File size: 33,428 Bytes
8daa4df |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 |
#!/usr/bin/env python3
"""
Interactive Benchmark Explorer
A comprehensive web application for exploring OpenThoughts benchmark correlations and model performance
"""
import streamlit as st
import pandas as pd
import numpy as np
import plotly.express as px
import plotly.graph_objects as go
from plotly.subplots import make_subplots
import seaborn as sns
import matplotlib.pyplot as plt
from scipy.stats import pearsonr, spearmanr, kendalltau
from scipy.optimize import minimize
import ast
import io
import base64
from itertools import combinations
import warnings
warnings.filterwarnings('ignore')
# Configure page
st.set_page_config(
page_title="OpenThoughts Evalchemy Benchmark Explorer",
page_icon="π",
layout="wide",
initial_sidebar_state="expanded"
)
# Custom CSS for better styling
st.markdown("""
<style>
.main-header {
font-size: 2.5rem;
font-weight: bold;
color: #1f77b4;
text-align: center;
margin-bottom: 2rem;
}
.metric-card {
background-color: #f8f9fa;
padding: 1rem;
border-radius: 0.5rem;
border-left: 4px solid #1f77b4;
margin: 0.5rem 0;
}
.correlation-high { color: #d73027; font-weight: bold; }
.correlation-medium { color: #fdae61; font-weight: bold; }
.correlation-low { color: #4575b4; font-weight: bold; }
.category-math { color: #d73027; font-weight: bold; }
.category-code { color: #1f78b4; font-weight: bold; }
.category-science { color: #33a02c; font-weight: bold; }
.category-general { color: #ff7f00; font-weight: bold; }
</style>
""", unsafe_allow_html=True)
@st.cache_data
def load_comprehensive_data():
"""Load and clean the comprehensive benchmark data."""
try:
df = pd.read_csv("comprehensive_benchmark_scores.csv", index_col=0)
# Clean the data - handle list-like values stored as strings
for col in df.columns:
def extract_value(x):
if pd.isna(x):
return np.nan
if isinstance(x, str) and x.startswith('['):
try:
return ast.literal_eval(x)[0]
except:
return np.nan
return x
df[col] = df[col].apply(extract_value)
df[col] = pd.to_numeric(df[col], errors='coerce')
# Filter to only models that have data for at least a few benchmarks
min_benchmarks = 3
df = df.dropna(thresh=min_benchmarks, axis=0)
return df
except FileNotFoundError:
st.error("Could not find comprehensive_benchmark_scores.csv. Please ensure the data file exists.")
return pd.DataFrame()
@st.cache_data
def load_stderr_data():
"""Load and clean standard error data."""
try:
stderr_df = pd.read_csv("benchmark_standard_errors.csv", index_col=0)
# Clean the data
for col in stderr_df.columns:
def extract_value(x):
if pd.isna(x):
return np.nan
if isinstance(x, str) and x.startswith('['):
try:
return ast.literal_eval(x)[0]
except:
return np.nan
return x
stderr_df[col] = stderr_df[col].apply(extract_value)
stderr_df[col] = pd.to_numeric(stderr_df[col], errors='coerce')
return stderr_df
except FileNotFoundError:
return None
def clean_benchmark_name(name):
"""Clean benchmark names for consistent display."""
return (name.replace("LiveCodeBench_accuracy_avg", "LiveCodeBenchv2")
.replace('_accuracy_avg', '')
.replace('_accuracy', '')
.replace('LiveCodeBench', 'LCB')
.replace('GPQADiamond', 'GPQAD')
)
def get_focused_benchmark_mapping():
"""Define the target benchmarks and categories."""
target_benchmarks = {
# Math benchmarks
'AIME24': 'AIME24_accuracy_avg',
'AIME25': 'AIME25_accuracy_avg',
'AMC23': 'AMC23_accuracy_avg',
'MATH500': 'MATH500_accuracy',
# Code benchmarks
'CodeElo': 'CodeElo_accuracy_avg',
'CodeForces': 'CodeForces_accuracy_avg',
'LCBv2': 'LiveCodeBench_accuracy_avg',
'LCBv5': 'LiveCodeBenchv5_accuracy_avg',
# Science benchmarks
'GPQADiamond': 'GPQADiamond_accuracy_avg',
'JEEBench': 'JEEBench_accuracy_avg',
# General benchmarks
'MMLUPro': 'MMLUPro_accuracy_avg',
'HLE': 'HLE_accuracy_avg'
}
benchmark_categories = {
'Math': ['AIME24', 'AIME25', 'AMC23', 'MATH500'],
'Code': ['CodeElo', 'CodeForces', 'LCBv2', 'LCBv5'],
'Science': ['GPQADiamond', 'JEEBench'],
'General': ['MMLUPro', 'HLE']
}
colors = {'Math': '#d73027', 'Code': '#1f78b4', 'Science': '#33a02c', 'General': '#ff7f00'}
# Create reverse mapping
col_to_category = {}
for category, bench_list in benchmark_categories.items():
for bench_name in bench_list:
actual_name = target_benchmarks.get(bench_name)
if actual_name:
col_to_category[actual_name] = category
return target_benchmarks, benchmark_categories, colors, col_to_category
def compute_correlations(df, method='pearson'):
"""Compute correlation matrix with the specified method."""
if method == 'pearson':
return df.corr(method='pearson')
elif method == 'spearman':
return df.corr(method='spearman')
elif method == 'kendall':
return df.corr(method='kendall')
def create_interactive_heatmap(corr_matrix, title="Correlation Heatmap"):
"""Create an interactive correlation heatmap using Plotly."""
target_benchmarks, benchmark_categories, colors, col_to_category = get_focused_benchmark_mapping()
# Get clean names for display
clean_names = [clean_benchmark_name(name) for name in corr_matrix.columns]
# Convert to percentages for display
corr_matrix_pct = (corr_matrix * 100).round(1)
# Create hover text
hover_text = []
for i, bench1 in enumerate(corr_matrix.columns):
hover_row = []
for j, bench2 in enumerate(corr_matrix.columns):
if i == j:
hover_row.append(f"{clean_names[i]}<br>Reliability: 100%")
else:
corr_val = corr_matrix_pct.iloc[i, j]
if pd.isna(corr_val):
hover_row.append(f"{clean_names[i]} vs {clean_names[j]}<br>No data")
else:
hover_row.append(f"{clean_names[i]} vs {clean_names[j]}<br>Correlation: {corr_val:.1f}%")
hover_text.append(hover_row)
# Create the heatmap
fig = go.Figure(data=go.Heatmap(
z=corr_matrix.values,
x=clean_names,
y=clean_names,
colorscale='RdBu_r',
zmid=0,
text=corr_matrix_pct.values,
texttemplate="%{text}",
textfont={"size": 10},
hoverinfo='text',
hovertext=hover_text,
colorbar=dict(title="Correlation", tickformat=".2f")
))
# Update layout
fig.update_layout(
title=title,
xaxis_title="",
yaxis_title="",
width=800,
height=800,
font=dict(size=12)
)
# Color the axis labels by category
for i, bench in enumerate(corr_matrix.columns):
category = col_to_category.get(bench, 'Unknown')
color = colors.get(category, 'black')
return fig
def create_scatter_plot(df, x_bench, y_bench, stderr_df=None):
"""Create an interactive scatter plot between two benchmarks."""
if x_bench not in df.columns or y_bench not in df.columns:
return None
# Get common data
common_data = df[[x_bench, y_bench]].dropna()
if len(common_data) < 3:
return None
x_vals = common_data[x_bench]
y_vals = common_data[y_bench]
# Calculate correlation
corr, p_val = pearsonr(x_vals, y_vals)
# Create figure
fig = go.Figure()
# Add scatter points
fig.add_trace(go.Scatter(
x=x_vals,
y=y_vals,
mode='markers',
text=common_data.index,
hovertemplate=(
"<b>%{text}</b><br>" +
f"{clean_benchmark_name(x_bench)}: %{{x:.3f}}<br>" +
f"{clean_benchmark_name(y_bench)}: %{{y:.3f}}<br>" +
"<extra></extra>"
),
marker=dict(size=8, opacity=0.7, color='steelblue')
))
# Add regression line
z = np.polyfit(x_vals, y_vals, 1)
p = np.poly1d(z)
x_line = np.linspace(x_vals.min(), x_vals.max(), 100)
fig.add_trace(go.Scatter(
x=x_line,
y=p(x_line),
mode='lines',
name=f'r = {corr:.3f}, p = {p_val:.3f}',
line=dict(color='red', dash='dash')
))
# Update layout
fig.update_layout(
title=f"{clean_benchmark_name(y_bench)} vs {clean_benchmark_name(x_bench)}",
xaxis_title=clean_benchmark_name(x_bench),
yaxis_title=clean_benchmark_name(y_bench),
showlegend=True,
width=600,
height=500
)
return fig
def filter_target_benchmarks(df):
"""Filter dataframe to only include target benchmarks."""
target_benchmarks, _, _, _ = get_focused_benchmark_mapping()
available_benchmarks = []
for display_name, actual_name in target_benchmarks.items():
if actual_name in df.columns:
available_benchmarks.append(actual_name)
return df[available_benchmarks].copy()
def main():
"""Main application."""
# Header
st.markdown('<div class="main-header">π¬ OpenThoughts Evalchemy Benchmark Explorer</div>', unsafe_allow_html=True)
st.markdown("**Explore correlations and relationships between OpenThoughts model performance across different benchmarks**")
# Load data
with st.spinner("Loading benchmark data..."):
df = load_comprehensive_data()
stderr_df = load_stderr_data()
if df.empty:
st.error("No data available. Please check that the data files exist.")
return
# Filter to target benchmarks
df_filtered = filter_target_benchmarks(df)
target_benchmarks, benchmark_categories, colors, col_to_category = get_focused_benchmark_mapping()
# Sidebar
st.sidebar.header("ποΈ Controls")
# Analysis mode selection
analysis_mode = st.sidebar.selectbox(
"Choose Analysis Mode",
["π Overview Dashboard", "π₯ Interactive Heatmap", "π Scatter Plot Explorer",
"π― Model Performance", "π Statistical Summary", "π¬ Uncertainty Analysis"]
)
# Data filtering options
st.sidebar.subheader("Data Filters")
# Category filter
selected_categories = st.sidebar.multiselect(
"Select Benchmark Categories",
list(benchmark_categories.keys()),
default=list(benchmark_categories.keys())
)
# Filter benchmarks based on selected categories
filtered_benchmarks = []
for category in selected_categories:
for bench_name in benchmark_categories[category]:
actual_name = target_benchmarks.get(bench_name)
if actual_name in df_filtered.columns:
filtered_benchmarks.append(actual_name)
if filtered_benchmarks:
df_display = df_filtered[filtered_benchmarks].copy()
else:
df_display = df_filtered.copy()
# Zero filtering
filter_zeros = st.sidebar.checkbox("Filter out zero/near-zero values", value=False)
if filter_zeros:
for col in df_display.columns:
df_display.loc[(df_display[col] == 0) | (df_display[col] < 0.01), col] = np.nan
# Minimum data points filter
coverage_counts = [df_display[col].notna().sum() for col in df_display.columns]
if coverage_counts:
min_coverage = min(coverage_counts)
max_coverage = max(coverage_counts)
default_min = max(10, min_coverage) # Default to at least 10 or minimum available
min_models = st.sidebar.slider(
"Minimum models per benchmark",
min_value=min_coverage,
max_value=max_coverage,
value=default_min,
help=f"Range: {min_coverage} to {max_coverage} models"
)
else:
min_models = 10
# Apply the minimum models filter
valid_benchmarks = []
for col in df_display.columns:
if df_display[col].notna().sum() >= min_models:
valid_benchmarks.append(col)
df_display = df_display[valid_benchmarks]
# Main content based on analysis mode
if analysis_mode == "π Overview Dashboard":
show_overview_dashboard(df_display, stderr_df)
elif analysis_mode == "π₯ Interactive Heatmap":
show_interactive_heatmap(df_display)
elif analysis_mode == "π Scatter Plot Explorer":
show_scatter_explorer(df_display, stderr_df)
elif analysis_mode == "π― Model Performance":
show_model_performance(df_display)
elif analysis_mode == "π Statistical Summary":
show_statistical_summary(df_display)
elif analysis_mode == "π¬ Uncertainty Analysis":
show_uncertainty_analysis(df_display, stderr_df)
def show_overview_dashboard(df, stderr_df):
"""Show the overview dashboard."""
st.header("π Overview Dashboard")
# Key metrics
col1, col2, col3, col4 = st.columns(4)
with col1:
st.metric("Models", len(df))
with col2:
st.metric("Benchmarks", len(df.columns))
with col3:
total_evals = df.notna().sum().sum()
st.metric("Total Evaluations", f"{total_evals:,}")
with col4:
avg_coverage = (df.notna().sum() / len(df)).mean() * 100
st.metric("Avg Coverage", f"{avg_coverage:.1f}%")
# Benchmark coverage chart
st.subheader("Benchmark Coverage")
coverage_data = []
target_benchmarks, benchmark_categories, colors, col_to_category = get_focused_benchmark_mapping()
for col in df.columns:
coverage = df[col].notna().sum()
category = col_to_category.get(col, 'Unknown')
clean_name = clean_benchmark_name(col)
coverage_data.append({
'Benchmark': clean_name,
'Coverage': coverage,
'Percentage': coverage / len(df) * 100,
'Category': category
})
coverage_df = pd.DataFrame(coverage_data).sort_values('Coverage', ascending=True)
fig = px.bar(coverage_df,
x='Coverage',
y='Benchmark',
color='Category',
color_discrete_map=colors,
title="Model Coverage by Benchmark",
labels={'Coverage': 'Number of Models'},
orientation='h')
fig.update_layout(height=400)
st.plotly_chart(fig, use_container_width=True)
# Quick correlation insights
st.subheader("Quick Correlation Insights")
corr_matrix = compute_correlations(df, 'pearson')
# Get top correlations
pairs = []
for i, bench1 in enumerate(corr_matrix.columns):
for j, bench2 in enumerate(corr_matrix.columns[i+1:], i+1):
if not pd.isna(corr_matrix.iloc[i, j]):
cat1 = col_to_category.get(bench1, 'Unknown')
cat2 = col_to_category.get(bench2, 'Unknown')
pairs.append((bench1, bench2, corr_matrix.iloc[i, j], cat1, cat2))
pairs.sort(key=lambda x: abs(x[2]), reverse=True)
col1, col2 = st.columns(2)
with col1:
st.markdown("**π₯ Top 5 Highest Correlations**")
for i, (bench1, bench2, corr, cat1, cat2) in enumerate(pairs[:5]):
same_cat = "β
" if cat1 == cat2 else "π"
st.write(f"{i+1}. {clean_benchmark_name(bench1)} β {clean_benchmark_name(bench2)}")
st.write(f" r = {corr:.3f} {same_cat}")
with col2:
st.markdown("**π Category Analysis**")
within_cat = [p[2] for p in pairs if p[3] == p[4]]
across_cat = [p[2] for p in pairs if p[3] != p[4]]
if within_cat:
st.write(f"Within-category avg: {np.mean(within_cat):.3f}")
if across_cat:
st.write(f"Across-category avg: {np.mean(across_cat):.3f}")
st.write(f"Total pairs analyzed: {len(pairs)}")
def show_interactive_heatmap(df):
"""Show the interactive heatmap."""
st.header("π₯ Interactive Correlation Heatmap")
# Correlation method selection
col1, col2 = st.columns([3, 1])
with col2:
corr_method = st.selectbox(
"Correlation Method",
["pearson", "spearman", "kendall"]
)
# Compute correlation matrix
corr_matrix = compute_correlations(df, corr_method)
# Create and display heatmap
fig = create_interactive_heatmap(corr_matrix, f"{corr_method.capitalize()} Correlation Matrix")
st.plotly_chart(fig, use_container_width=True)
# Correlation statistics
st.subheader("Correlation Statistics")
# Get all off-diagonal correlations
mask = np.triu(np.ones_like(corr_matrix, dtype=bool), k=1)
corr_values = corr_matrix.where(mask).stack().dropna()
col1, col2, col3, col4 = st.columns(4)
with col1:
st.metric("Mean Correlation", f"{corr_values.mean():.3f}")
with col2:
st.metric("Median Correlation", f"{corr_values.median():.3f}")
with col3:
st.metric("Max Correlation", f"{corr_values.max():.3f}")
with col4:
st.metric("Min Correlation", f"{corr_values.min():.3f}")
# Distribution of correlations
st.subheader("Correlation Distribution")
fig = px.histogram(corr_values,
nbins=20,
title="Distribution of Pairwise Correlations",
labels={'value': 'Correlation Coefficient', 'count': 'Frequency'})
st.plotly_chart(fig, use_container_width=True)
def show_scatter_explorer(df, stderr_df):
"""Show the scatter plot explorer."""
st.header("π Scatter Plot Explorer")
# Benchmark selection
col1, col2 = st.columns(2)
with col1:
x_benchmark = st.selectbox(
"X-axis Benchmark",
df.columns,
format_func=clean_benchmark_name
)
with col2:
y_benchmark = st.selectbox(
"Y-axis Benchmark",
df.columns,
index=1 if len(df.columns) > 1 else 0,
format_func=clean_benchmark_name
)
if x_benchmark and y_benchmark and x_benchmark != y_benchmark:
# Create scatter plot
fig = create_scatter_plot(df, x_benchmark, y_benchmark, stderr_df)
if fig:
st.plotly_chart(fig, use_container_width=True)
# Additional statistics
common_data = df[[x_benchmark, y_benchmark]].dropna()
if len(common_data) >= 3:
col1, col2, col3 = st.columns(3)
# Correlation metrics
pearson_r, pearson_p = pearsonr(common_data[x_benchmark], common_data[y_benchmark])
spearman_r, spearman_p = spearmanr(common_data[x_benchmark], common_data[y_benchmark])
kendall_r, kendall_p = kendalltau(common_data[x_benchmark], common_data[y_benchmark])
with col1:
st.metric("Pearson r", f"{pearson_r:.3f}")
st.caption(f"p = {pearson_p:.3f}")
with col2:
st.metric("Spearman Ο", f"{spearman_r:.3f}")
st.caption(f"p = {spearman_p:.3f}")
with col3:
st.metric("Kendall Ο", f"{kendall_r:.3f}")
st.caption(f"p = {kendall_p:.3f}")
# Show data table
st.subheader("Data Points")
display_data = common_data.copy()
display_data.columns = [clean_benchmark_name(col) for col in display_data.columns]
st.dataframe(display_data, use_container_width=True)
else:
st.warning("Insufficient data for the selected benchmark pair.")
else:
st.info("Please select two different benchmarks to compare.")
def show_model_performance(df):
"""Show model performance analysis."""
st.header("π― Model Performance Analysis")
# Model search
search_term = st.text_input("π Search for models", placeholder="Enter model name or part of name")
if search_term:
matching_models = df.index[df.index.str.contains(search_term, case=False, na=False)]
if len(matching_models) > 0:
df_display = df.loc[matching_models]
else:
st.warning(f"No models found matching '{search_term}'")
df_display = df
else:
df_display = df
# Performance ranking
st.subheader("Model Rankings")
# Calculate average performance (excluding NaN)
model_avg_scores = df_display.mean(axis=1, skipna=True).sort_values(ascending=False)
# Top performers
col1, col2 = st.columns(2)
with col1:
st.markdown("**π Top 10 Models (by average score)**")
for i, (model, score) in enumerate(model_avg_scores.head(10).items()):
st.write(f"{i+1}. {model.split('/')[-1]}: {score:.3f}")
with col2:
st.markdown("**π Performance Distribution**")
fig = px.histogram(model_avg_scores,
nbins=20,
title="Distribution of Average Model Scores")
st.plotly_chart(fig, use_container_width=True)
# Model comparison
st.subheader("Model Comparison")
selected_models = st.multiselect(
"Select models to compare",
df_display.index.tolist(),
default=model_avg_scores.head(3).index.tolist()
)
if selected_models:
comparison_data = df_display.loc[selected_models].T
comparison_data.index = [clean_benchmark_name(idx) for idx in comparison_data.index]
# Radar chart
if len(selected_models) <= 5: # Only for manageable number of models
fig = go.Figure()
for model in selected_models:
model_data = df_display.loc[model].dropna()
benchmarks = [clean_benchmark_name(b) for b in model_data.index]
values = model_data.values.tolist()
# Close the radar chart
values += values[:1]
benchmarks += benchmarks[:1]
fig.add_trace(go.Scatterpolar(
r=values,
theta=benchmarks,
fill='toself',
name=model.split('/')[-1]
))
fig.update_layout(
polar=dict(
radialaxis=dict(
visible=True,
range=[0, 1]
)),
showlegend=True,
title="Model Performance Radar Chart"
)
st.plotly_chart(fig, use_container_width=True)
# Detailed comparison table
st.subheader("Detailed Comparison")
st.dataframe(comparison_data, use_container_width=True)
def show_statistical_summary(df):
"""Show statistical summary."""
st.header("π Statistical Summary")
# Overall statistics
st.subheader("Dataset Statistics")
col1, col2 = st.columns(2)
with col1:
st.markdown("**Data Coverage**")
total_possible = len(df) * len(df.columns)
total_actual = df.notna().sum().sum()
coverage_pct = (total_actual / total_possible) * 100
st.write(f"Total possible evaluations: {total_possible:,}")
st.write(f"Actual evaluations: {total_actual:,}")
st.write(f"Overall coverage: {coverage_pct:.1f}%")
with col2:
st.markdown("**Score Statistics**")
all_scores = df.values.flatten()
all_scores = all_scores[~pd.isna(all_scores)]
st.write(f"Mean score: {np.mean(all_scores):.3f}")
st.write(f"Median score: {np.median(all_scores):.3f}")
st.write(f"Std deviation: {np.std(all_scores):.3f}")
# Benchmark-wise statistics
st.subheader("Benchmark Statistics")
benchmark_stats = []
target_benchmarks, benchmark_categories, colors, col_to_category = get_focused_benchmark_mapping()
for col in df.columns:
scores = df[col].dropna()
if len(scores) > 0:
benchmark_stats.append({
'Benchmark': clean_benchmark_name(col),
'Category': col_to_category.get(col, 'Unknown'),
'Count': len(scores),
'Mean': scores.mean(),
'Median': scores.median(),
'Std': scores.std(),
'Min': scores.min(),
'Max': scores.max(),
'Range': scores.max() - scores.min()
})
stats_df = pd.DataFrame(benchmark_stats)
st.dataframe(stats_df, use_container_width=True)
# Correlation summary
st.subheader("Correlation Analysis Summary")
for method in ['pearson', 'spearman', 'kendall']:
corr_matrix = compute_correlations(df, method)
# Get all off-diagonal correlations
mask = np.triu(np.ones_like(corr_matrix, dtype=bool), k=1)
corr_values = corr_matrix.where(mask).stack().dropna()
st.write(f"**{method.capitalize()} Correlations:**")
col1, col2, col3, col4 = st.columns(4)
with col1:
st.metric("Mean", f"{corr_values.mean():.3f}")
with col2:
st.metric("Median", f"{corr_values.median():.3f}")
with col3:
st.metric("Max", f"{corr_values.max():.3f}")
with col4:
st.metric("Min", f"{corr_values.min():.3f}")
def show_uncertainty_analysis(df, stderr_df):
"""Show uncertainty analysis if standard error data is available."""
st.header("π¬ Uncertainty Analysis")
if stderr_df is None:
st.warning("Standard error data not available. This analysis requires benchmark_standard_errors.csv")
return
st.info("This section analyzes measurement uncertainty and reliability of benchmark evaluations.")
# Match benchmarks with standard errors
matched_benchmarks = []
for score_col in df.columns:
# Try to find matching stderr column
potential_stderr_cols = [
f"{score_col}_std_err",
f"{score_col.replace('_accuracy', '_accuracy_std_err')}",
f"{score_col.replace('_accuracy_avg', '_accuracy_std_err')}"
]
for stderr_col in potential_stderr_cols:
if stderr_col in stderr_df.columns:
matched_benchmarks.append((score_col, stderr_col))
break
if not matched_benchmarks:
st.warning("No matching standard error data found for the selected benchmarks.")
return
st.success(f"Found standard error data for {len(matched_benchmarks)} benchmarks.")
# Measurement precision analysis
st.subheader("Measurement Precision")
precision_data = []
for score_col, stderr_col in matched_benchmarks:
scores = df[score_col].dropna()
stderrs = stderr_df[stderr_col].dropna()
if len(stderrs) > 0:
mean_stderr = stderrs.mean()
median_stderr = stderrs.median()
# Signal-to-noise ratio
if len(scores) > 0:
signal_std = scores.std()
snr = signal_std / mean_stderr if mean_stderr > 0 else float('inf')
else:
snr = 0
precision_data.append({
'Benchmark': clean_benchmark_name(score_col),
'Mean StdErr': mean_stderr,
'Median StdErr': median_stderr,
'Signal/Noise': snr,
'N Models': len(stderrs)
})
if precision_data:
precision_df = pd.DataFrame(precision_data)
st.dataframe(precision_df, use_container_width=True)
# Visualization
fig = px.scatter(precision_df,
x='Mean StdErr',
y='Signal/Noise',
size='N Models',
hover_name='Benchmark',
title="Measurement Precision: Signal-to-Noise vs Standard Error",
labels={'Signal/Noise': 'Signal-to-Noise Ratio'})
st.plotly_chart(fig, use_container_width=True)
# Uncertainty-aware scatter plot
st.subheader("Uncertainty-Aware Scatter Plot")
# Let user select benchmarks with stderr data
available_benchmarks = [score_col for score_col, _ in matched_benchmarks]
col1, col2 = st.columns(2)
with col1:
x_bench = st.selectbox(
"X-axis Benchmark (with uncertainty)",
available_benchmarks,
format_func=clean_benchmark_name
)
with col2:
y_bench = st.selectbox(
"Y-axis Benchmark (with uncertainty)",
available_benchmarks,
index=1 if len(available_benchmarks) > 1 else 0,
format_func=clean_benchmark_name
)
if x_bench and y_bench and x_bench != y_bench:
# Find corresponding stderr columns
x_stderr_col = None
y_stderr_col = None
for score_col, stderr_col in matched_benchmarks:
if score_col == x_bench:
x_stderr_col = stderr_col
if score_col == y_bench:
y_stderr_col = stderr_col
if x_stderr_col and y_stderr_col:
# Get data
x_scores = df[x_bench]
y_scores = df[y_bench]
x_err = stderr_df[x_stderr_col]
y_err = stderr_df[y_stderr_col]
# Find common valid data
valid_mask = ~(x_scores.isna() | y_scores.isna() | x_err.isna() | y_err.isna())
if valid_mask.sum() >= 3:
x_clean = x_scores[valid_mask]
y_clean = y_scores[valid_mask]
x_err_clean = x_err[valid_mask]
y_err_clean = y_err[valid_mask]
# Create uncertainty scatter plot
fig = go.Figure()
# Add error bars
fig.add_trace(go.Scatter(
x=x_clean,
y=y_clean,
error_x=dict(
type='data',
array=1.96 * x_err_clean, # 95% CI
visible=True
),
error_y=dict(
type='data',
array=1.96 * y_err_clean, # 95% CI
visible=True
),
mode='markers',
text=x_clean.index,
hovertemplate=(
"<b>%{text}</b><br>" +
f"{clean_benchmark_name(x_bench)}: %{{x:.3f}} Β± %{{error_x:.3f}}<br>" +
f"{clean_benchmark_name(y_bench)}: %{{y:.3f}} Β± %{{error_y:.3f}}<br>" +
"<extra></extra>"
),
marker=dict(size=8, opacity=0.7),
name='Models'
))
# Add regression line
corr, p_val = pearsonr(x_clean, y_clean)
z = np.polyfit(x_clean, y_clean, 1)
p = np.poly1d(z)
x_line = np.linspace(x_clean.min(), x_clean.max(), 100)
fig.add_trace(go.Scatter(
x=x_line,
y=p(x_line),
mode='lines',
name=f'r = {corr:.3f}, p = {p_val:.3f}',
line=dict(color='red', dash='dash')
))
fig.update_layout(
title=f"Uncertainty-Aware Correlation: {clean_benchmark_name(y_bench)} vs {clean_benchmark_name(x_bench)}",
xaxis_title=f"{clean_benchmark_name(x_bench)} (Β±95% CI)",
yaxis_title=f"{clean_benchmark_name(y_bench)} (Β±95% CI)",
showlegend=True
)
st.plotly_chart(fig, use_container_width=True)
st.info(f"Showing {len(x_clean)} models with both score and uncertainty data. Error bars represent 95% confidence intervals.")
else:
st.warning("Insufficient data with uncertainty estimates for the selected benchmark pair.")
if __name__ == "__main__":
main() |