False
|
274 |
+
# )
|
275 |
+
# for i, similar_items in enumerate(progress.tqdm(results, desc="Processing duplicates", total=len(embeddings_a))):
|
276 |
+
# for sim_idx, _ in similar_items:
|
277 |
+
# sim_idx = int(sim_idx)
|
278 |
+
# if sim_idx != i and sim_idx not in duplicate_to_original:
|
279 |
+
# duplicate_to_original[sim_idx] = i
|
280 |
+
# deduplicated_indices = set(range(len(embeddings_a))) - set(duplicate_to_original.keys())
|
281 |
+
# return deduplicated_indices, duplicate_to_original
|
282 |
+
# else:
|
283 |
+
# reach = Reach(vectors=embeddings_a, items=[str(i) for i in range(len(embeddings_a))])
|
284 |
+
# duplicate_indices_in_b = []
|
285 |
+
# duplicate_to_original = {}
|
286 |
+
# results = reach.nearest_neighbor_threshold(
|
287 |
+
# embeddings_b, threshold=threshold, batch_size=batch_size, show_progressbar=False
|
288 |
+
# )
|
289 |
+
# for i, similar_items in enumerate(progress.tqdm(results, desc="Processing duplicates", total=len(embeddings_b))):
|
290 |
+
# if similar_items:
|
291 |
+
# duplicate_indices_in_b.append(i)
|
292 |
+
# duplicate_to_original[i] = int(similar_items[0][0])
|
293 |
+
# return duplicate_indices_in_b, duplicate_to_original
|
294 |
+
|
295 |
+
# def display_word_differences(x: str, y: str) -> str:
|
296 |
+
# """Display differences between two texts."""
|
297 |
+
# diff = ndiff(x.split(), y.split())
|
298 |
+
# return " ".join(word for word in diff if word.startswith(("+", "-")))
|
299 |
+
|
300 |
+
# def load_dataset_texts(dataset_name, dataset_split, text_column):
|
301 |
+
# """Load texts from a specified dataset."""
|
302 |
+
# ds = load_dataset(dataset_name, split=dataset_split)
|
303 |
+
# return [example[text_column] for example in ds]
|
304 |
+
|
305 |
+
# def perform_deduplication(
|
306 |
+
# deduplication_type,
|
307 |
+
# dataset1_name,
|
308 |
+
# dataset1_split,
|
309 |
+
# dataset1_text_column,
|
310 |
+
# dataset2_name="",
|
311 |
+
# dataset2_split="",
|
312 |
+
# dataset2_text_column="",
|
313 |
+
# threshold=default_threshold,
|
314 |
+
# progress=gr.Progress(track_tqdm=True),
|
315 |
+
# ):
|
316 |
+
# try:
|
317 |
+
# threshold = float(threshold)
|
318 |
+
|
319 |
+
# # Load and process Dataset 1
|
320 |
+
# yield "Loading Dataset 1...", ""
|
321 |
+
# texts1 = load_dataset_texts(dataset1_name, dataset1_split, dataset1_text_column)
|
322 |
+
# yield "Computing embeddings for Dataset 1...", ""
|
323 |
+
# #embeddings1 = compute_embeddings(texts1, batch_size=64, progress=progress, desc="Dataset 1 embeddings")
|
324 |
+
# embeddings1 = model.encode(texts1, show_progressbar=True)
|
325 |
+
# if deduplication_type == "Single dataset":
|
326 |
+
# # Deduplicate within Dataset 1
|
327 |
+
# yield "Deduplicating within Dataset 1...", ""
|
328 |
+
# deduplicated_indices, duplicate_mapping = deduplicate_embeddings(
|
329 |
+
# embeddings1, threshold=threshold, progress=progress
|
330 |
+
# )
|
331 |
+
|
332 |
+
# num_duplicates = len(duplicate_mapping)
|
333 |
+
# result_text = (
|
334 |
+
# f"**Total documents:** {len(texts1)}\n\n"
|
335 |
+
# f"**Duplicates found:** {num_duplicates}\n\n"
|
336 |
+
# f"**Unique documents after deduplication:** {len(deduplicated_indices)}\n\n"
|
337 |
+
# )
|
338 |
+
|
339 |
+
# if num_duplicates > 0:
|
340 |
+
# result_text += "**Sample duplicates:**\n\n"
|
341 |
+
# for dup_idx, orig_idx in list(duplicate_mapping.items())[:5]:
|
342 |
+
# orig_text = texts1[orig_idx]
|
343 |
+
# dup_text = texts1[dup_idx]
|
344 |
+
# differences = display_word_differences(orig_text, dup_text)
|
345 |
+
# result_text += (
|
346 |
+
# f"**Original:**\n{orig_text}\n\n"
|
347 |
+
# f"**Duplicate:**\n{dup_text}\n\n"
|
348 |
+
# f"**Differences:**\n{differences}\n"
|
349 |
+
# + "-" * 50 + "\n\n"
|
350 |
+
# )
|
351 |
+
# else:
|
352 |
+
# result_text += "No duplicates found."
|
353 |
+
|
354 |
+
# yield "Deduplication completed.", result_text
|
355 |
+
|
356 |
+
# else:
|
357 |
+
# # Load and process Dataset 2
|
358 |
+
# yield "Loading Dataset 2...", ""
|
359 |
+
# texts2 = load_dataset_texts(dataset2_name, dataset2_split, dataset2_text_column)
|
360 |
+
# yield "Computing embeddings for Dataset 2...", ""
|
361 |
+
# #embeddings2 = compute_embeddings(texts2, batch_size=64, progress=progress, desc="Dataset 2 embeddings")
|
362 |
+
# embeddings2 = model.encode(texts2, show_progressbar=True)
|
363 |
+
# # Deduplicate Dataset 2 against Dataset 1
|
364 |
+
# yield "Deduplicating Dataset 2 against Dataset 1...", ""
|
365 |
+
# duplicate_indices, duplicate_mapping = deduplicate_embeddings(
|
366 |
+
# embeddings1, embeddings_b=embeddings2, threshold=threshold, progress=progress
|
367 |
+
# )
|
368 |
+
|
369 |
+
# num_duplicates = len(duplicate_indices)
|
370 |
+
# result_text = (
|
371 |
+
# f"**Total documents in {dataset2_name}/{dataset2_split}:** {len(texts2)}\n\n"
|
372 |
+
# f"**Duplicates found in Dataset 2:** {num_duplicates}\n\n"
|
373 |
+
# f"**Unique documents after deduplication:** {len(texts2) - num_duplicates}\n\n"
|
374 |
+
# )
|
375 |
+
|
376 |
+
# if num_duplicates > 0:
|
377 |
+
# result_text += "**Sample duplicates from Dataset 2:**\n\n"
|
378 |
+
# for idx in duplicate_indices[:5]:
|
379 |
+
# orig_text = texts1[duplicate_mapping[idx]]
|
380 |
+
# dup_text = texts2[idx]
|
381 |
+
# differences = display_word_differences(orig_text, dup_text)
|
382 |
+
# result_text += (
|
383 |
+
# f"**Original (Dataset 1):**\n{orig_text}\n\n"
|
384 |
+
# f"**Duplicate (Dataset 2):**\n{dup_text}\n\n"
|
385 |
+
# f"**Differences:**\n{differences}\n"
|
386 |
+
# + "-" * 50 + "\n\n"
|
387 |
+
# )
|
388 |
+
# else:
|
389 |
+
# result_text += "No duplicates found."
|
390 |
+
|
391 |
+
# yield "Deduplication completed.", result_text
|
392 |
+
|
393 |
+
# except Exception as e:
|
394 |
+
# yield f"An error occurred: {e}", ""
|
395 |
+
# raise e
|
396 |
+
|
397 |
+
# with gr.Blocks(css="#status_output { height: 50px; overflow: auto; }") as demo:
|
398 |
+
# gr.Markdown("# Semantic Deduplication")
|
399 |
+
|
400 |
+
# deduplication_type = gr.Radio(
|
401 |
+
# choices=["Single dataset", "Cross-dataset"],
|
402 |
+
# label="Deduplication Type",
|
403 |
+
# value="Single dataset",
|
404 |
+
# )
|
405 |
+
|
406 |
+
# with gr.Row():
|
407 |
+
# dataset1_name = gr.Textbox(value=default_dataset_name, label="Dataset 1 Name")
|
408 |
+
# dataset1_split = gr.Textbox(value=default_dataset_split, label="Dataset 1 Split")
|
409 |
+
# dataset1_text_column = gr.Textbox(value=default_text_column, label="Text Column Name")
|
410 |
+
|
411 |
+
# dataset2_inputs = gr.Column(visible=False)
|
412 |
+
# with dataset2_inputs:
|
413 |
+
# gr.Markdown("### Dataset 2")
|
414 |
+
# with gr.Row():
|
415 |
+
# dataset2_name = gr.Textbox(value=default_dataset_name, label="Dataset 2 Name")
|
416 |
+
# dataset2_split = gr.Textbox(value=default_dataset_split, label="Dataset 2 Split")
|
417 |
+
# dataset2_text_column = gr.Textbox(value=default_text_column, label="Text Column Name")
|
418 |
+
|
419 |
+
# threshold = gr.Slider(0.0, 1.0, value=default_threshold, label="Similarity Threshold")
|
420 |
+
# compute_button = gr.Button("Compute")
|
421 |
+
# status_output = gr.Markdown(elem_id="status_output")
|
422 |
+
# result_output = gr.Markdown()
|
423 |
+
|
424 |
+
# def update_visibility(choice):
|
425 |
+
# return gr.update(visible=choice == "Cross-dataset")
|
426 |
+
|
427 |
+
# deduplication_type.change(update_visibility, inputs=deduplication_type, outputs=dataset2_inputs)
|
428 |
+
|
429 |
+
# compute_button.click(
|
430 |
+
# fn=perform_deduplication,
|
431 |
+
# inputs=[
|
432 |
+
# deduplication_type,
|
433 |
+
# dataset1_name,
|
434 |
+
# dataset1_split,
|
435 |
+
# dataset1_text_column,
|
436 |
+
# dataset2_name,
|
437 |
+
# dataset2_split,
|
438 |
+
# dataset2_text_column,
|
439 |
+
# threshold,
|
440 |
+
# ],
|
441 |
+
# outputs=[status_output, result_output],
|
442 |
+
# )
|
443 |
+
|
444 |
+
# demo.launch()
|