Spaces:
Sleeping
Sleeping
File size: 17,033 Bytes
142faac c4c7997 142faac c4c7997 142faac c4c7997 142faac c4c7997 142faac 6190131 142faac fbb993d 142faac |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 |
"""
Streamlit app for human evaluation of model outputs.
Allows users to select two models, compare their responses to the same inputs,
and record preferences for subsequent analysis.
"""
import os
import json
import csv
from datetime import datetime
import streamlit as st
import pandas as pd
import tempfile
st.set_page_config(page_title="Model Comparison Evaluation", layout="wide")
SCRIPT_DIR = os.path.dirname(os.path.abspath(__file__))
DATA_DIR = os.path.join(SCRIPT_DIR, "for_experiments_prediction")
@st.cache_data
def load_models(data_dir):
"""
Discover prediction JSON files named 'predicted_vs_gt.json' and load flattened records for each model.
Returns a dict mapping model name to dict of {id: record}.
"""
model_paths = {}
for root, _, files in os.walk(data_dir):
for fname in files:
if fname == 'predicted_vs_gt.json':
path = os.path.join(root, fname)
rel = os.path.relpath(root, data_dir)
model_paths[rel] = path
models = {}
for model_name, path in sorted(model_paths.items()):
with open(path, 'r', encoding='utf-8') as f:
data = json.load(f)
records = {}
for section in data.values():
if isinstance(section, dict):
for sub in section.values():
for rec in sub:
records[rec['id']] = rec
elif isinstance(section, list):
for rec in section:
records[rec['id']] = rec
models[model_name] = records
return models
def append_feedback(feedback_file, header, row):
"""
Append a single feedback row to TSV, writing header if file does not exist.
"""
write_header = not os.path.exists(feedback_file)
with open(feedback_file, 'a', newline='', encoding='utf-8') as f:
writer = csv.writer(f, delimiter='\t', quoting=csv.QUOTE_ALL)
if write_header:
writer.writerow(header)
writer.writerow(row)
@st.cache_data
def load_eval_tables(data_dir):
"""
Discover evaluation_table.parquet files under each model directory and load each into a pandas DataFrame.
Returns a dict mapping model name to its evaluation DataFrame.
"""
tables = {}
for root, _, files in os.walk(data_dir):
if 'evaluation_table.parquet' in files:
path = os.path.join(root, 'evaluation_table.parquet')
rel = os.path.relpath(root, data_dir)
tables[rel] = pd.read_parquet(path)
return tables
def main():
st.title("Model Comparison Evaluation")
print(DATA_DIR)
models = load_models(DATA_DIR)
eval_tables = load_eval_tables(DATA_DIR)
all_cols = set()
for df in eval_tables.values():
all_cols.update(df.columns)
key_columns = {'gt_sac_id', 'gt_title'}
metric_columns = sorted(all_cols - key_columns)
fixed_metrics = [
'chemicals_accuracy',
'chemicals_f1_score',
'chemicals_precision',
'chemicals_recall',
'metal_accuracy',
'metal_f1_score',
'metal_precision',
'metal_recall',
'procedure_procedure_completeness_score',
'procedure_procedure_order_score',
'procedure_procedure_accuracy_score',
'support_accuracy',
'support_f1_score',
'support_precision',
'support_recall',
]
other_metrics = sorted([c for c in metric_columns if c not in fixed_metrics and not c.startswith('gt_')])
model_names = list(models.keys())
st.sidebar.header("Configuration")
def reset_index():
st.session_state.idx = 0
# Reset feedback file when models change
feedback_path = os.path.join(tempfile.gettempdir(), 'feedback.tsv')
if os.path.exists(feedback_path):
os.remove(feedback_path)
selected = st.sidebar.multiselect(
"Select exactly two models to compare",
options=model_names,
key='models',
help="Choose two model variants for side-by-side comparison",
on_change=reset_index
)
if len(selected) != 2:
st.sidebar.info("Please select exactly two models.")
st.stop()
# Download button for feedback TSV
feedback_path = os.path.join(tempfile.gettempdir(), 'feedback.tsv')
if os.path.exists(feedback_path):
with open(feedback_path, 'r', encoding='utf-8') as f:
tsv_data = f.read()
st.sidebar.download_button(
label="Download Feedback TSV",
data=tsv_data,
file_name="feedback.tsv",
mime="text/tab-separated-values"
)
m1, m2 = selected
recs1 = models[m1]
recs2 = models[m2]
common_ids = sorted(set(recs1.keys()) & set(recs2.keys()))
if not common_ids:
st.error("No common records between the selected models.")
st.stop()
if 'idx' not in st.session_state:
st.session_state.idx = 0
if 'feedback_saved' not in st.session_state:
st.session_state.feedback_saved = False
# Initialize fresh feedback file for new session
if 'session_initialized' not in st.session_state:
feedback_path = os.path.join(tempfile.gettempdir(), 'feedback.tsv')
if os.path.exists(feedback_path):
os.remove(feedback_path)
st.session_state.session_initialized = True
total = len(common_ids)
idx = st.session_state.idx
if idx < 0:
idx = 0
if idx >= total:
st.write("### Evaluation complete! Thank you for your feedback.")
st.stop()
current_id = common_ids[idx]
rec1 = recs1[current_id]
rec2 = recs2[current_id]
st.markdown(f"**Record {idx+1}/{total} — ID: {current_id}**")
st.markdown("---")
st.subheader("Input Prompt")
st.code(rec1.get('input', ''), language='')
st.subheader("Model Responses and Ground Truth")
col1, col2, col3 = st.columns(3)
with col1:
st.markdown(f"**{m1}**")
st.text_area("", rec1.get('predicted', ''), height=600, key=f"resp1_{idx}")
with col2:
st.markdown(f"**{m2}**")
st.text_area("", rec2.get('predicted', ''), height=600, key=f"resp2_{idx}")
with col3:
st.markdown("**Ground Truth**")
st.text_area("", rec1.get('ground_truth', ''), height=600, key=f"gt_{idx}")
fcol1, fcol2, fcol3 = st.columns(3)
with fcol1:
df1 = eval_tables.get(m1)
if df1 is not None:
if 'gt_sac_id' in df1.columns:
key_val = rec1.get('gt_sac_id', rec1.get('sac_id'))
key_col = 'gt_sac_id'
elif 'gt_title' in df1.columns:
key_val = rec1.get('gt_title', rec1.get('title'))
key_col = 'gt_title'
else:
key_col = key_val = None
if key_col and key_val is not None:
row = df1[df1[key_col] == key_val]
if not row.empty:
fm_df = row[fixed_metrics].T
fm_df.columns = ['value']
st.table(fm_df)
else:
st.info("No fixed metrics for this record.")
else:
st.info("No evaluation table available for this model.")
with fcol2:
df2 = eval_tables.get(m2)
if df2 is not None:
if 'gt_sac_id' in df2.columns:
key_val = rec1.get('gt_sac_id', rec1.get('sac_id'))
key_col = 'gt_sac_id'
elif 'gt_title' in df2.columns:
key_val = rec1.get('gt_title', rec1.get('title'))
key_col = 'gt_title'
else:
key_col = key_val = None
if key_col and key_val is not None:
row = df2[df2[key_col] == key_val]
if not row.empty:
fm_df = row[fixed_metrics].T.astype(float).mean(axis=1)
fm_df.columns = ['value']
st.table(fm_df)
else:
st.info("No fixed metrics for this record.")
else:
st.info("No evaluation table available for this model.")
if other_metrics:
selected_metric = st.selectbox(
"Select additional metric to display",
options=other_metrics,
key=f"metric_sel_{idx}"
)
else:
selected_metric = None
if selected_metric:
mcol1, mcol2, mcol3 = st.columns(3)
with mcol1:
df1 = eval_tables.get(m1)
if df1 is not None and selected_metric in df1.columns:
if 'gt_sac_id' in df1.columns:
key_val = rec1.get('gt_sac_id', rec1.get('sac_id'))
key_col = 'gt_sac_id'
elif 'gt_title' in df1.columns:
key_val = rec1.get('gt_title', rec1.get('title'))
key_col = 'gt_title'
else:
key_col = key_val = None
if key_col and key_val is not None:
row = df1[df1[key_col] == key_val]
if not row.empty:
value = row[selected_metric].iloc[0]
try:
# Try to parse as JSON first
parsed_json = json.loads(str(value))
formatted_json = json.dumps(parsed_json, indent=2)
st.markdown(f"**{selected_metric}:**")
st.code(formatted_json, language='json')
except json.JSONDecodeError:
try:
# If JSON fails, try to evaluate as Python literal (handles single quotes)
import ast
parsed_json = ast.literal_eval(str(value))
formatted_json = json.dumps(parsed_json, indent=2)
st.markdown(f"**{selected_metric}:**")
st.code(formatted_json, language='json')
except (ValueError, SyntaxError):
# If all parsing fails, show as raw text
st.markdown(f"**{selected_metric}:** {value}")
except (TypeError, ValueError):
st.markdown(f"**{selected_metric}:** {value}")
else:
st.markdown(f"**{selected_metric}:** N/A")
else:
st.markdown(f"**{selected_metric}:** N/A")
with mcol2:
df2 = eval_tables.get(m2)
if df2 is not None and selected_metric in df2.columns:
if 'gt_sac_id' in df2.columns:
key_val = rec1.get('gt_sac_id', rec1.get('sac_id'))
key_col = 'gt_sac_id'
elif 'gt_title' in df2.columns:
key_val = rec1.get('gt_title', rec1.get('title'))
key_col = 'gt_title'
else:
key_col = key_val = None
if key_col and key_val is not None:
row = df2[df2[key_col] == key_val]
if not row.empty:
value = row[selected_metric].iloc[0]
try:
# Try to parse as JSON first
parsed_json = json.loads(str(value))
formatted_json = json.dumps(parsed_json, indent=2)
st.markdown(f"**{selected_metric}:**")
st.code(formatted_json, language='json')
except json.JSONDecodeError:
try:
# If JSON fails, try to evaluate as Python literal (handles single quotes)
import ast
parsed_json = ast.literal_eval(str(value))
formatted_json = json.dumps(parsed_json, indent=2)
st.markdown(f"**{selected_metric}:**")
st.code(formatted_json, language='json')
except (ValueError, SyntaxError):
# If all parsing fails, show as raw text
st.markdown(f"**{selected_metric}:** {value}")
except (TypeError, ValueError):
st.markdown(f"**{selected_metric}:** {value}")
else:
st.markdown(f"**{selected_metric}:** N/A")
else:
st.markdown(f"**{selected_metric}:** N/A")
with mcol3:
st.markdown("**Ground Truth Metrics**")
df_for_gt = eval_tables.get(m1)
if df_for_gt is None:
df_for_gt = eval_tables.get(m2)
if df_for_gt is not None:
if 'gt_sac_id' in df_for_gt.columns:
key_val = rec1.get('gt_sac_id', rec1.get('sac_id'))
key_col = 'gt_sac_id'
elif 'gt_title' in df_for_gt.columns:
key_val = rec1.get('gt_title', rec1.get('title'))
key_col = 'gt_title'
else:
key_col = key_val = None
if key_col and key_val is not None:
row = df_for_gt[df_for_gt[key_col] == key_val]
if not row.empty:
excluded_gt_fields = {'gt_procedure', 'gt_dspy_uuid', 'gt_dspy_split'}
gt_columns = [col for col in df_for_gt.columns if col.startswith('gt_') and col not in key_columns and col not in excluded_gt_fields]
if gt_columns:
for gt_col in gt_columns:
value = row[gt_col].iloc[0]
try:
# Try to parse as JSON first
parsed_json = json.loads(str(value))
formatted_json = json.dumps(parsed_json, indent=2)
st.markdown(f"**{gt_col}:**")
st.code(formatted_json, language='json')
except json.JSONDecodeError:
try:
# If JSON fails, try to evaluate as Python literal (handles single quotes)
import ast
parsed_json = ast.literal_eval(str(value))
formatted_json = json.dumps(parsed_json, indent=2)
st.markdown(f"**{gt_col}:**")
st.code(formatted_json, language='json')
except (ValueError, SyntaxError):
# If all parsing fails, show as raw text
st.markdown(f"**{gt_col}:** {value}")
except (TypeError, ValueError):
st.markdown(f"**{gt_col}:** {value}")
else:
st.info("No additional ground truth metrics available.")
else:
st.info("No ground truth metrics for this record.")
else:
st.info("No evaluation table available for ground truth metrics.")
st.subheader("Your Preference")
pref = st.radio(
"Which response do you prefer?", options=[m1, m2], key=f"pref_{idx}"
)
st.subheader("Comments (Optional)")
comments = st.text_area(
"Add any comments or notes about your preference:",
height=100,
key=f"comments_{idx}",
placeholder="Optional: Explain your reasoning or add any observations..."
)
if st.session_state.feedback_saved:
st.success("Feedback saved.")
st.session_state.feedback_saved = False
header = [
'timestamp', 'record_id', 'model_1', 'model_2', 'preference',
'input', 'response_1', 'response_2', 'ground_truth', 'comments'
]
row = [
datetime.now().isoformat(), current_id, m1, m2, pref,
rec1.get('input', ''), rec1.get('predicted', ''), rec2.get('predicted', ''),
rec1.get('ground_truth', ''), comments
]
def submit_feedback():
# Get the current text box content at the time of submission
current_comments = st.session_state.get(f"comments_{idx}", "")
# Update the row with the current comments
row[9] = current_comments # comments is at index 9
append_feedback(feedback_path, header, row)
st.session_state.idx += 1
st.session_state.feedback_saved = True
st.button("Submit and Next", on_click=submit_feedback)
if __name__ == '__main__':
main() |