File size: 17,108 Bytes
bb95205
fe5ff1b
8bd698c
 
 
 
 
 
fd3f0f2
 
 
 
8bd698c
 
 
 
 
 
 
 
 
 
fd3f0f2
8bd698c
 
 
fd3f0f2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8bd698c
 
 
 
fd3f0f2
8bd698c
 
 
fd3f0f2
8bd698c
 
fd3f0f2
8bd698c
 
 
fd3f0f2
 
8bd698c
 
fd3f0f2
8bd698c
 
fd3f0f2
8bd698c
 
 
fd3f0f2
 
8bd698c
 
 
 
 
 
 
 
 
fd3f0f2
8bd698c
 
 
 
 
 
 
 
 
 
 
36b7e3f
8bd698c
 
 
98fe021
fd3f0f2
 
8bd698c
 
 
 
 
 
 
fd3f0f2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8bd698c
fd3f0f2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8bd698c
fd3f0f2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8bd698c
fd3f0f2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8bd698c
 
 
 
 
 
fd3f0f2
8bd698c
 
 
fd3f0f2
8bd698c
 
 
fd3f0f2
 
 
 
36b7e3f
fd3f0f2
 
 
 
8bd698c
 
 
 
fd3f0f2
8bd698c
fd3f0f2
bb95205
 
8bd698c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
fd3f0f2
8bd698c
 
 
 
 
 
 
 
 
c5a0831
bb95205
a9fb05a
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
import gradio as gr
import pandas as pd
from datasets import load_dataset, get_dataset_split_names
from huggingface_hub import HfApi
import os
import pathlib
import uuid
import logging
import threading
import time
import socket
import uvicorn

# --- Setup Logging ---
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')

# --- Embedding Atlas Imports ---
from embedding_atlas.data_source import DataSource
from embedding_atlas.server import make_server
from embedding_atlas.projection import compute_text_projection
from embedding_atlas.utils import Hasher

# --- Helper functions ---
def find_column_name(existing_names, candidate):
    if candidate not in existing_names:
        return candidate
    index = 1
    while True:
        s = f"{candidate}_{index}"
        if s not in existing_names:
            return s
        index += 1

def find_available_port(start_port: int, max_attempts: int = 100):
    """Finds an available TCP port on the host."""
    for port in range(start_port, start_port + max_attempts):
        with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
            if s.connect_ex(('127.0.0.1', port)) != 0:
                logging.info(f"Found available port: {port}")
                return port
    raise RuntimeError("Could not find an available port.")

def run_atlas_server(app, port):
    """Target function for the background thread to run the Uvicorn server."""
    logging.info(f"Starting Atlas server on http://127.0.0.1:{port}")
    uvicorn.run(app, host="127.0.0.1", port=port, log_level="warning")

# --- Hugging Face API Helpers ---
hf_api = HfApi()

def get_user_datasets(username: str):
    logging.info(f"Fetching datasets for user: {username}")
    if not username: return gr.update(choices=[], value=None, interactive=False)
    try:
        datasets = hf_api.list_datasets(author=username, full=True)
        dataset_ids = [d.id for d in datasets if not d.private]
        logging.info(f"Found {len(dataset_ids)} datasets.")
        return gr.update(choices=sorted(dataset_ids), value=None, interactive=True)
    except Exception as e:
        logging.error(f"Failed to fetch datasets: {e}")
        return gr.update(choices=[], value=None, interactive=False)

def get_dataset_splits(dataset_id: str):
    logging.info(f"Fetching splits for: {dataset_id}")
    if not dataset_id: return gr.update(choices=[], value=None, interactive=False)
    try:
        splits = get_dataset_split_names(dataset_id)
        logging.info(f"Found splits: {splits}")
        return gr.update(choices=splits, value=splits[0] if splits else None, interactive=True)
    except Exception as e:
        logging.error(f"Failed to fetch splits: {e}")
        return gr.update(choices=[], value=None, interactive=False)

def get_split_columns(dataset_id: str, split: str):
    logging.info(f"Fetching columns for: {dataset_id}/{split}")
    if not dataset_id or not split: return gr.update(choices=[], value=None, interactive=False)
    try:
        dataset_sample = load_dataset(dataset_id, split=split, streaming=True)
        first_row = next(iter(dataset_sample))
        columns = list(first_row.keys())
        logging.info(f"Found columns: {columns}")
        preferred_cols = ['text', 'content', 'instruction', 'question', 'document', 'prompt']
        best_col = next((col for col in preferred_cols if col in columns), columns[0] if columns else None)
        return gr.update(choices=columns, value=best_col, interactive=True)
    except Exception as e:
        logging.error(f"Failed to get columns: {e}", exc_info=True)
        return gr.update(choices=[], value=None, interactive=False)

# --- Main Atlas Generation Logic ---
def generate_atlas(
    dataset_name: str,
    split: str,
    text_column: str,
    sample_size: int,
    model_name: str,
    umap_neighbors: int,
    umap_min_dist: float,
    progress=gr.Progress(track_tqdm=True)
):
    if not all([dataset_name, split, text_column]):
        raise gr.Error("Please ensure a Dataset, Split, and Text Column are selected.")
        
    progress(0, desc="Loading dataset...")
    df = load_dataset(dataset_name, split=split).to_pandas()
    if sample_size > 0 and sample_size < len(df):
        df = df.sample(n=sample_size, random_state=42).reset_index(drop=True)
    
    progress(0.2, desc="Computing embeddings and UMAP...")
    x_col = find_column_name(df.columns, "projection_x")
    y_col = find_column_name(df.columns, "projection_y")
    neighbors_col = find_column_name(df.columns, "__neighbors")
    compute_text_projection(
        df, text_column, x=x_col, y=y_col, neighbors=neighbors_col, model=model_name,
        umap_args={"n_neighbors": umap_neighbors, "min_dist": umap_min_dist, "metric": "cosine", "random_state": 42},
    )

    progress(0.8, desc="Preparing Atlas data source...")
    id_col = find_column_name(df.columns, "_row_index")
    df[id_col] = range(df.shape[0])
    metadata = {"columns": {"id": id_col, "text": text_column, "embedding": {"x": x_col, "y": y_col}, "neighbors": neighbors_col}}
    hasher = Hasher()
    hasher.update(f"{dataset_name}-{split}-{text_column}-{sample_size}-{model_name}-{uuid.uuid4()}")
    identifier = hasher.hexdigest()
    atlas_dataset = DataSource(identifier, df, metadata)
    
    progress(0.9, desc="Starting Atlas server...")
    static_path = str((pathlib.Path(__import__('embedding_atlas').__file__).parent / "static").resolve())
    atlas_app = make_server(atlas_dataset, static_path=static_path, duckdb_uri="wasm")
    
    # Find an open port and run the server in a background thread
    port = find_available_port(start_port=8001)
    thread = threading.Thread(target=run_atlas_server, args=(atlas_app, port), daemon=True)
    thread.start()
    
    # Give the server a moment to start up
    time.sleep(2) 

    iframe_html = f"<iframe src='http://127.0.0.1:{port}' width='100%' height='800px' frameborder='0'></iframe>"
    return gr.HTML(iframe_html)

# --- Gradio UI Definition ---
with gr.Blocks(theme=gr.themes.Soft(), title="Embedding Atlas Explorer") as app:
    # UI elements...
    gr.Markdown("# Embedding Atlas Explorer")
    # ... (rest of the UI is the same as before) ...
    with gr.Row():
        with gr.Column(scale=1):
            gr.Markdown("### 1. Select Data")
            hf_user_input = gr.Textbox(label="Hugging Face User or Org Name", value="Trendyol", placeholder="e.g., 'gradio' or 'google'")
            dataset_input = gr.Dropdown(label="Select a Dataset", interactive=False)
            split_input = gr.Dropdown(label="Select a Split", interactive=False)
            text_column_input = gr.Dropdown(label="Select a Text Column", interactive=False)
            
            gr.Markdown("### 2. Configure Visualization")
            sample_size_input = gr.Slider(label="Number of Samples", minimum=0, maximum=10000, value=2000, step=100)
            
            with gr.Accordion("Advanced Settings", open=False):
                model_input = gr.Dropdown(label="Embedding Model", choices=["all-MiniLM-L6-v2", "all-mpnet-base-v2", "multi-qa-MiniLM-L6-cos-v1"], value="all-MiniLM-L6-v2")
                umap_neighbors_input = gr.Slider(label="UMAP Neighbors", minimum=2, maximum=100, value=15, step=1, info="Controls local vs. global structure.")
                umap_min_dist_input = gr.Slider(label="UMAP Min Distance", minimum=0.0, maximum=0.99, value=0.1, step=0.01, info="Controls how tightly points are packed.")

            generate_button = gr.Button("Generate Atlas", variant="primary")

        with gr.Column(scale=3):
            gr.Markdown("### 3. Explore Atlas")
            output_html = gr.HTML("<div style='display:flex; justify-content:center; align-items:center; height:800px; border: 1px solid #ddd; border-radius: 5px;'><p>Atlas will be displayed here after generation.</p></div>")

    # --- Event Listeners ---
    hf_user_input.submit(fn=get_user_datasets, inputs=hf_user_input, outputs=dataset_input)
    dataset_input.change(fn=get_dataset_splits, inputs=dataset_input, outputs=split_input)
    split_input.change(fn=get_split_columns, inputs=[dataset_input, split_input], outputs=text_column_input)
    generate_button.click(
        fn=generate_atlas,
        inputs=[dataset_input, split_input, text_column_input, sample_size_input, model_input, umap_neighbors_input, umap_min_dist_input],
        outputs=[output_html],
    )
    app.load(fn=get_user_datasets, inputs=hf_user_input, outputs=dataset_input)

if __name__ == "__main__":
    app.launch(debug=True)import gradio as gr
import pandas as pd
from datasets import load_dataset, get_dataset_split_names
from huggingface_hub import HfApi
import os
import pathlib
import uuid
import logging
import threading
import time
import socket
import uvicorn

# --- Setup Logging ---
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')

# --- Embedding Atlas Imports ---
from embedding_atlas.data_source import DataSource
from embedding_atlas.server import make_server
from embedding_atlas.projection import compute_text_projection
from embedding_atlas.utils import Hasher

# --- Helper functions ---
def find_column_name(existing_names, candidate):
    if candidate not in existing_names:
        return candidate
    index = 1
    while True:
        s = f"{candidate}_{index}"
        if s not in existing_names:
            return s
        index += 1

def find_available_port(start_port: int, max_attempts: int = 100):
    """Finds an available TCP port on the host."""
    for port in range(start_port, start_port + max_attempts):
        with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
            if s.connect_ex(('127.0.0.1', port)) != 0:
                logging.info(f"Found available port: {port}")
                return port
    raise RuntimeError("Could not find an available port.")

def run_atlas_server(app, port):
    """Target function for the background thread to run the Uvicorn server."""
    logging.info(f"Starting Atlas server on http://127.0.0.1:{port}")
    uvicorn.run(app, host="127.0.0.1", port=port, log_level="warning")

# --- Hugging Face API Helpers ---
hf_api = HfApi()

def get_user_datasets(username: str):
    logging.info(f"Fetching datasets for user: {username}")
    if not username: return gr.update(choices=[], value=None, interactive=False)
    try:
        datasets = hf_api.list_datasets(author=username, full=True)
        dataset_ids = [d.id for d in datasets if not d.private]
        logging.info(f"Found {len(dataset_ids)} datasets.")
        return gr.update(choices=sorted(dataset_ids), value=None, interactive=True)
    except Exception as e:
        logging.error(f"Failed to fetch datasets: {e}")
        return gr.update(choices=[], value=None, interactive=False)

def get_dataset_splits(dataset_id: str):
    logging.info(f"Fetching splits for: {dataset_id}")
    if not dataset_id: return gr.update(choices=[], value=None, interactive=False)
    try:
        splits = get_dataset_split_names(dataset_id)
        logging.info(f"Found splits: {splits}")
        return gr.update(choices=splits, value=splits[0] if splits else None, interactive=True)
    except Exception as e:
        logging.error(f"Failed to fetch splits: {e}")
        return gr.update(choices=[], value=None, interactive=False)

def get_split_columns(dataset_id: str, split: str):
    logging.info(f"Fetching columns for: {dataset_id}/{split}")
    if not dataset_id or not split: return gr.update(choices=[], value=None, interactive=False)
    try:
        dataset_sample = load_dataset(dataset_id, split=split, streaming=True)
        first_row = next(iter(dataset_sample))
        columns = list(first_row.keys())
        logging.info(f"Found columns: {columns}")
        preferred_cols = ['text', 'content', 'instruction', 'question', 'document', 'prompt']
        best_col = next((col for col in preferred_cols if col in columns), columns[0] if columns else None)
        return gr.update(choices=columns, value=best_col, interactive=True)
    except Exception as e:
        logging.error(f"Failed to get columns: {e}", exc_info=True)
        return gr.update(choices=[], value=None, interactive=False)

# --- Main Atlas Generation Logic ---
def generate_atlas(
    dataset_name: str,
    split: str,
    text_column: str,
    sample_size: int,
    model_name: str,
    umap_neighbors: int,
    umap_min_dist: float,
    progress=gr.Progress(track_tqdm=True)
):
    if not all([dataset_name, split, text_column]):
        raise gr.Error("Please ensure a Dataset, Split, and Text Column are selected.")
        
    progress(0, desc="Loading dataset...")
    df = load_dataset(dataset_name, split=split).to_pandas()
    if sample_size > 0 and sample_size < len(df):
        df = df.sample(n=sample_size, random_state=42).reset_index(drop=True)
    
    progress(0.2, desc="Computing embeddings and UMAP...")
    x_col = find_column_name(df.columns, "projection_x")
    y_col = find_column_name(df.columns, "projection_y")
    neighbors_col = find_column_name(df.columns, "__neighbors")
    compute_text_projection(
        df, text_column, x=x_col, y=y_col, neighbors=neighbors_col, model=model_name,
        umap_args={"n_neighbors": umap_neighbors, "min_dist": umap_min_dist, "metric": "cosine", "random_state": 42},
    )

    progress(0.8, desc="Preparing Atlas data source...")
    id_col = find_column_name(df.columns, "_row_index")
    df[id_col] = range(df.shape[0])
    metadata = {"columns": {"id": id_col, "text": text_column, "embedding": {"x": x_col, "y": y_col}, "neighbors": neighbors_col}}
    hasher = Hasher()
    hasher.update(f"{dataset_name}-{split}-{text_column}-{sample_size}-{model_name}-{uuid.uuid4()}")
    identifier = hasher.hexdigest()
    atlas_dataset = DataSource(identifier, df, metadata)
    
    progress(0.9, desc="Starting Atlas server...")
    static_path = str((pathlib.Path(__import__('embedding_atlas').__file__).parent / "static").resolve())
    atlas_app = make_server(atlas_dataset, static_path=static_path, duckdb_uri="wasm")
    
    # Find an open port and run the server in a background thread
    port = find_available_port(start_port=8001)
    thread = threading.Thread(target=run_atlas_server, args=(atlas_app, port), daemon=True)
    thread.start()
    
    # Give the server a moment to start up
    time.sleep(2) 

    iframe_html = f"<iframe src='http://127.0.0.1:{port}' width='100%' height='800px' frameborder='0'></iframe>"
    return gr.HTML(iframe_html)

# --- Gradio UI Definition ---
with gr.Blocks(theme=gr.themes.Soft(), title="Embedding Atlas Explorer") as app:
    # UI elements...
    gr.Markdown("# Embedding Atlas Explorer")
    # ... (rest of the UI is the same as before) ...
    with gr.Row():
        with gr.Column(scale=1):
            gr.Markdown("### 1. Select Data")
            hf_user_input = gr.Textbox(label="Hugging Face User or Org Name", value="Trendyol", placeholder="e.g., 'gradio' or 'google'")
            dataset_input = gr.Dropdown(label="Select a Dataset", interactive=False)
            split_input = gr.Dropdown(label="Select a Split", interactive=False)
            text_column_input = gr.Dropdown(label="Select a Text Column", interactive=False)
            
            gr.Markdown("### 2. Configure Visualization")
            sample_size_input = gr.Slider(label="Number of Samples", minimum=0, maximum=10000, value=2000, step=100)
            
            with gr.Accordion("Advanced Settings", open=False):
                model_input = gr.Dropdown(label="Embedding Model", choices=["all-MiniLM-L6-v2", "all-mpnet-base-v2", "multi-qa-MiniLM-L6-cos-v1"], value="all-MiniLM-L6-v2")
                umap_neighbors_input = gr.Slider(label="UMAP Neighbors", minimum=2, maximum=100, value=15, step=1, info="Controls local vs. global structure.")
                umap_min_dist_input = gr.Slider(label="UMAP Min Distance", minimum=0.0, maximum=0.99, value=0.1, step=0.01, info="Controls how tightly points are packed.")

            generate_button = gr.Button("Generate Atlas", variant="primary")

        with gr.Column(scale=3):
            gr.Markdown("### 3. Explore Atlas")
            output_html = gr.HTML("<div style='display:flex; justify-content:center; align-items:center; height:800px; border: 1px solid #ddd; border-radius: 5px;'><p>Atlas will be displayed here after generation.</p></div>")

    # --- Event Listeners ---
    hf_user_input.submit(fn=get_user_datasets, inputs=hf_user_input, outputs=dataset_input)
    dataset_input.change(fn=get_dataset_splits, inputs=dataset_input, outputs=split_input)
    split_input.change(fn=get_split_columns, inputs=[dataset_input, split_input], outputs=text_column_input)
    generate_button.click(
        fn=generate_atlas,
        inputs=[dataset_input, split_input, text_column_input, sample_size_input, model_input, umap_neighbors_input, umap_min_dist_input],
        outputs=[output_html],
    )
    app.load(fn=get_user_datasets, inputs=hf_user_input, outputs=dataset_input)

if __name__ == "__main__":
    app.launch(debug=True)