Updated app with code for deduplication
Browse files
app.py
CHANGED
@@ -26,13 +26,13 @@ def deduplicate(embedding_matrix: np.ndarray, threshold: float, batch_size: int
|
|
26 |
"""
|
27 |
Deduplicate embeddings and return the deduplicated indices and a mapping of removed indices to their corresponding original indices.
|
28 |
"""
|
29 |
-
#
|
30 |
reach = Reach(vectors=embedding_matrix, items=[str(i) for i in range(len(embedding_matrix))])
|
31 |
|
32 |
deduplicated_indices = set(range(len(embedding_matrix)))
|
33 |
duplicate_to_original_mapping = {}
|
34 |
|
35 |
-
#
|
36 |
results = reach.nearest_neighbor_threshold(
|
37 |
embedding_matrix,
|
38 |
threshold=threshold,
|
@@ -40,7 +40,7 @@ def deduplicate(embedding_matrix: np.ndarray, threshold: float, batch_size: int
|
|
40 |
show_progressbar=True # Allow internal progress bar
|
41 |
)
|
42 |
|
43 |
-
#
|
44 |
for i, similar_items in enumerate(results):
|
45 |
if i not in deduplicated_indices:
|
46 |
continue
|
@@ -58,13 +58,13 @@ def deduplicate_across_datasets(embedding_matrix_1: np.ndarray, embedding_matrix
|
|
58 |
"""
|
59 |
Deduplicate embeddings across two datasets and return the indices of duplicates between them.
|
60 |
"""
|
61 |
-
#
|
62 |
reach = Reach(vectors=embedding_matrix_1, items=[str(i) for i in range(len(embedding_matrix_1))])
|
63 |
|
64 |
duplicate_indices_in_test = []
|
65 |
duplicate_to_original_mapping = {}
|
66 |
|
67 |
-
#
|
68 |
results = reach.nearest_neighbor_threshold(
|
69 |
embedding_matrix_2,
|
70 |
threshold=threshold,
|
@@ -72,7 +72,7 @@ def deduplicate_across_datasets(embedding_matrix_1: np.ndarray, embedding_matrix
|
|
72 |
show_progressbar=True # Allow internal progress bar
|
73 |
)
|
74 |
|
75 |
-
#
|
76 |
for i, similar_items in enumerate(results):
|
77 |
similar_indices = [int(item[0]) for item in similar_items if item[1] >= threshold]
|
78 |
|
@@ -97,17 +97,35 @@ def perform_deduplication(
|
|
97 |
threshold=default_threshold,
|
98 |
progress=gr.Progress(track_tqdm=True)
|
99 |
):
|
100 |
-
#
|
101 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
102 |
tqdm.tqdm = progress.tqdm
|
103 |
-
|
104 |
-
|
105 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
106 |
|
107 |
try:
|
108 |
# Convert threshold to float
|
109 |
threshold = float(threshold)
|
110 |
-
|
111 |
# Initialize status message
|
112 |
status = ""
|
113 |
|
@@ -119,33 +137,33 @@ def perform_deduplication(
|
|
119 |
ds = ds_default1
|
120 |
else:
|
121 |
ds = load_dataset(dataset1_name, split=dataset1_split)
|
122 |
-
|
123 |
# Extract texts
|
124 |
status = "Extracting texts from Dataset 1..."
|
125 |
yield status, ""
|
126 |
texts = [example[dataset1_text_column] for example in ds]
|
127 |
-
|
128 |
# Compute embeddings
|
129 |
status = "Computing embeddings for Dataset 1..."
|
130 |
yield status, ""
|
131 |
embedding_matrix = model.encode(texts, show_progressbar=True) # Enable internal progress bar
|
132 |
-
|
133 |
# Deduplicate
|
134 |
status = "Deduplicating embeddings..."
|
135 |
yield status, ""
|
136 |
deduplicated_indices, duplicate_to_original_mapping = deduplicate(
|
137 |
embedding_matrix, threshold
|
138 |
)
|
139 |
-
|
140 |
# Prepare the results
|
141 |
num_duplicates = len(duplicate_to_original_mapping)
|
142 |
num_total = len(texts)
|
143 |
num_deduplicated = len(deduplicated_indices)
|
144 |
-
|
145 |
result_text = f"**Total documents:** {num_total}\n"
|
146 |
result_text += f"**Number of duplicates found:** {num_duplicates}\n"
|
147 |
result_text += f"**Number of unique documents after deduplication:** {num_deduplicated}\n\n"
|
148 |
-
|
149 |
# Show deduplicated examples
|
150 |
if num_duplicates > 0:
|
151 |
result_text += "**Examples of duplicates found:**\n\n"
|
@@ -160,11 +178,11 @@ def perform_deduplication(
|
|
160 |
result_text += "-" * 50 + "\n\n"
|
161 |
else:
|
162 |
result_text += "No duplicates found."
|
163 |
-
|
164 |
# Final status
|
165 |
status = "Deduplication completed."
|
166 |
yield status, result_text
|
167 |
-
|
168 |
elif deduplication_type == "Cross-dataset":
|
169 |
# Load Dataset 1
|
170 |
status = "Loading Dataset 1..."
|
@@ -173,7 +191,7 @@ def perform_deduplication(
|
|
173 |
ds1 = ds_default1
|
174 |
else:
|
175 |
ds1 = load_dataset(dataset1_name, split=dataset1_split)
|
176 |
-
|
177 |
# Load Dataset 2
|
178 |
status = "Loading Dataset 2..."
|
179 |
yield status, ""
|
@@ -181,42 +199,42 @@ def perform_deduplication(
|
|
181 |
ds2 = ds_default2
|
182 |
else:
|
183 |
ds2 = load_dataset(dataset2_name, split=dataset2_split)
|
184 |
-
|
185 |
# Extract texts from Dataset 1
|
186 |
status = "Extracting texts from Dataset 1..."
|
187 |
yield status, ""
|
188 |
texts1 = [example[dataset1_text_column] for example in ds1]
|
189 |
-
|
190 |
# Extract texts from Dataset 2
|
191 |
status = "Extracting texts from Dataset 2..."
|
192 |
yield status, ""
|
193 |
texts2 = [example[dataset2_text_column] for example in ds2]
|
194 |
-
|
195 |
# Compute embeddings for Dataset 1
|
196 |
status = "Computing embeddings for Dataset 1..."
|
197 |
yield status, ""
|
198 |
embedding_matrix1 = model.encode(texts1, show_progressbar=True)
|
199 |
-
|
200 |
# Compute embeddings for Dataset 2
|
201 |
status = "Computing embeddings for Dataset 2..."
|
202 |
yield status, ""
|
203 |
embedding_matrix2 = model.encode(texts2, show_progressbar=True)
|
204 |
-
|
205 |
# Deduplicate across datasets
|
206 |
status = "Deduplicating embeddings across datasets..."
|
207 |
yield status, ""
|
208 |
duplicate_indices_in_ds2, duplicate_to_original_mapping = deduplicate_across_datasets(
|
209 |
embedding_matrix1, embedding_matrix2, threshold
|
210 |
)
|
211 |
-
|
212 |
num_duplicates = len(duplicate_indices_in_ds2)
|
213 |
num_total_ds2 = len(texts2)
|
214 |
num_unique_ds2 = num_total_ds2 - num_duplicates
|
215 |
-
|
216 |
result_text = f"**Total documents in {dataset2_name}/{dataset2_split}:** {num_total_ds2}\n"
|
217 |
result_text += f"**Number of duplicates found in {dataset2_name}/{dataset2_split}:** {num_duplicates}\n"
|
218 |
result_text += f"**Number of unique documents in {dataset2_name}/{dataset2_split} after deduplication:** {num_unique_ds2}\n\n"
|
219 |
-
|
220 |
# Show deduplicated examples
|
221 |
if num_duplicates > 0:
|
222 |
result_text += "**Examples of duplicates found in Dataset 2:**\n\n"
|
@@ -232,17 +250,18 @@ def perform_deduplication(
|
|
232 |
result_text += "-" * 50 + "\n\n"
|
233 |
else:
|
234 |
result_text += "No duplicates found."
|
235 |
-
|
236 |
# Final status
|
237 |
status = "Deduplication completed."
|
238 |
yield status, result_text
|
239 |
|
240 |
finally:
|
241 |
-
# Restore original tqdm
|
242 |
-
tqdm.tqdm =
|
243 |
-
|
244 |
-
|
245 |
-
|
|
|
246 |
|
247 |
with gr.Blocks() as demo:
|
248 |
gr.Markdown("# Semantic Deduplication")
|
@@ -305,10 +324,321 @@ with gr.Blocks() as demo:
|
|
305 |
],
|
306 |
outputs=[status_output, result_output]
|
307 |
)
|
308 |
-
|
309 |
demo.launch()
|
310 |
|
311 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
312 |
# import gradio as gr
|
313 |
# from datasets import load_dataset
|
314 |
# import numpy as np
|
|
|
26 |
"""
|
27 |
Deduplicate embeddings and return the deduplicated indices and a mapping of removed indices to their corresponding original indices.
|
28 |
"""
|
29 |
+
# Build the index
|
30 |
reach = Reach(vectors=embedding_matrix, items=[str(i) for i in range(len(embedding_matrix))])
|
31 |
|
32 |
deduplicated_indices = set(range(len(embedding_matrix)))
|
33 |
duplicate_to_original_mapping = {}
|
34 |
|
35 |
+
# Find nearest neighbors
|
36 |
results = reach.nearest_neighbor_threshold(
|
37 |
embedding_matrix,
|
38 |
threshold=threshold,
|
|
|
40 |
show_progressbar=True # Allow internal progress bar
|
41 |
)
|
42 |
|
43 |
+
# Process duplicates
|
44 |
for i, similar_items in enumerate(results):
|
45 |
if i not in deduplicated_indices:
|
46 |
continue
|
|
|
58 |
"""
|
59 |
Deduplicate embeddings across two datasets and return the indices of duplicates between them.
|
60 |
"""
|
61 |
+
# Build the index from Dataset 1
|
62 |
reach = Reach(vectors=embedding_matrix_1, items=[str(i) for i in range(len(embedding_matrix_1))])
|
63 |
|
64 |
duplicate_indices_in_test = []
|
65 |
duplicate_to_original_mapping = {}
|
66 |
|
67 |
+
# Find nearest neighbors between datasets
|
68 |
results = reach.nearest_neighbor_threshold(
|
69 |
embedding_matrix_2,
|
70 |
threshold=threshold,
|
|
|
72 |
show_progressbar=True # Allow internal progress bar
|
73 |
)
|
74 |
|
75 |
+
# Process duplicates
|
76 |
for i, similar_items in enumerate(results):
|
77 |
similar_indices = [int(item[0]) for item in similar_items if item[1] >= threshold]
|
78 |
|
|
|
97 |
threshold=default_threshold,
|
98 |
progress=gr.Progress(track_tqdm=True)
|
99 |
):
|
100 |
+
# Custom tqdm class that wraps progress.tqdm and includes module-level attributes
|
101 |
+
class TqdmWrapper(tqdm.std.tqdm):
|
102 |
+
def __init__(self, *args, **kwargs):
|
103 |
+
super().__init__(*args, **kwargs)
|
104 |
+
|
105 |
+
# Copy module-level attributes from original tqdm module
|
106 |
+
TqdmWrapper.format_interval = staticmethod(tqdm.format_interval)
|
107 |
+
TqdmWrapper.format_num = staticmethod(tqdm.format_num)
|
108 |
+
TqdmWrapper.format_sizeof = staticmethod(tqdm.format_sizeof)
|
109 |
+
|
110 |
+
# Monkey-patch tqdm.tqdm with our wrapper
|
111 |
+
original_tqdm_tqdm = tqdm.tqdm
|
112 |
tqdm.tqdm = progress.tqdm
|
113 |
+
|
114 |
+
# Monkey-patch model2vec's tqdm reference if needed
|
115 |
+
import model2vec.model
|
116 |
+
if hasattr(model2vec.model, 'tqdm'):
|
117 |
+
original_model2vec_tqdm = model2vec.model.tqdm
|
118 |
+
model2vec.model.tqdm = TqdmWrapper
|
119 |
+
|
120 |
+
# Monkey-patch reach's tqdm reference if needed
|
121 |
+
if hasattr(Reach, 'tqdm'):
|
122 |
+
original_reach_tqdm = Reach.tqdm
|
123 |
+
Reach.tqdm = TqdmWrapper
|
124 |
|
125 |
try:
|
126 |
# Convert threshold to float
|
127 |
threshold = float(threshold)
|
128 |
+
|
129 |
# Initialize status message
|
130 |
status = ""
|
131 |
|
|
|
137 |
ds = ds_default1
|
138 |
else:
|
139 |
ds = load_dataset(dataset1_name, split=dataset1_split)
|
140 |
+
|
141 |
# Extract texts
|
142 |
status = "Extracting texts from Dataset 1..."
|
143 |
yield status, ""
|
144 |
texts = [example[dataset1_text_column] for example in ds]
|
145 |
+
|
146 |
# Compute embeddings
|
147 |
status = "Computing embeddings for Dataset 1..."
|
148 |
yield status, ""
|
149 |
embedding_matrix = model.encode(texts, show_progressbar=True) # Enable internal progress bar
|
150 |
+
|
151 |
# Deduplicate
|
152 |
status = "Deduplicating embeddings..."
|
153 |
yield status, ""
|
154 |
deduplicated_indices, duplicate_to_original_mapping = deduplicate(
|
155 |
embedding_matrix, threshold
|
156 |
)
|
157 |
+
|
158 |
# Prepare the results
|
159 |
num_duplicates = len(duplicate_to_original_mapping)
|
160 |
num_total = len(texts)
|
161 |
num_deduplicated = len(deduplicated_indices)
|
162 |
+
|
163 |
result_text = f"**Total documents:** {num_total}\n"
|
164 |
result_text += f"**Number of duplicates found:** {num_duplicates}\n"
|
165 |
result_text += f"**Number of unique documents after deduplication:** {num_deduplicated}\n\n"
|
166 |
+
|
167 |
# Show deduplicated examples
|
168 |
if num_duplicates > 0:
|
169 |
result_text += "**Examples of duplicates found:**\n\n"
|
|
|
178 |
result_text += "-" * 50 + "\n\n"
|
179 |
else:
|
180 |
result_text += "No duplicates found."
|
181 |
+
|
182 |
# Final status
|
183 |
status = "Deduplication completed."
|
184 |
yield status, result_text
|
185 |
+
|
186 |
elif deduplication_type == "Cross-dataset":
|
187 |
# Load Dataset 1
|
188 |
status = "Loading Dataset 1..."
|
|
|
191 |
ds1 = ds_default1
|
192 |
else:
|
193 |
ds1 = load_dataset(dataset1_name, split=dataset1_split)
|
194 |
+
|
195 |
# Load Dataset 2
|
196 |
status = "Loading Dataset 2..."
|
197 |
yield status, ""
|
|
|
199 |
ds2 = ds_default2
|
200 |
else:
|
201 |
ds2 = load_dataset(dataset2_name, split=dataset2_split)
|
202 |
+
|
203 |
# Extract texts from Dataset 1
|
204 |
status = "Extracting texts from Dataset 1..."
|
205 |
yield status, ""
|
206 |
texts1 = [example[dataset1_text_column] for example in ds1]
|
207 |
+
|
208 |
# Extract texts from Dataset 2
|
209 |
status = "Extracting texts from Dataset 2..."
|
210 |
yield status, ""
|
211 |
texts2 = [example[dataset2_text_column] for example in ds2]
|
212 |
+
|
213 |
# Compute embeddings for Dataset 1
|
214 |
status = "Computing embeddings for Dataset 1..."
|
215 |
yield status, ""
|
216 |
embedding_matrix1 = model.encode(texts1, show_progressbar=True)
|
217 |
+
|
218 |
# Compute embeddings for Dataset 2
|
219 |
status = "Computing embeddings for Dataset 2..."
|
220 |
yield status, ""
|
221 |
embedding_matrix2 = model.encode(texts2, show_progressbar=True)
|
222 |
+
|
223 |
# Deduplicate across datasets
|
224 |
status = "Deduplicating embeddings across datasets..."
|
225 |
yield status, ""
|
226 |
duplicate_indices_in_ds2, duplicate_to_original_mapping = deduplicate_across_datasets(
|
227 |
embedding_matrix1, embedding_matrix2, threshold
|
228 |
)
|
229 |
+
|
230 |
num_duplicates = len(duplicate_indices_in_ds2)
|
231 |
num_total_ds2 = len(texts2)
|
232 |
num_unique_ds2 = num_total_ds2 - num_duplicates
|
233 |
+
|
234 |
result_text = f"**Total documents in {dataset2_name}/{dataset2_split}:** {num_total_ds2}\n"
|
235 |
result_text += f"**Number of duplicates found in {dataset2_name}/{dataset2_split}:** {num_duplicates}\n"
|
236 |
result_text += f"**Number of unique documents in {dataset2_name}/{dataset2_split} after deduplication:** {num_unique_ds2}\n\n"
|
237 |
+
|
238 |
# Show deduplicated examples
|
239 |
if num_duplicates > 0:
|
240 |
result_text += "**Examples of duplicates found in Dataset 2:**\n\n"
|
|
|
250 |
result_text += "-" * 50 + "\n\n"
|
251 |
else:
|
252 |
result_text += "No duplicates found."
|
253 |
+
|
254 |
# Final status
|
255 |
status = "Deduplication completed."
|
256 |
yield status, result_text
|
257 |
|
258 |
finally:
|
259 |
+
# Restore original tqdm functions
|
260 |
+
tqdm.tqdm = original_tqdm_tqdm
|
261 |
+
if hasattr(model2vec.model, 'tqdm'):
|
262 |
+
model2vec.model.tqdm = original_model2vec_tqdm
|
263 |
+
if hasattr(Reach, 'tqdm'):
|
264 |
+
Reach.tqdm = original_reach_tqdm
|
265 |
|
266 |
with gr.Blocks() as demo:
|
267 |
gr.Markdown("# Semantic Deduplication")
|
|
|
324 |
],
|
325 |
outputs=[status_output, result_output]
|
326 |
)
|
327 |
+
|
328 |
demo.launch()
|
329 |
|
330 |
|
331 |
+
# import gradio as gr
|
332 |
+
# from datasets import load_dataset
|
333 |
+
# import numpy as np
|
334 |
+
# from model2vec import StaticModel
|
335 |
+
# from reach import Reach
|
336 |
+
# from difflib import ndiff
|
337 |
+
# import sys
|
338 |
+
# import tqdm
|
339 |
+
|
340 |
+
# # Load the model at startup
|
341 |
+
# model = StaticModel.from_pretrained("minishlab/M2V_base_output")
|
342 |
+
|
343 |
+
# # Update default dataset to 'sst2' and set default threshold to 0.9
|
344 |
+
# default_dataset1_name = "sst2"
|
345 |
+
# default_dataset1_split = "train"
|
346 |
+
# default_dataset2_name = "sst2"
|
347 |
+
# default_dataset2_split = "validation"
|
348 |
+
# default_text_column = "sentence"
|
349 |
+
# default_threshold = 0.9
|
350 |
+
|
351 |
+
# # Load the default datasets at startup
|
352 |
+
# ds_default1 = load_dataset(default_dataset1_name, split=default_dataset1_split)
|
353 |
+
# ds_default2 = load_dataset(default_dataset2_name, split=default_dataset2_split)
|
354 |
+
|
355 |
+
# def deduplicate(embedding_matrix: np.ndarray, threshold: float, batch_size: int = 1024) -> tuple[np.ndarray, dict[int, int]]:
|
356 |
+
# """
|
357 |
+
# Deduplicate embeddings and return the deduplicated indices and a mapping of removed indices to their corresponding original indices.
|
358 |
+
# """
|
359 |
+
# # Building the index
|
360 |
+
# reach = Reach(vectors=embedding_matrix, items=[str(i) for i in range(len(embedding_matrix))])
|
361 |
+
|
362 |
+
# deduplicated_indices = set(range(len(embedding_matrix)))
|
363 |
+
# duplicate_to_original_mapping = {}
|
364 |
+
|
365 |
+
# # Finding nearest neighbors
|
366 |
+
# results = reach.nearest_neighbor_threshold(
|
367 |
+
# embedding_matrix,
|
368 |
+
# threshold=threshold,
|
369 |
+
# batch_size=batch_size,
|
370 |
+
# show_progressbar=True # Allow internal progress bar
|
371 |
+
# )
|
372 |
+
|
373 |
+
# # Processing duplicates
|
374 |
+
# for i, similar_items in enumerate(results):
|
375 |
+
# if i not in deduplicated_indices:
|
376 |
+
# continue
|
377 |
+
|
378 |
+
# similar_indices = [int(item[0]) for item in similar_items if int(item[0]) != i]
|
379 |
+
|
380 |
+
# for sim_idx in similar_indices:
|
381 |
+
# if sim_idx in deduplicated_indices:
|
382 |
+
# deduplicated_indices.remove(sim_idx)
|
383 |
+
# duplicate_to_original_mapping[sim_idx] = i
|
384 |
+
|
385 |
+
# return np.array(list(deduplicated_indices)), duplicate_to_original_mapping
|
386 |
+
|
387 |
+
# def deduplicate_across_datasets(embedding_matrix_1: np.ndarray, embedding_matrix_2: np.ndarray, threshold: float, batch_size: int = 1024) -> tuple[list[int], dict[int, int]]:
|
388 |
+
# """
|
389 |
+
# Deduplicate embeddings across two datasets and return the indices of duplicates between them.
|
390 |
+
# """
|
391 |
+
# # Building the index from Dataset 1
|
392 |
+
# reach = Reach(vectors=embedding_matrix_1, items=[str(i) for i in range(len(embedding_matrix_1))])
|
393 |
+
|
394 |
+
# duplicate_indices_in_test = []
|
395 |
+
# duplicate_to_original_mapping = {}
|
396 |
+
|
397 |
+
# # Finding nearest neighbors between datasets
|
398 |
+
# results = reach.nearest_neighbor_threshold(
|
399 |
+
# embedding_matrix_2,
|
400 |
+
# threshold=threshold,
|
401 |
+
# batch_size=batch_size,
|
402 |
+
# show_progressbar=True # Allow internal progress bar
|
403 |
+
# )
|
404 |
+
|
405 |
+
# # Processing duplicates
|
406 |
+
# for i, similar_items in enumerate(results):
|
407 |
+
# similar_indices = [int(item[0]) for item in similar_items if item[1] >= threshold]
|
408 |
+
|
409 |
+
# if similar_indices:
|
410 |
+
# duplicate_indices_in_test.append(i)
|
411 |
+
# duplicate_to_original_mapping[i] = similar_indices[0]
|
412 |
+
|
413 |
+
# return duplicate_indices_in_test, duplicate_to_original_mapping
|
414 |
+
|
415 |
+
# def display_word_differences(x: str, y: str) -> str:
|
416 |
+
# diff = ndiff(x.split(), y.split())
|
417 |
+
# return " ".join([word for word in diff if word.startswith(('+', '-'))])
|
418 |
+
|
419 |
+
# def perform_deduplication(
|
420 |
+
# deduplication_type,
|
421 |
+
# dataset1_name,
|
422 |
+
# dataset1_split,
|
423 |
+
# dataset1_text_column,
|
424 |
+
# dataset2_name="",
|
425 |
+
# dataset2_split="",
|
426 |
+
# dataset2_text_column="",
|
427 |
+
# threshold=default_threshold,
|
428 |
+
# progress=gr.Progress(track_tqdm=True)
|
429 |
+
# ):
|
430 |
+
# # Deep Monkey-Patching of tqdm
|
431 |
+
# original_tqdm = tqdm.tqdm
|
432 |
+
# tqdm.tqdm = progress.tqdm
|
433 |
+
# for mod_name in list(sys.modules.keys()):
|
434 |
+
# if 'tqdm' in mod_name:
|
435 |
+
# sys.modules[mod_name].tqdm = progress.tqdm
|
436 |
+
|
437 |
+
# try:
|
438 |
+
# # Convert threshold to float
|
439 |
+
# threshold = float(threshold)
|
440 |
+
|
441 |
+
# # Initialize status message
|
442 |
+
# status = ""
|
443 |
+
|
444 |
+
# if deduplication_type == "Single dataset":
|
445 |
+
# # Load Dataset 1
|
446 |
+
# status = "Loading Dataset 1..."
|
447 |
+
# yield status, ""
|
448 |
+
# if dataset1_name == default_dataset1_name and dataset1_split == default_dataset1_split:
|
449 |
+
# ds = ds_default1
|
450 |
+
# else:
|
451 |
+
# ds = load_dataset(dataset1_name, split=dataset1_split)
|
452 |
+
|
453 |
+
# # Extract texts
|
454 |
+
# status = "Extracting texts from Dataset 1..."
|
455 |
+
# yield status, ""
|
456 |
+
# texts = [example[dataset1_text_column] for example in ds]
|
457 |
+
|
458 |
+
# # Compute embeddings
|
459 |
+
# status = "Computing embeddings for Dataset 1..."
|
460 |
+
# yield status, ""
|
461 |
+
# embedding_matrix = model.encode(texts, show_progressbar=True) # Enable internal progress bar
|
462 |
+
|
463 |
+
# # Deduplicate
|
464 |
+
# status = "Deduplicating embeddings..."
|
465 |
+
# yield status, ""
|
466 |
+
# deduplicated_indices, duplicate_to_original_mapping = deduplicate(
|
467 |
+
# embedding_matrix, threshold
|
468 |
+
# )
|
469 |
+
|
470 |
+
# # Prepare the results
|
471 |
+
# num_duplicates = len(duplicate_to_original_mapping)
|
472 |
+
# num_total = len(texts)
|
473 |
+
# num_deduplicated = len(deduplicated_indices)
|
474 |
+
|
475 |
+
# result_text = f"**Total documents:** {num_total}\n"
|
476 |
+
# result_text += f"**Number of duplicates found:** {num_duplicates}\n"
|
477 |
+
# result_text += f"**Number of unique documents after deduplication:** {num_deduplicated}\n\n"
|
478 |
+
|
479 |
+
# # Show deduplicated examples
|
480 |
+
# if num_duplicates > 0:
|
481 |
+
# result_text += "**Examples of duplicates found:**\n\n"
|
482 |
+
# num_examples = min(5, num_duplicates)
|
483 |
+
# for duplicate_idx, original_idx in list(duplicate_to_original_mapping.items())[:num_examples]:
|
484 |
+
# original_text = texts[original_idx]
|
485 |
+
# duplicate_text = texts[duplicate_idx]
|
486 |
+
# differences = display_word_differences(original_text, duplicate_text)
|
487 |
+
# result_text += f"**Original text:**\n{original_text}\n\n"
|
488 |
+
# result_text += f"**Duplicate text:**\n{duplicate_text}\n\n"
|
489 |
+
# result_text += f"**Differences:**\n{differences}\n"
|
490 |
+
# result_text += "-" * 50 + "\n\n"
|
491 |
+
# else:
|
492 |
+
# result_text += "No duplicates found."
|
493 |
+
|
494 |
+
# # Final status
|
495 |
+
# status = "Deduplication completed."
|
496 |
+
# yield status, result_text
|
497 |
+
|
498 |
+
# elif deduplication_type == "Cross-dataset":
|
499 |
+
# # Load Dataset 1
|
500 |
+
# status = "Loading Dataset 1..."
|
501 |
+
# yield status, ""
|
502 |
+
# if dataset1_name == default_dataset1_name and dataset1_split == default_dataset1_split:
|
503 |
+
# ds1 = ds_default1
|
504 |
+
# else:
|
505 |
+
# ds1 = load_dataset(dataset1_name, split=dataset1_split)
|
506 |
+
|
507 |
+
# # Load Dataset 2
|
508 |
+
# status = "Loading Dataset 2..."
|
509 |
+
# yield status, ""
|
510 |
+
# if dataset2_name == default_dataset2_name and dataset2_split == default_dataset2_split:
|
511 |
+
# ds2 = ds_default2
|
512 |
+
# else:
|
513 |
+
# ds2 = load_dataset(dataset2_name, split=dataset2_split)
|
514 |
+
|
515 |
+
# # Extract texts from Dataset 1
|
516 |
+
# status = "Extracting texts from Dataset 1..."
|
517 |
+
# yield status, ""
|
518 |
+
# texts1 = [example[dataset1_text_column] for example in ds1]
|
519 |
+
|
520 |
+
# # Extract texts from Dataset 2
|
521 |
+
# status = "Extracting texts from Dataset 2..."
|
522 |
+
# yield status, ""
|
523 |
+
# texts2 = [example[dataset2_text_column] for example in ds2]
|
524 |
+
|
525 |
+
# # Compute embeddings for Dataset 1
|
526 |
+
# status = "Computing embeddings for Dataset 1..."
|
527 |
+
# yield status, ""
|
528 |
+
# embedding_matrix1 = model.encode(texts1, show_progressbar=True)
|
529 |
+
|
530 |
+
# # Compute embeddings for Dataset 2
|
531 |
+
# status = "Computing embeddings for Dataset 2..."
|
532 |
+
# yield status, ""
|
533 |
+
# embedding_matrix2 = model.encode(texts2, show_progressbar=True)
|
534 |
+
|
535 |
+
# # Deduplicate across datasets
|
536 |
+
# status = "Deduplicating embeddings across datasets..."
|
537 |
+
# yield status, ""
|
538 |
+
# duplicate_indices_in_ds2, duplicate_to_original_mapping = deduplicate_across_datasets(
|
539 |
+
# embedding_matrix1, embedding_matrix2, threshold
|
540 |
+
# )
|
541 |
+
|
542 |
+
# num_duplicates = len(duplicate_indices_in_ds2)
|
543 |
+
# num_total_ds2 = len(texts2)
|
544 |
+
# num_unique_ds2 = num_total_ds2 - num_duplicates
|
545 |
+
|
546 |
+
# result_text = f"**Total documents in {dataset2_name}/{dataset2_split}:** {num_total_ds2}\n"
|
547 |
+
# result_text += f"**Number of duplicates found in {dataset2_name}/{dataset2_split}:** {num_duplicates}\n"
|
548 |
+
# result_text += f"**Number of unique documents in {dataset2_name}/{dataset2_split} after deduplication:** {num_unique_ds2}\n\n"
|
549 |
+
|
550 |
+
# # Show deduplicated examples
|
551 |
+
# if num_duplicates > 0:
|
552 |
+
# result_text += "**Examples of duplicates found in Dataset 2:**\n\n"
|
553 |
+
# num_examples = min(5, num_duplicates)
|
554 |
+
# for duplicate_idx in duplicate_indices_in_ds2[:num_examples]:
|
555 |
+
# original_idx = duplicate_to_original_mapping[duplicate_idx]
|
556 |
+
# original_text = texts1[original_idx]
|
557 |
+
# duplicate_text = texts2[duplicate_idx]
|
558 |
+
# differences = display_word_differences(original_text, duplicate_text)
|
559 |
+
# result_text += f"**Original text (Dataset 1):**\n{original_text}\n\n"
|
560 |
+
# result_text += f"**Duplicate text (Dataset 2):**\n{duplicate_text}\n\n"
|
561 |
+
# result_text += f"**Differences:**\n{differences}\n"
|
562 |
+
# result_text += "-" * 50 + "\n\n"
|
563 |
+
# else:
|
564 |
+
# result_text += "No duplicates found."
|
565 |
+
|
566 |
+
# # Final status
|
567 |
+
# status = "Deduplication completed."
|
568 |
+
# yield status, result_text
|
569 |
+
|
570 |
+
# finally:
|
571 |
+
# # Restore original tqdm
|
572 |
+
# tqdm.tqdm = original_tqdm
|
573 |
+
# for mod_name in list(sys.modules.keys()):
|
574 |
+
# if 'tqdm' in mod_name:
|
575 |
+
# sys.modules[mod_name].tqdm = original_tqdm
|
576 |
+
|
577 |
+
# with gr.Blocks() as demo:
|
578 |
+
# gr.Markdown("# Semantic Deduplication")
|
579 |
+
|
580 |
+
# deduplication_type = gr.Radio(
|
581 |
+
# choices=["Single dataset", "Cross-dataset"],
|
582 |
+
# label="Deduplication Type",
|
583 |
+
# value="Single dataset"
|
584 |
+
# )
|
585 |
+
|
586 |
+
# with gr.Row():
|
587 |
+
# dataset1_name = gr.Textbox(value=default_dataset1_name, label="Dataset 1 Name")
|
588 |
+
# dataset1_split = gr.Textbox(value=default_dataset1_split, label="Dataset 1 Split")
|
589 |
+
# dataset1_text_column = gr.Textbox(value=default_text_column, label="Text Column Name")
|
590 |
+
|
591 |
+
# dataset2_inputs = gr.Column(visible=False)
|
592 |
+
# with dataset2_inputs:
|
593 |
+
# gr.Markdown("### Dataset 2")
|
594 |
+
# with gr.Row():
|
595 |
+
# dataset2_name = gr.Textbox(value=default_dataset2_name, label="Dataset 2 Name")
|
596 |
+
# dataset2_split = gr.Textbox(value=default_dataset2_split, label="Dataset 2 Split")
|
597 |
+
# dataset2_text_column = gr.Textbox(value=default_text_column, label="Text Column Name")
|
598 |
+
|
599 |
+
# threshold = gr.Slider(
|
600 |
+
# minimum=0.0,
|
601 |
+
# maximum=1.0,
|
602 |
+
# value=default_threshold,
|
603 |
+
# label="Similarity Threshold"
|
604 |
+
# )
|
605 |
+
|
606 |
+
# compute_button = gr.Button("Compute")
|
607 |
+
|
608 |
+
# status_output = gr.Markdown()
|
609 |
+
# result_output = gr.Markdown()
|
610 |
+
|
611 |
+
# # Function to update the visibility of dataset2_inputs
|
612 |
+
# def update_visibility(deduplication_type_value):
|
613 |
+
# if deduplication_type_value == "Cross-dataset":
|
614 |
+
# return gr.update(visible=True)
|
615 |
+
# else:
|
616 |
+
# return gr.update(visible=False)
|
617 |
+
|
618 |
+
# deduplication_type.change(
|
619 |
+
# update_visibility,
|
620 |
+
# inputs=deduplication_type,
|
621 |
+
# outputs=dataset2_inputs
|
622 |
+
# )
|
623 |
+
|
624 |
+
# compute_button.click(
|
625 |
+
# fn=perform_deduplication,
|
626 |
+
# inputs=[
|
627 |
+
# deduplication_type,
|
628 |
+
# dataset1_name,
|
629 |
+
# dataset1_split,
|
630 |
+
# dataset1_text_column,
|
631 |
+
# dataset2_name,
|
632 |
+
# dataset2_split,
|
633 |
+
# dataset2_text_column,
|
634 |
+
# threshold
|
635 |
+
# ],
|
636 |
+
# outputs=[status_output, result_output]
|
637 |
+
# )
|
638 |
+
|
639 |
+
# demo.launch()
|
640 |
+
|
641 |
+
|
642 |
# import gradio as gr
|
643 |
# from datasets import load_dataset
|
644 |
# import numpy as np
|