Spaces:
Sleeping
Sleeping
Export PNG from plot
Browse files- app.py +62 -49
- requirements.txt +1 -0
app.py
CHANGED
@@ -12,7 +12,7 @@ from bertopic import BERTopic
|
|
12 |
from bertopic.representation import KeyBERTInspired
|
13 |
from cuml.manifold import UMAP
|
14 |
from cuml.cluster import HDBSCAN
|
15 |
-
|
16 |
from sklearn.feature_extraction.text import CountVectorizer
|
17 |
from sentence_transformers import SentenceTransformer
|
18 |
|
@@ -25,12 +25,10 @@ import gradio as gr
|
|
25 |
|
26 |
"""
|
27 |
TODOs:
|
28 |
-
- Improve DataMapPlot plot arguments
|
29 |
-
- Add export button for final plot
|
30 |
-
- Export and serve an interactive HTML plot?
|
31 |
- Try with more rows
|
32 |
-
|
33 |
- Add TextGenerationLayer
|
|
|
|
|
34 |
- Make it run on Zero GPU
|
35 |
"""
|
36 |
|
@@ -38,6 +36,12 @@ load_dotenv()
|
|
38 |
HF_TOKEN = os.getenv("HF_TOKEN")
|
39 |
assert HF_TOKEN is not None, "You need to set HF_TOKEN in your environment variables"
|
40 |
|
|
|
|
|
|
|
|
|
|
|
|
|
41 |
logging.basicConfig(
|
42 |
level=logging.INFO, format="%(asctime)s - %(name)s - %(levelname)s - %(message)s"
|
43 |
)
|
@@ -145,6 +149,27 @@ def fit_model(docs, embeddings, n_neighbors, n_components):
|
|
145 |
logging.info("Global model updated")
|
146 |
|
147 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
148 |
def generate_topics(dataset, config, split, column, nested_column, plot_type):
|
149 |
logging.info(
|
150 |
f"Generating topics for {dataset} with config {config} {split} {column} {nested_column}"
|
@@ -159,7 +184,7 @@ def generate_topics(dataset, config, split, column, nested_column, plot_type):
|
|
159 |
|
160 |
reduce_umap_model = UMAP(
|
161 |
n_neighbors=n_neighbors,
|
162 |
-
n_components=2, # For visualization, keeping it
|
163 |
min_dist=0.0,
|
164 |
metric="cosine",
|
165 |
random_state=42,
|
@@ -183,6 +208,7 @@ def generate_topics(dataset, config, split, column, nested_column, plot_type):
|
|
183 |
gr.DataFrame(value=[], interactive=False, visible=True),
|
184 |
gr.Plot(value=None, visible=True),
|
185 |
gr.Label({message: rows_processed / limit}, visible=True),
|
|
|
186 |
)
|
187 |
while offset < limit:
|
188 |
docs = get_docs_from_parquet(parquet_urls, column, offset, CHUNK_SIZE)
|
@@ -216,59 +242,32 @@ def generate_topics(dataset, config, split, column, nested_column, plot_type):
|
|
216 |
topics_info = base_model.get_topic_info()
|
217 |
all_topics, _ = base_model.transform(all_docs)
|
218 |
all_topics = np.array(all_topics)
|
219 |
-
# topic_plot, _ = datamapplot.create_plot(
|
220 |
-
# data_map_coords=reduced_embeddings_array,
|
221 |
-
# labels=all_topics.astype(str),
|
222 |
-
# use_medoids=True,
|
223 |
-
# figsize=(12, 12),
|
224 |
-
# dpi=100,
|
225 |
-
# title="PubMed - Literature review",
|
226 |
-
# sub_title="A data map of papers representing artificial intelligence and machine learning in ophthalmology",
|
227 |
-
# title_keywords={"fontsize": 36, "fontfamily": "Roboto Black"},
|
228 |
-
# sub_title_keywords={
|
229 |
-
# "fontsize": 18,
|
230 |
-
# },
|
231 |
-
# highlight_label_keywords={
|
232 |
-
# "fontsize": 12,
|
233 |
-
# "fontweight": "bold",
|
234 |
-
# "bbox": {"boxstyle": "round"},
|
235 |
-
# },
|
236 |
-
# label_font_size=8,
|
237 |
-
# label_wrap_width=16,
|
238 |
-
# label_linespacing=1.25,
|
239 |
-
# label_direction_bias=1.3,
|
240 |
-
# label_margin_factor=2.0,
|
241 |
-
# label_base_radius=15.0,
|
242 |
-
# point_size=4,
|
243 |
-
# marker_type="o",
|
244 |
-
# arrowprops={
|
245 |
-
# "arrowstyle": "wedge,tail_width=0.5",
|
246 |
-
# "connectionstyle": "arc3,rad=0.05",
|
247 |
-
# "linewidth": 0,
|
248 |
-
# "fc": "#33333377",
|
249 |
-
# },
|
250 |
-
# add_glow=True,
|
251 |
-
# glow_keywords={
|
252 |
-
# "kernel_bandwidth": 0.75, # controls how wide the glow spreads.
|
253 |
-
# "kernel": "cosine", # controls the kernel type. Default is "gaussian". See https://scikit-learn.org/stable/modules/density.html#kernel-density.
|
254 |
-
# "n_levels": 32, # controls how many "levels" there are in the contour plot.
|
255 |
-
# "max_alpha": 0.9, # controls the translucency of the glow.
|
256 |
-
# },
|
257 |
-
# darkmode=False,
|
258 |
-
# )
|
259 |
|
260 |
topic_plot = (
|
261 |
base_model.visualize_document_datamap(
|
262 |
docs=all_docs,
|
263 |
reduced_embeddings=reduced_embeddings_array,
|
264 |
-
title=
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
265 |
)
|
266 |
if plot_type == "DataMapPlot"
|
267 |
else base_model.visualize_documents(
|
268 |
docs=all_docs,
|
269 |
reduced_embeddings=reduced_embeddings_array,
|
270 |
custom_labels=True,
|
271 |
-
title=
|
272 |
)
|
273 |
)
|
274 |
|
@@ -286,12 +285,23 @@ def generate_topics(dataset, config, split, column, nested_column, plot_type):
|
|
286 |
topics_info,
|
287 |
topic_plot,
|
288 |
gr.Label({message: progress}, visible=True),
|
|
|
289 |
)
|
290 |
|
291 |
offset += CHUNK_SIZE
|
292 |
|
293 |
logging.info("Finished processing all data")
|
294 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
295 |
yield (
|
296 |
gr.Accordion(open=False),
|
297 |
topics_info,
|
@@ -299,6 +309,7 @@ def generate_topics(dataset, config, split, column, nested_column, plot_type):
|
|
299 |
gr.Label(
|
300 |
{f"✅ Done: {rows_processed} rows have been processed": 1.0}, visible=True
|
301 |
),
|
|
|
302 |
)
|
303 |
cuda.empty_cache()
|
304 |
|
@@ -339,7 +350,7 @@ with gr.Blocks() as demo:
|
|
339 |
)
|
340 |
plot_type_radio = gr.Radio(
|
341 |
["DataMapPlot", "Plotly"],
|
342 |
-
value="
|
343 |
label="Choose the plot type",
|
344 |
interactive=True,
|
345 |
)
|
@@ -347,6 +358,7 @@ with gr.Blocks() as demo:
|
|
347 |
|
348 |
gr.Markdown("## Data map")
|
349 |
full_topics_generation_label = gr.Label(visible=False, show_label=False)
|
|
|
350 |
topics_plot = gr.Plot()
|
351 |
with gr.Accordion("Topics Info", open=False):
|
352 |
topics_df = gr.DataFrame(interactive=False, visible=True)
|
@@ -365,6 +377,7 @@ with gr.Blocks() as demo:
|
|
365 |
topics_df,
|
366 |
topics_plot,
|
367 |
full_topics_generation_label,
|
|
|
368 |
],
|
369 |
)
|
370 |
|
|
|
12 |
from bertopic.representation import KeyBERTInspired
|
13 |
from cuml.manifold import UMAP
|
14 |
from cuml.cluster import HDBSCAN
|
15 |
+
from huggingface_hub import HfApi
|
16 |
from sklearn.feature_extraction.text import CountVectorizer
|
17 |
from sentence_transformers import SentenceTransformer
|
18 |
|
|
|
25 |
|
26 |
"""
|
27 |
TODOs:
|
|
|
|
|
|
|
28 |
- Try with more rows
|
|
|
29 |
- Add TextGenerationLayer
|
30 |
+
- Try with more rows
|
31 |
+
- Export and serve an interactive HTML plot?
|
32 |
- Make it run on Zero GPU
|
33 |
"""
|
34 |
|
|
|
36 |
HF_TOKEN = os.getenv("HF_TOKEN")
|
37 |
assert HF_TOKEN is not None, "You need to set HF_TOKEN in your environment variables"
|
38 |
|
39 |
+
|
40 |
+
EXPORTS_REPOSITORY = os.getenv("EXPORTS_REPOSITORY")
|
41 |
+
assert (
|
42 |
+
EXPORTS_REPOSITORY is not None
|
43 |
+
), "You need to set EXPORTS_REPOSITORY in your environment variables"
|
44 |
+
|
45 |
logging.basicConfig(
|
46 |
level=logging.INFO, format="%(asctime)s - %(name)s - %(levelname)s - %(message)s"
|
47 |
)
|
|
|
149 |
logging.info("Global model updated")
|
150 |
|
151 |
|
152 |
+
def _push_to_hub(
|
153 |
+
dataset_id,
|
154 |
+
file_path,
|
155 |
+
):
|
156 |
+
logging.info(f"Pushing file to hub: {dataset_id} on file {file_path}")
|
157 |
+
|
158 |
+
file_name = file_path.split("/")[-1]
|
159 |
+
api = HfApi(token=HF_TOKEN)
|
160 |
+
try:
|
161 |
+
logging.info(f"About to push {file_path} - {dataset_id}")
|
162 |
+
api.upload_file(
|
163 |
+
path_or_fileobj=file_path,
|
164 |
+
path_in_repo=file_name,
|
165 |
+
repo_id=EXPORTS_REPOSITORY,
|
166 |
+
repo_type="dataset",
|
167 |
+
)
|
168 |
+
except Exception as e:
|
169 |
+
logging.info("Failed to push file", e)
|
170 |
+
raise
|
171 |
+
|
172 |
+
|
173 |
def generate_topics(dataset, config, split, column, nested_column, plot_type):
|
174 |
logging.info(
|
175 |
f"Generating topics for {dataset} with config {config} {split} {column} {nested_column}"
|
|
|
184 |
|
185 |
reduce_umap_model = UMAP(
|
186 |
n_neighbors=n_neighbors,
|
187 |
+
n_components=2, # For visualization, keeping it for 2D
|
188 |
min_dist=0.0,
|
189 |
metric="cosine",
|
190 |
random_state=42,
|
|
|
208 |
gr.DataFrame(value=[], interactive=False, visible=True),
|
209 |
gr.Plot(value=None, visible=True),
|
210 |
gr.Label({message: rows_processed / limit}, visible=True),
|
211 |
+
"",
|
212 |
)
|
213 |
while offset < limit:
|
214 |
docs = get_docs_from_parquet(parquet_urls, column, offset, CHUNK_SIZE)
|
|
|
242 |
topics_info = base_model.get_topic_info()
|
243 |
all_topics, _ = base_model.transform(all_docs)
|
244 |
all_topics = np.array(all_topics)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
245 |
|
246 |
topic_plot = (
|
247 |
base_model.visualize_document_datamap(
|
248 |
docs=all_docs,
|
249 |
reduced_embeddings=reduced_embeddings_array,
|
250 |
+
title=dataset,
|
251 |
+
width=800,
|
252 |
+
height=700,
|
253 |
+
# arrowprops={
|
254 |
+
# "arrowstyle": "wedge,tail_width=0.5",
|
255 |
+
# "connectionstyle": "arc3,rad=0.05",
|
256 |
+
# "linewidth": 0,
|
257 |
+
# "fc": "#33333377",
|
258 |
+
# },
|
259 |
+
label_wrap_width=12,
|
260 |
+
label_over_points=True,
|
261 |
+
dynamic_label_size=True,
|
262 |
+
max_font_size=36,
|
263 |
+
min_font_size=4,
|
264 |
)
|
265 |
if plot_type == "DataMapPlot"
|
266 |
else base_model.visualize_documents(
|
267 |
docs=all_docs,
|
268 |
reduced_embeddings=reduced_embeddings_array,
|
269 |
custom_labels=True,
|
270 |
+
title=dataset,
|
271 |
)
|
272 |
)
|
273 |
|
|
|
285 |
topics_info,
|
286 |
topic_plot,
|
287 |
gr.Label({message: progress}, visible=True),
|
288 |
+
"",
|
289 |
)
|
290 |
|
291 |
offset += CHUNK_SIZE
|
292 |
|
293 |
logging.info("Finished processing all data")
|
294 |
|
295 |
+
plot_png = f"{dataset.replace('/', '-')}-{plot_type.lower()}.png"
|
296 |
+
if plot_type == "DataMapPlot":
|
297 |
+
topic_plot.savefig(plot_png, format="png", dpi=300)
|
298 |
+
else:
|
299 |
+
topic_plot.write_image(plot_png)
|
300 |
+
|
301 |
+
_push_to_hub(dataset, plot_png)
|
302 |
+
plot_png_link = (
|
303 |
+
f"https://huggingface.co/datasets/{EXPORTS_REPOSITORY}/blob/main/{plot_png}"
|
304 |
+
)
|
305 |
yield (
|
306 |
gr.Accordion(open=False),
|
307 |
topics_info,
|
|
|
309 |
gr.Label(
|
310 |
{f"✅ Done: {rows_processed} rows have been processed": 1.0}, visible=True
|
311 |
),
|
312 |
+
f"[![Download as PNG](https://img.shields.io/badge/Download_as-PNG-red)]({plot_png_link})",
|
313 |
)
|
314 |
cuda.empty_cache()
|
315 |
|
|
|
350 |
)
|
351 |
plot_type_radio = gr.Radio(
|
352 |
["DataMapPlot", "Plotly"],
|
353 |
+
value="DataMapPlot",
|
354 |
label="Choose the plot type",
|
355 |
interactive=True,
|
356 |
)
|
|
|
358 |
|
359 |
gr.Markdown("## Data map")
|
360 |
full_topics_generation_label = gr.Label(visible=False, show_label=False)
|
361 |
+
open_png_label = gr.Markdown()
|
362 |
topics_plot = gr.Plot()
|
363 |
with gr.Accordion("Topics Info", open=False):
|
364 |
topics_df = gr.DataFrame(interactive=False, visible=True)
|
|
|
377 |
topics_df,
|
378 |
topics_plot,
|
379 |
full_topics_generation_label,
|
380 |
+
open_png_label,
|
381 |
],
|
382 |
)
|
383 |
|
requirements.txt
CHANGED
@@ -12,3 +12,4 @@ pandas
|
|
12 |
torch
|
13 |
numpy
|
14 |
python-dotenv
|
|
|
|
12 |
torch
|
13 |
numpy
|
14 |
python-dotenv
|
15 |
+
kaleido
|