Spaces:
Running
Running
fix to have baseline run from the runs table
Browse files- app.py +53 -20
- data_access.py +12 -11
- load_ground_truth.py +0 -0
- eval_tables.py → scripts/eval_tables.py +0 -0
app.py
CHANGED
@@ -1,8 +1,8 @@
|
|
1 |
import asyncio
|
|
|
2 |
|
3 |
import gradio as gr
|
4 |
import pandas as pd
|
5 |
-
import logging
|
6 |
|
7 |
from data_access import get_questions, get_source_finders, get_run_ids, get_baseline_rankers, \
|
8 |
get_unified_sources, get_source_text, calculate_cumulative_statistics_for_all_questions, get_metadata, \
|
@@ -10,6 +10,8 @@ from data_access import get_questions, get_source_finders, get_run_ids, get_base
|
|
10 |
|
11 |
logger = logging.getLogger(__name__)
|
12 |
|
|
|
|
|
13 |
# Initialize data at the module level
|
14 |
questions = []
|
15 |
source_finders = []
|
@@ -22,9 +24,11 @@ run_ids = []
|
|
22 |
available_run_id_dict = {}
|
23 |
finder_options = []
|
24 |
previous_run_id = "initial_run"
|
|
|
25 |
|
26 |
run_id_dropdown = None
|
27 |
|
|
|
28 |
# Get all questions
|
29 |
|
30 |
# Initialize data in a single async function
|
@@ -36,7 +40,6 @@ async def initialize_data():
|
|
36 |
source_finders = await get_source_finders(conn)
|
37 |
baseline_rankers = await get_baseline_rankers(conn)
|
38 |
|
39 |
-
baseline_rankers_dict = {f["name"]: f["id"] for f in baseline_rankers}
|
40 |
# Convert to dictionaries for easier lookup
|
41 |
questions_dict = {q["text"]: q["id"] for q in questions}
|
42 |
baseline_rankers_dict = {f["name"]: f["id"] for f in baseline_rankers}
|
@@ -46,9 +49,32 @@ async def initialize_data():
|
|
46 |
question_options = [q['text'] for q in questions]
|
47 |
finder_options = [s["name"] for s in source_finders]
|
48 |
baseline_ranker_options = [b["name"] for b in baseline_rankers]
|
|
|
|
|
49 |
|
|
|
|
|
50 |
|
51 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
52 |
global previous_run_id
|
53 |
if evt:
|
54 |
logger.info(f"event: {evt.target.elem_id}")
|
@@ -70,27 +96,30 @@ async def update_sources_list_async(question_option, source_finder_name, run_id,
|
|
70 |
if type(baseline_ranker_name) == list:
|
71 |
baseline_ranker_name = baseline_ranker_name[0]
|
72 |
|
73 |
-
baseline_ranker_id_int = 1 if len(baseline_ranker_name) == 0 else baseline_rankers_dict.get(
|
|
|
74 |
|
75 |
if len(source_finder_name):
|
76 |
finder_id_int = source_finders_dict.get(source_finder_name)
|
77 |
else:
|
78 |
finder_id_int = None
|
79 |
|
80 |
-
if question_option ==
|
81 |
if finder_id_int:
|
82 |
if run_id is None:
|
83 |
available_run_id_dict = await get_run_ids(conn, finder_id_int)
|
84 |
run_id = list(available_run_id_dict.keys())[0]
|
85 |
previous_run_id = run_id
|
86 |
run_id_int = available_run_id_dict.get(run_id)
|
87 |
-
all_stats = await calculate_cumulative_statistics_for_all_questions(conn, run_id_int,
|
|
|
88 |
|
89 |
else:
|
90 |
run_id_options = list(available_run_id_dict.keys())
|
91 |
all_stats = None
|
92 |
run_id_options = list(available_run_id_dict.keys())
|
93 |
-
return None, all_stats, gr.Dropdown(choices=run_id_options,
|
|
|
94 |
|
95 |
# Extract question ID from selection
|
96 |
question_id = questions_dict.get(question_option)
|
@@ -102,8 +131,6 @@ async def update_sources_list_async(question_option, source_finder_name, run_id,
|
|
102 |
previous_run_id = run_id
|
103 |
run_id_int = available_run_id_dict.get(run_id)
|
104 |
|
105 |
-
|
106 |
-
|
107 |
source_runs = None
|
108 |
stats = None
|
109 |
# Get source runs data
|
@@ -116,7 +143,8 @@ async def update_sources_list_async(question_option, source_finder_name, run_id,
|
|
116 |
return None, None, run_id_options, "No results found for the selected filters",
|
117 |
|
118 |
# Format table columns
|
119 |
-
columns_to_display = ['sugya_id', 'in_baseline', 'baseline_rank', 'in_source_run', 'source_run_rank',
|
|
|
120 |
'folio', 'reason']
|
121 |
df_display = df[columns_to_display] if all(col in df.columns for col in columns_to_display) else df
|
122 |
|
@@ -147,6 +175,7 @@ async def handle_row_selection_async(evt: gr.SelectData):
|
|
147 |
def handle_row_selection(evt: gr.SelectData):
|
148 |
return asyncio.run(handle_row_selection_async(evt))
|
149 |
|
|
|
150 |
# Create Gradio app
|
151 |
|
152 |
# Ensure we clean up when done
|
@@ -162,7 +191,7 @@ async def main():
|
|
162 |
with gr.Column(scale=1):
|
163 |
# Main content area
|
164 |
question_dropdown = gr.Dropdown(
|
165 |
-
choices=[
|
166 |
label="Select Question",
|
167 |
value=None,
|
168 |
interactive=True,
|
@@ -186,7 +215,7 @@ async def main():
|
|
186 |
)
|
187 |
with gr.Column(scale=1):
|
188 |
run_id_dropdown = gr.Dropdown(
|
189 |
-
choices=
|
190 |
allow_custom_value=True,
|
191 |
label="Run id for Question and source finder",
|
192 |
interactive=True,
|
@@ -201,7 +230,6 @@ async def main():
|
|
201 |
gr.Markdown(f"Total Questions: {len(questions)}")
|
202 |
gr.Markdown(f"Source Finders: {len(source_finders)}")
|
203 |
|
204 |
-
|
205 |
with gr.Row():
|
206 |
result_text = gr.Markdown("Select a question to view source runs")
|
207 |
with gr.Row():
|
@@ -221,14 +249,15 @@ async def main():
|
|
221 |
metadata_text = gr.TextArea(
|
222 |
label="Metadata of Source Finder for Selected Question",
|
223 |
elem_id="metadata",
|
224 |
-
lines
|
225 |
)
|
226 |
with gr.Row():
|
227 |
gr.Markdown("# Sources Found")
|
228 |
with gr.Row():
|
229 |
with gr.Column(scale=3):
|
230 |
results_table = gr.DataFrame(
|
231 |
-
headers=['id', 'tractate', 'folio', 'in_baseline', 'baseline_rank', 'in_source_run',
|
|
|
232 |
interactive=False
|
233 |
)
|
234 |
with gr.Column(scale=1):
|
@@ -246,8 +275,6 @@ async def main():
|
|
246 |
# visible=True
|
247 |
# )
|
248 |
|
249 |
-
|
250 |
-
|
251 |
# Set up event handlers
|
252 |
results_table.select(
|
253 |
handle_row_selection,
|
@@ -255,15 +282,22 @@ async def main():
|
|
255 |
outputs=source_text
|
256 |
)
|
257 |
|
258 |
-
|
259 |
update_sources_list,
|
260 |
inputs=[question_dropdown, source_finder_dropdown, run_id_dropdown, baseline_rankers_dropdown],
|
261 |
outputs=[results_table, statistics_table, run_id_dropdown, result_text, metadata_text]
|
|
|
262 |
)
|
263 |
|
264 |
-
|
265 |
update_sources_list,
|
266 |
inputs=[question_dropdown, source_finder_dropdown, run_id_dropdown, baseline_rankers_dropdown],
|
|
|
|
|
|
|
|
|
|
|
|
|
267 |
# outputs=[run_id_dropdown, results_table, result_text, download_button]
|
268 |
outputs=[results_table, statistics_table, run_id_dropdown, result_text, metadata_text]
|
269 |
)
|
@@ -274,7 +308,6 @@ async def main():
|
|
274 |
outputs=[results_table, statistics_table, run_id_dropdown, result_text, metadata_text]
|
275 |
)
|
276 |
|
277 |
-
|
278 |
app.queue()
|
279 |
app.launch()
|
280 |
|
|
|
1 |
import asyncio
|
2 |
+
import logging
|
3 |
|
4 |
import gradio as gr
|
5 |
import pandas as pd
|
|
|
6 |
|
7 |
from data_access import get_questions, get_source_finders, get_run_ids, get_baseline_rankers, \
|
8 |
get_unified_sources, get_source_text, calculate_cumulative_statistics_for_all_questions, get_metadata, \
|
|
|
10 |
|
11 |
logger = logging.getLogger(__name__)
|
12 |
|
13 |
+
ALL_QUESTIONS_STR = "All questions"
|
14 |
+
|
15 |
# Initialize data at the module level
|
16 |
questions = []
|
17 |
source_finders = []
|
|
|
24 |
available_run_id_dict = {}
|
25 |
finder_options = []
|
26 |
previous_run_id = "initial_run"
|
27 |
+
run_id_options = []
|
28 |
|
29 |
run_id_dropdown = None
|
30 |
|
31 |
+
|
32 |
# Get all questions
|
33 |
|
34 |
# Initialize data in a single async function
|
|
|
40 |
source_finders = await get_source_finders(conn)
|
41 |
baseline_rankers = await get_baseline_rankers(conn)
|
42 |
|
|
|
43 |
# Convert to dictionaries for easier lookup
|
44 |
questions_dict = {q["text"]: q["id"] for q in questions}
|
45 |
baseline_rankers_dict = {f["name"]: f["id"] for f in baseline_rankers}
|
|
|
49 |
question_options = [q['text'] for q in questions]
|
50 |
finder_options = [s["name"] for s in source_finders]
|
51 |
baseline_ranker_options = [b["name"] for b in baseline_rankers]
|
52 |
+
update_run_ids(ALL_QUESTIONS_STR, list(source_finders_dict.keys())[0])
|
53 |
+
|
54 |
|
55 |
+
def update_run_ids(question_option, source_finder_name):
|
56 |
+
return asyncio.run(update_run_ids_async(question_option, source_finder_name))
|
57 |
|
58 |
+
|
59 |
+
async def update_run_ids_async(question_option, source_finder_name):
|
60 |
+
global previous_run_id, available_run_id_dict, run_id_options
|
61 |
+
async with get_async_connection() as conn:
|
62 |
+
finder_id_int = source_finders_dict.get(source_finder_name)
|
63 |
+
if question_option and question_option != ALL_QUESTIONS_STR:
|
64 |
+
question_id = questions_dict.get(question_option)
|
65 |
+
available_run_id_dict = await get_run_ids(conn, finder_id_int, question_id)
|
66 |
+
else:
|
67 |
+
available_run_id_dict = await get_run_ids(conn, finder_id_int)
|
68 |
+
|
69 |
+
|
70 |
+
run_id = list(available_run_id_dict.keys())[0]
|
71 |
+
previous_run_id = run_id
|
72 |
+
run_id_options = list(available_run_id_dict.keys())
|
73 |
+
return None, None, gr.Dropdown(choices=run_id_options,
|
74 |
+
value=run_id), "Select Question to see results", ""
|
75 |
+
|
76 |
+
def update_sources_list(question_option, source_finder_id, run_id: str, baseline_ranker_id: str,
|
77 |
+
evt: gr.EventData = None):
|
78 |
global previous_run_id
|
79 |
if evt:
|
80 |
logger.info(f"event: {evt.target.elem_id}")
|
|
|
96 |
if type(baseline_ranker_name) == list:
|
97 |
baseline_ranker_name = baseline_ranker_name[0]
|
98 |
|
99 |
+
baseline_ranker_id_int = 1 if len(baseline_ranker_name) == 0 else baseline_rankers_dict.get(
|
100 |
+
baseline_ranker_name)
|
101 |
|
102 |
if len(source_finder_name):
|
103 |
finder_id_int = source_finders_dict.get(source_finder_name)
|
104 |
else:
|
105 |
finder_id_int = None
|
106 |
|
107 |
+
if question_option == ALL_QUESTIONS_STR:
|
108 |
if finder_id_int:
|
109 |
if run_id is None:
|
110 |
available_run_id_dict = await get_run_ids(conn, finder_id_int)
|
111 |
run_id = list(available_run_id_dict.keys())[0]
|
112 |
previous_run_id = run_id
|
113 |
run_id_int = available_run_id_dict.get(run_id)
|
114 |
+
all_stats = await calculate_cumulative_statistics_for_all_questions(conn, run_id_int,
|
115 |
+
baseline_ranker_id_int)
|
116 |
|
117 |
else:
|
118 |
run_id_options = list(available_run_id_dict.keys())
|
119 |
all_stats = None
|
120 |
run_id_options = list(available_run_id_dict.keys())
|
121 |
+
return None, all_stats, gr.Dropdown(choices=run_id_options,
|
122 |
+
value=run_id), "Select Run Id and source finder to see results", ""
|
123 |
|
124 |
# Extract question ID from selection
|
125 |
question_id = questions_dict.get(question_option)
|
|
|
131 |
previous_run_id = run_id
|
132 |
run_id_int = available_run_id_dict.get(run_id)
|
133 |
|
|
|
|
|
134 |
source_runs = None
|
135 |
stats = None
|
136 |
# Get source runs data
|
|
|
143 |
return None, None, run_id_options, "No results found for the selected filters",
|
144 |
|
145 |
# Format table columns
|
146 |
+
columns_to_display = ['sugya_id', 'in_baseline', 'baseline_rank', 'in_source_run', 'source_run_rank',
|
147 |
+
'tractate',
|
148 |
'folio', 'reason']
|
149 |
df_display = df[columns_to_display] if all(col in df.columns for col in columns_to_display) else df
|
150 |
|
|
|
175 |
def handle_row_selection(evt: gr.SelectData):
|
176 |
return asyncio.run(handle_row_selection_async(evt))
|
177 |
|
178 |
+
|
179 |
# Create Gradio app
|
180 |
|
181 |
# Ensure we clean up when done
|
|
|
191 |
with gr.Column(scale=1):
|
192 |
# Main content area
|
193 |
question_dropdown = gr.Dropdown(
|
194 |
+
choices=[ALL_QUESTIONS_STR] + question_options,
|
195 |
label="Select Question",
|
196 |
value=None,
|
197 |
interactive=True,
|
|
|
215 |
)
|
216 |
with gr.Column(scale=1):
|
217 |
run_id_dropdown = gr.Dropdown(
|
218 |
+
choices=run_id_options,
|
219 |
allow_custom_value=True,
|
220 |
label="Run id for Question and source finder",
|
221 |
interactive=True,
|
|
|
230 |
gr.Markdown(f"Total Questions: {len(questions)}")
|
231 |
gr.Markdown(f"Source Finders: {len(source_finders)}")
|
232 |
|
|
|
233 |
with gr.Row():
|
234 |
result_text = gr.Markdown("Select a question to view source runs")
|
235 |
with gr.Row():
|
|
|
249 |
metadata_text = gr.TextArea(
|
250 |
label="Metadata of Source Finder for Selected Question",
|
251 |
elem_id="metadata",
|
252 |
+
lines=2
|
253 |
)
|
254 |
with gr.Row():
|
255 |
gr.Markdown("# Sources Found")
|
256 |
with gr.Row():
|
257 |
with gr.Column(scale=3):
|
258 |
results_table = gr.DataFrame(
|
259 |
+
headers=['id', 'tractate', 'folio', 'in_baseline', 'baseline_rank', 'in_source_run',
|
260 |
+
'source_run_rank', 'source_reason', 'metadata'],
|
261 |
interactive=False
|
262 |
)
|
263 |
with gr.Column(scale=1):
|
|
|
275 |
# visible=True
|
276 |
# )
|
277 |
|
|
|
|
|
278 |
# Set up event handlers
|
279 |
results_table.select(
|
280 |
handle_row_selection,
|
|
|
282 |
outputs=source_text
|
283 |
)
|
284 |
|
285 |
+
baseline_rankers_dropdown.change(
|
286 |
update_sources_list,
|
287 |
inputs=[question_dropdown, source_finder_dropdown, run_id_dropdown, baseline_rankers_dropdown],
|
288 |
outputs=[results_table, statistics_table, run_id_dropdown, result_text, metadata_text]
|
289 |
+
|
290 |
)
|
291 |
|
292 |
+
question_dropdown.change(
|
293 |
update_sources_list,
|
294 |
inputs=[question_dropdown, source_finder_dropdown, run_id_dropdown, baseline_rankers_dropdown],
|
295 |
+
outputs=[results_table, statistics_table, run_id_dropdown, result_text, metadata_text]
|
296 |
+
)
|
297 |
+
|
298 |
+
source_finder_dropdown.change(
|
299 |
+
update_run_ids,
|
300 |
+
inputs=[question_dropdown, source_finder_dropdown],
|
301 |
# outputs=[run_id_dropdown, results_table, result_text, download_button]
|
302 |
outputs=[results_table, statistics_table, run_id_dropdown, result_text, metadata_text]
|
303 |
)
|
|
|
308 |
outputs=[results_table, statistics_table, run_id_dropdown, result_text, metadata_text]
|
309 |
)
|
310 |
|
|
|
311 |
app.queue()
|
312 |
app.launch()
|
313 |
|
data_access.py
CHANGED
@@ -15,6 +15,7 @@ load_dotenv()
|
|
15 |
@asynccontextmanager
|
16 |
async def get_async_connection(schema="talmudexplore"):
|
17 |
"""Get a connection for the current request."""
|
|
|
18 |
try:
|
19 |
# Create a single connection without relying on a shared pool
|
20 |
conn = await asyncpg.connect(
|
@@ -27,7 +28,8 @@ async def get_async_connection(schema="talmudexplore"):
|
|
27 |
await conn.execute(f'SET search_path TO {schema}')
|
28 |
yield conn
|
29 |
finally:
|
30 |
-
|
|
|
31 |
|
32 |
|
33 |
async def get_questions(conn: asyncpg.Connection):
|
@@ -73,8 +75,13 @@ async def get_run_ids(conn: asyncpg.Connection, source_finder_id: int, question_
|
|
73 |
|
74 |
|
75 |
async def get_baseline_rankers(conn: asyncpg.Connection):
|
76 |
-
|
77 |
-
|
|
|
|
|
|
|
|
|
|
|
78 |
|
79 |
async def calculate_baseline_vs_source_stats_for_question(conn: asyncpg.Connection, baseline_sources , source_runs_sources):
|
80 |
# for a given question_id and source_finder_id and run_id calculate the baseline vs source stats
|
@@ -203,14 +210,8 @@ async def get_unified_sources(conn: asyncpg.Connection, question_id: int, source
|
|
203 |
"""
|
204 |
source_runs = await conn.fetch(query_runs, question_id, source_finder_run_id)
|
205 |
# Get sources from baseline_sources
|
206 |
-
|
207 |
-
|
208 |
-
FROM baseline_sources bs
|
209 |
-
join talmud_bavli tb on bs.sugya_id = tb.xml_id
|
210 |
-
WHERE bs.question_id = $1
|
211 |
-
AND bs.ranker_id = $2
|
212 |
-
"""
|
213 |
-
baseline_sources = await conn.fetch(query_baseline, question_id, ranker_id)
|
214 |
stats_df = await calculate_baseline_vs_source_stats_for_question(conn, baseline_sources, source_runs)
|
215 |
# Convert to dictionaries for easier lookup
|
216 |
source_runs_dict = {s["id"]: dict(s) for s in source_runs}
|
|
|
15 |
@asynccontextmanager
|
16 |
async def get_async_connection(schema="talmudexplore"):
|
17 |
"""Get a connection for the current request."""
|
18 |
+
conn = None
|
19 |
try:
|
20 |
# Create a single connection without relying on a shared pool
|
21 |
conn = await asyncpg.connect(
|
|
|
28 |
await conn.execute(f'SET search_path TO {schema}')
|
29 |
yield conn
|
30 |
finally:
|
31 |
+
if conn:
|
32 |
+
await conn.close()
|
33 |
|
34 |
|
35 |
async def get_questions(conn: asyncpg.Connection):
|
|
|
75 |
|
76 |
|
77 |
async def get_baseline_rankers(conn: asyncpg.Connection):
|
78 |
+
query = """
|
79 |
+
select sfr.id, sf.source_finder_type, sfr.description from talmudexplore.source_finder_runs sfr
|
80 |
+
join source_finders sf on sf.id = sfr.source_finder_id
|
81 |
+
order by sf.id
|
82 |
+
"""
|
83 |
+
rankers = await conn.fetch(query)
|
84 |
+
return [{"id": r["id"], "name": f"{r['source_finder_type']} : {r['description']}"} for r in rankers]
|
85 |
|
86 |
async def calculate_baseline_vs_source_stats_for_question(conn: asyncpg.Connection, baseline_sources , source_runs_sources):
|
87 |
# for a given question_id and source_finder_id and run_id calculate the baseline vs source stats
|
|
|
210 |
"""
|
211 |
source_runs = await conn.fetch(query_runs, question_id, source_finder_run_id)
|
212 |
# Get sources from baseline_sources
|
213 |
+
baseline_query = query_runs.replace("source_rank", "baseline_rank")
|
214 |
+
baseline_sources = await conn.fetch(baseline_query, question_id, ranker_id)
|
|
|
|
|
|
|
|
|
|
|
|
|
215 |
stats_df = await calculate_baseline_vs_source_stats_for_question(conn, baseline_sources, source_runs)
|
216 |
# Convert to dictionaries for easier lookup
|
217 |
source_runs_dict = {s["id"]: dict(s) for s in source_runs}
|
load_ground_truth.py
DELETED
File without changes
|
eval_tables.py → scripts/eval_tables.py
RENAMED
File without changes
|