Spaces:
Sleeping
Sleeping
File size: 18,137 Bytes
587f33e | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 | # Copyright 2026 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import streamlit as st
import json
import base64
from io import BytesIO
from PIL import Image
import os
import sys
import asyncio
import importlib
import re
# Ensure local imports work
sys.path.append(os.getcwd())
st.set_page_config(layout="wide", page_title="PaperVizAgent Referenced Eval Visualizer", page_icon="π")
def detect_task_type(data):
"""Detect whether data is for diagram or plot task."""
if not data:
return "diagram"
sample = data[0]
# Check for plot-specific fields
if "content" in sample and isinstance(sample.get("content"), dict):
return "plot"
# Check for diagram-specific fields
# Now directly accessible from top level
return "diagram"
# Default to diagram
return "diagram"
@st.cache_data
def load_data(path):
"""Read JSONL or JSON data."""
data = []
if not os.path.exists(path):
return []
# Detect file format by extension
file_ext = os.path.splitext(path)[1].lower()
try:
with open(path, "r", encoding="utf-8") as f:
if file_ext == ".json":
# Load as a single JSON array
try:
data = json.load(f)
if not isinstance(data, list):
st.error("JSON file must contain an array at the top level")
return []
except json.JSONDecodeError as e:
st.error(f"Invalid JSON format: {e}")
return []
else:
# Load as JSONL (line-by-line)
for line in f:
line = line.strip()
if not line:
continue
try:
data.append(json.loads(line))
except json.JSONDecodeError:
continue
except Exception as e:
st.error(f"Error reading file: {e}")
return []
return data
def calculate_stats(data, dimensions):
"""Calculate win rates for each dimension."""
outcomes = ["Model", "Human", "Both are good", "Both are bad", "Tie", "Error", "Unknown"]
stats = {dim: {out: 0 for out in outcomes} for dim in dimensions}
for item in data:
for dim in dimensions:
outcome = item.get(f"{dim.lower()}_outcome", "Unknown")
if outcome in outcomes:
stats[dim][outcome] += 1
else:
stats[dim]["Unknown"] += 1
return stats
def base64_to_image(b64_str):
if not b64_str:
return None
try:
if "," in b64_str:
b64_str = b64_str.split(",")[1]
image_data = base64.b64decode(b64_str)
return Image.open(BytesIO(image_data))
except Exception:
return None
def load_local_image(path):
if path and os.path.exists(path):
return Image.open(path)
return None
def display_outcome(outcome):
if outcome == "Model":
return f":blue[**{outcome}**]"
if outcome == "Human":
return f":green[**{outcome}**]"
if outcome == "Both are good":
return f":orange[**{outcome}**]"
if outcome == "Both are bad":
return f":red[**{outcome}**]"
if outcome == "Tie":
return f":violet[**{outcome}**]"
return f":gray[**{outcome}**]"
def format_reasoning(text):
"""Format reasoning string for better readability in Streamlit."""
if not text:
return ""
# Common headers used in prompts
headers = [
"Faithfulness of Human", "Faithfulness of Model",
"Conciseness of Human", "Conciseness of Model",
"Readability of Human", "Readability of Model",
"Aesthetics of Human", "Aesthetics of Model",
"Overall Quality of Human", "Overall Quality of Model",
"Conclusion"
]
formatted_text = text
# Ensure headers at the start or after a space/punctuation are bolded and preceded by newlines
for header in headers:
# Match header followed by colon, case-insensitive
pattern = re.compile(rf"({re.escape(header)}):", re.IGNORECASE)
# Use \n\n to ensure a clear paragraph break
formatted_text = pattern.sub(r"\n\n**\1**:", formatted_text)
# Clean up: remove semicolon if it's right before our new bolded section
formatted_text = re.sub(r";\s*\n\n", r"\n\n", formatted_text)
# Final trim
return formatted_text.strip()
async def run_eval_on_sample(sample, task_name="diagram"):
"""Hot-reload prompts and run eval."""
import prompts.diagram_eval_prompts
import prompts.plots_eval_prompts
import utils.eval_toolkits
importlib.reload(prompts.diagram_eval_prompts)
importlib.reload(prompts.plots_eval_prompts)
importlib.reload(utils.eval_toolkits)
from utils.eval_toolkits import get_score_for_image_referenced
# Ensure eval_image_field is set
if "eval_image_field" not in sample:
# Try to infer from available fields or use default
if task_name == "plot":
if "target_plot_desc0_base64_jpg" in sample:
sample["eval_image_field"] = "target_plot_desc0_base64_jpg"
elif "target_plot_stylist_desc0_base64_jpg" in sample:
sample["eval_image_field"] = "target_plot_stylist_desc0_base64_jpg"
else:
if "target_diagram_critic_desc0_base64_jpg" in sample:
sample["eval_image_field"] = "target_diagram_critic_desc0_base64_jpg"
elif "target_diagram_stylist_desc0_base64_jpg" in sample:
sample["eval_image_field"] = "target_diagram_stylist_desc0_base64_jpg"
elif "target_diagram_desc0_base64_jpg" in sample:
sample["eval_image_field"] = "target_diagram_desc0_base64_jpg"
else:
sample["eval_image_field"] = "vanilla_image_base64" # fallback
return await get_score_for_image_referenced(sample, task_name=task_name)
def main():
st.sidebar.title("π PaperVizAgent Referenced Eval")
file_path = st.sidebar.text_input("Results JSONL Path", placeholder="Enter path to results file...")
# --- Debug Tool Section in Sidebar ---
if "debug_sample" in st.session_state:
st.sidebar.divider()
st.sidebar.subheader("π οΈ Debug Target")
debug_sample = st.session_state.debug_sample
identifier = debug_sample.get('id')
st.sidebar.info(f"Active: {identifier}\nIndex: {st.session_state.debug_idx}")
if st.sidebar.button("π Re-run Eval (Hot-Reload Prompts)", type="primary"):
with st.spinner("Running live evaluation..."):
try:
# Pass task_name if available
task_name = st.session_state.get("task_type", "diagram")
new_result = asyncio.run(run_eval_on_sample(debug_sample.copy(), task_name))
st.session_state.debug_result = new_result
st.sidebar.success("Evaluation Complete!")
except Exception as e:
st.sidebar.error(f"Eval Failed: {e}")
if st.sidebar.button("π§Ή Clear Debug State"):
if "debug_sample" in st.session_state: del st.session_state.debug_sample
if "debug_idx" in st.session_state: del st.session_state.debug_idx
if "debug_result" in st.session_state: del st.session_state.debug_result
st.rerun()
if st.sidebar.button("π Refresh Data"):
load_data.clear()
st.rerun()
if not file_path:
st.info("π Please enter a file path to begin")
st.stop()
if not os.path.exists(file_path):
st.error(f"File not found: {file_path}")
st.stop()
data = load_data(file_path)
# Detect task type
task_type = detect_task_type(data)
st.session_state["task_type"] = task_type
# --- Display Mode Selection ---
if task_type == "plot":
display_mode = st.sidebar.selectbox(
"Model Display Mode",
["Auto", "Vanilla", "Stylist"],
help="Select which stage of the model output to display."
)
mode_to_keys = {
"Vanilla": ("target_plot_desc0_base64_jpg", "target_plot_desc0"),
"Stylist": ("target_plot_stylist_desc0_base64_jpg", "target_plot_stylist_desc0"),
}
else: # diagram
display_mode = st.sidebar.selectbox(
"Model Display Mode",
["Auto", "Vanilla", "Stylist", "Critic"],
help="Select which stage of the model output to display."
)
mode_to_keys = {
"Vanilla": ("target_diagram_desc0_base64_jpg", "target_diagram_desc0"),
"Stylist": ("target_diagram_stylist_desc0_base64_jpg", "target_diagram_stylist_desc0"),
"Critic": ("target_diagram_critic_desc0_base64_jpg", "target_diagram_critic_desc0"),
}
# --- Search Functionality ---
search_field = "id"
search_query = st.sidebar.text_input(f"π Search {search_field.title()}", value="", help=f"Filter by {search_field} (case-insensitive)")
if search_query:
data = [item for item in data if search_query.lower() in str(item.get(search_field, "")).lower()]
st.sidebar.caption(f"Found {len(data)} matching cases")
total_items = len(data)
if total_items == 0:
if search_query:
st.warning(f"No samples found matching '{search_query}'.")
else:
st.warning("Data is empty or format is incorrect.")
return
st.title(f"Referenced Evaluation Visualizer")
# --- Global Stats ---
dimensions = ["Faithfulness", "Conciseness", "Readability", "Aesthetics", "Overall"]
stats = calculate_stats(data, dimensions)
with st.expander("π Global Statistics", expanded=False):
cols = st.columns(len(dimensions))
for i, dim in enumerate(dimensions):
with cols[i]:
st.info(f"**{dim}**")
s = stats[dim]
total = sum(s.values()) or 1
st.metric("Model Win", f"{(s['Model'])/total:.1%}")
st.metric("Human Win", f"{(s['Human'])/total:.1%}")
st.metric("Both Good", f"{(s['Both are good'])/total:.1%}")
st.metric("Both Bad", f"{(s['Both are bad'])/total:.1%}")
# Add Tie metric for Overall dimension
if dim == "Overall":
tie_count = s.get("Tie", 0)
st.metric("Tie", f"{tie_count/total:.1%}")
st.divider()
# --- Pagination ---
PAGE_SIZE = 10
if "page" not in st.session_state:
st.session_state.page = 0
total_pages = max((total_items + PAGE_SIZE - 1) // PAGE_SIZE, 1)
def on_page_change():
st.session_state.page = st.session_state.page_input - 1
st.sidebar.number_input(
"Page",
min_value=1,
max_value=total_pages,
value=st.session_state.page + 1,
key="page_input",
on_change=on_page_change
)
# Ensure page is within valid range (e.g. if search reduced results)
if st.session_state.page >= total_pages:
st.session_state.page = total_pages - 1
start_idx = st.session_state.page * PAGE_SIZE
end_idx = min(start_idx + PAGE_SIZE, total_items)
batch = data[start_idx:end_idx]
st.sidebar.markdown(f"Displaying {start_idx + 1} - {end_idx} of {total_items}")
for i, item in enumerate(batch):
idx = start_idx + i
# Extract metadata based on task type
identifier = item.get("id", "Unknown")
caption_or_desc = item.get("visual_intent") or item.get("brief_desc", "N/A")
if task_type == "plot":
raw_content_label = "Raw Data"
raw_content = json.dumps(item.get("content", {}), indent=2)
else: # diagram
raw_content_label = "Method Section"
raw_content = item.get("content", "N/A")
is_debugging = "debug_sample" in st.session_state and st.session_state.debug_idx == idx
with st.container(border=is_debugging):
col_title, col_btn = st.columns([0.8, 0.2])
with col_title:
st.subheader(f"#{idx + 1}: {caption_or_desc}")
with col_btn:
if st.button("π οΈ Debug", key=f"btn_debug_{idx}"):
st.session_state.debug_sample = item
st.session_state.debug_idx = idx
st.rerun()
st.caption(f"{search_field.title()}: `{identifier}`")
# --- Determine Image and Text for Model ---
if display_mode == "Auto":
eval_field = item.get("eval_image_field")
if eval_field:
model_b64_key = eval_field
model_text_key = eval_field.replace("_base64_jpg", "")
else:
# Fallback
if task_type == "plot":
model_b64_key = "target_plot_desc0_base64_jpg"
model_text_key = "target_plot_desc0"
else:
model_b64_key = "target_diagram_critic_desc0_base64_jpg"
model_text_key = "target_diagram_critic_desc0"
else:
model_b64_key, model_text_key = mode_to_keys[display_mode]
model_b64 = item.get(model_b64_key)
model_description = item.get(model_text_key, "N/A")
# Outcome Summary
outcome_cols = st.columns(len(dimensions))
for j, dim in enumerate(dimensions):
outcome_cols[j].markdown(f"**{dim}**\n{display_outcome(item.get(f'{dim.lower()}_outcome'))}")
# Debug Results Overlay
if is_debugging and "debug_result" in st.session_state:
st.markdown("---")
st.markdown("### π§ͺ **LIVE DEBUG RESULTS** (from current prompt)")
new_res = st.session_state.debug_result
new_cols = st.columns(len(dimensions))
for j, dim in enumerate(dimensions):
new_cols[j].markdown(f"**{dim}**\n{display_outcome(new_res.get(f'{dim.lower()}_outcome'))}")
st.markdown("---")
# Images
img_col1, img_col2 = st.columns(2)
with img_col1:
model_label = "Model Plot" if task_type == "plot" else "Model Diagram"
st.markdown(f"**{model_label}** ({display_mode})")
if model_b64:
st.image(base64_to_image(model_b64), use_container_width=True)
else:
st.error(f"Missing key: `{model_b64_key}`")
with st.expander("π Model Description", expanded=False):
st.write(model_description)
with img_col2:
human_label = "Human Plot" if task_type == "plot" else "Human Diagram"
st.markdown(f"**{human_label}** (Reference)")
# Get GT image path based on task type
if task_type == "plot":
gt_path = item.get("path_to_gt_image")
else:
gt_path = item.get("path_to_gt_image")
gt_img = load_local_image(gt_path)
if gt_img:
st.image(gt_img, use_container_width=True)
else:
st.error(f"Human image not found at: {gt_path}")
with st.expander("π Human/Caption Info", expanded=False):
st.markdown(f"**Caption/Description:** {caption_or_desc}")
if task_type == "diagram":
st.markdown(f"**Human Analysis:** {item.get('gt_diagram_desc0', 'N/A')}")
# Suggestions (if any)
suggestions = item.get("suggestions_diagram") or item.get("suggestions_plot")
if suggestions:
with st.expander("π‘ Polish Suggestions", expanded=False):
st.markdown(suggestions)
# Raw Content Section - spans full width
with st.expander(f"π {raw_content_label}", expanded=False):
if task_type == "plot":
st.code(raw_content, language="json")
else:
st.markdown(raw_content)
# Reasoning
with st.expander("π Original Reasoning", expanded=False):
tabs = st.tabs(dimensions)
for tab, dim in zip(tabs, dimensions):
with tab:
reasoning = item.get(f"{dim.lower()}_reasoning", "No reasoning provided.")
st.markdown(format_reasoning(reasoning))
if is_debugging and "debug_result" in st.session_state:
with st.expander("π§ͺ Debug Reasoning (Current Prompt)", expanded=True):
new_res = st.session_state.debug_result
tabs = st.tabs(dimensions)
for tab, dim in zip(tabs, dimensions):
with tab:
reasoning = new_res.get(f"{dim.lower()}_reasoning", "N/A")
st.markdown(format_reasoning(reasoning))
st.divider()
if __name__ == "__main__":
main()
|