thinkwee
commited on
Commit
·
7247a32
1
Parent(s):
845705b
update animation
Browse files- benchmarking.js +23 -1
- charts.js +77 -20
- index.html +38 -7
- styles.css +51 -0
benchmarking.js
CHANGED
|
@@ -5,9 +5,31 @@ document.addEventListener('DOMContentLoaded', function () {
|
|
| 5 |
return;
|
| 6 |
}
|
| 7 |
|
| 8 |
-
|
| 9 |
});
|
| 10 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
function renderBenchmarkingChart() {
|
| 12 |
const container = document.getElementById('benchmarking-chart');
|
| 13 |
if (!container) return;
|
|
|
|
| 5 |
return;
|
| 6 |
}
|
| 7 |
|
| 8 |
+
initBenchmarkingChart();
|
| 9 |
});
|
| 10 |
|
| 11 |
+
function initBenchmarkingChart() {
|
| 12 |
+
const container = document.getElementById('benchmarking-chart');
|
| 13 |
+
if (!container) return;
|
| 14 |
+
|
| 15 |
+
// Use Intersection Observer to trigger animation when chart comes into view
|
| 16 |
+
let hasAnimated = false;
|
| 17 |
+
|
| 18 |
+
const observer = new IntersectionObserver((entries) => {
|
| 19 |
+
entries.forEach(entry => {
|
| 20 |
+
if (entry.isIntersecting && !hasAnimated) {
|
| 21 |
+
hasAnimated = true;
|
| 22 |
+
renderBenchmarkingChart();
|
| 23 |
+
observer.disconnect(); // Stop observing after animation starts
|
| 24 |
+
}
|
| 25 |
+
});
|
| 26 |
+
}, {
|
| 27 |
+
threshold: 0.2 // Trigger when 20% of the chart is visible
|
| 28 |
+
});
|
| 29 |
+
|
| 30 |
+
observer.observe(container);
|
| 31 |
+
}
|
| 32 |
+
|
| 33 |
function renderBenchmarkingChart() {
|
| 34 |
const container = document.getElementById('benchmarking-chart');
|
| 35 |
if (!container) return;
|
charts.js
CHANGED
|
@@ -106,8 +106,8 @@ const lazyLoadObserver = new IntersectionObserver((entries) => {
|
|
| 106 |
}
|
| 107 |
});
|
| 108 |
}, {
|
| 109 |
-
rootMargin: '
|
| 110 |
-
threshold: 0.
|
| 111 |
});
|
| 112 |
|
| 113 |
// Debounce utility for hover effects
|
|
@@ -252,15 +252,16 @@ function initScalingCharts() {
|
|
| 252 |
const modelNormX = normTurns.slice(offset, offset + len);
|
| 253 |
offset += len;
|
| 254 |
|
|
|
|
| 255 |
traces.push({
|
| 256 |
x: modelNormX,
|
| 257 |
y: data[model].accuracy,
|
| 258 |
-
mode: '
|
| 259 |
-
name: model,
|
| 260 |
line: { color: DDR_DATA.modelColors[model] || '#888', width: 2 },
|
| 261 |
marker: { size: 6, color: DDR_DATA.modelColors[model] || '#888' },
|
| 262 |
hovertemplate: `<b>${model}</b><br>Turn: %{customdata}<br>Accuracy: %{y:.2f}%<extra></extra>`,
|
| 263 |
-
customdata: data[model].turns
|
| 264 |
});
|
| 265 |
});
|
| 266 |
|
|
@@ -273,9 +274,9 @@ function initScalingCharts() {
|
|
| 273 |
...darkLayout,
|
| 274 |
xaxis: {
|
| 275 |
...darkLayout.xaxis,
|
| 276 |
-
title: { text: 'Number of Interaction Turns', font: { size: 15, color: '#1d1d1f' } },
|
| 277 |
-
type: 'linear',
|
| 278 |
-
range: [-0.05, 1.05],
|
| 279 |
tickmode: 'array',
|
| 280 |
tickvals: tickVals,
|
| 281 |
ticktext: tickText,
|
|
@@ -283,14 +284,20 @@ function initScalingCharts() {
|
|
| 283 |
},
|
| 284 |
yaxis: {
|
| 285 |
...darkLayout.yaxis,
|
| 286 |
-
title: { text: 'Accuracy (%)', font: { size: 15, color: '#1d1d1f' } },
|
| 287 |
dtick: dtickVal,
|
| 288 |
range: yRange
|
| 289 |
},
|
| 290 |
-
showlegend: false
|
| 291 |
};
|
| 292 |
|
| 293 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 294 |
});
|
| 295 |
|
| 296 |
// Populate shared legend with models from first scenario
|
|
@@ -301,7 +308,64 @@ function initScalingCharts() {
|
|
| 301 |
}
|
| 302 |
|
| 303 |
// Apply hover effects after charts are rendered
|
| 304 |
-
setTimeout(() => applyHoverEffectsForSection('scaling'),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 305 |
}
|
| 306 |
|
| 307 |
function updateScalingCharts(dimension) {
|
|
@@ -1228,18 +1292,11 @@ function renderEntropyCharts(scenario) {
|
|
| 1228 |
// ============================================================================
|
| 1229 |
document.addEventListener('DOMContentLoaded', () => {
|
| 1230 |
// Register all sections for lazy loading
|
|
|
|
| 1231 |
const sections = document.querySelectorAll('section.section');
|
| 1232 |
sections.forEach(section => {
|
| 1233 |
lazyLoadObserver.observe(section);
|
| 1234 |
});
|
| 1235 |
-
|
| 1236 |
-
// Immediately initialize the first visible section (scaling) for instant feedback
|
| 1237 |
-
// Other sections will be lazy-loaded as user scrolls
|
| 1238 |
-
if (document.getElementById('scaling')) {
|
| 1239 |
-
initializedCharts.add('scaling');
|
| 1240 |
-
// Use setTimeout to not block the main thread
|
| 1241 |
-
setTimeout(() => initScalingCharts(), 0);
|
| 1242 |
-
}
|
| 1243 |
});
|
| 1244 |
|
| 1245 |
// Handle window resize with longer debounce for better performance
|
|
|
|
| 106 |
}
|
| 107 |
});
|
| 108 |
}, {
|
| 109 |
+
rootMargin: '0px 0px', // Start exactly when entering viewport
|
| 110 |
+
threshold: 0.15 // Trigger when 15% visible
|
| 111 |
});
|
| 112 |
|
| 113 |
// Debounce utility for hover effects
|
|
|
|
| 252 |
const modelNormX = normTurns.slice(offset, offset + len);
|
| 253 |
offset += len;
|
| 254 |
|
| 255 |
+
// Start with markers only (lines will be animated in)
|
| 256 |
traces.push({
|
| 257 |
x: modelNormX,
|
| 258 |
y: data[model].accuracy,
|
| 259 |
+
mode: 'markers', // Start with markers only
|
| 260 |
+
name: model,
|
| 261 |
line: { color: DDR_DATA.modelColors[model] || '#888', width: 2 },
|
| 262 |
marker: { size: 6, color: DDR_DATA.modelColors[model] || '#888' },
|
| 263 |
hovertemplate: `<b>${model}</b><br>Turn: %{customdata}<br>Accuracy: %{y:.2f}%<extra></extra>`,
|
| 264 |
+
customdata: data[model].turns
|
| 265 |
});
|
| 266 |
});
|
| 267 |
|
|
|
|
| 274 |
...darkLayout,
|
| 275 |
xaxis: {
|
| 276 |
...darkLayout.xaxis,
|
| 277 |
+
title: { text: 'Number of Interaction Turns', font: { size: 15, color: '#1d1d1f' } },
|
| 278 |
+
type: 'linear',
|
| 279 |
+
range: [-0.05, 1.05],
|
| 280 |
tickmode: 'array',
|
| 281 |
tickvals: tickVals,
|
| 282 |
ticktext: tickText,
|
|
|
|
| 284 |
},
|
| 285 |
yaxis: {
|
| 286 |
...darkLayout.yaxis,
|
| 287 |
+
title: { text: 'Accuracy (%)', font: { size: 15, color: '#1d1d1f' } },
|
| 288 |
dtick: dtickVal,
|
| 289 |
range: yRange
|
| 290 |
},
|
| 291 |
+
showlegend: false
|
| 292 |
};
|
| 293 |
|
| 294 |
+
// Create chart with markers only first
|
| 295 |
+
Plotly.newPlot(`scaling-${scenario}`, traces, layout, plotlyConfig).then(() => {
|
| 296 |
+
// After a short delay, animate in the lines
|
| 297 |
+
setTimeout(() => {
|
| 298 |
+
animateScalingLinesIn(`scaling-${scenario}`, models, data, normTurns);
|
| 299 |
+
}, 300);
|
| 300 |
+
});
|
| 301 |
});
|
| 302 |
|
| 303 |
// Populate shared legend with models from first scenario
|
|
|
|
| 308 |
}
|
| 309 |
|
| 310 |
// Apply hover effects after charts are rendered
|
| 311 |
+
setTimeout(() => applyHoverEffectsForSection('scaling'), 500);
|
| 312 |
+
}
|
| 313 |
+
|
| 314 |
+
// Animate lines drawing in for scaling charts
|
| 315 |
+
function animateScalingLinesIn(containerId, models, data, normTurns) {
|
| 316 |
+
const graphDiv = document.getElementById(containerId);
|
| 317 |
+
if (!graphDiv) return;
|
| 318 |
+
|
| 319 |
+
// Update to show lines+markers
|
| 320 |
+
let offset = 0;
|
| 321 |
+
const tracesWithLines = models.map(model => {
|
| 322 |
+
const len = data[model].turns.length;
|
| 323 |
+
const modelNormX = normTurns.slice(offset, offset + len);
|
| 324 |
+
offset += len;
|
| 325 |
+
|
| 326 |
+
return {
|
| 327 |
+
x: modelNormX,
|
| 328 |
+
y: data[model].accuracy,
|
| 329 |
+
mode: 'lines+markers',
|
| 330 |
+
name: model,
|
| 331 |
+
line: { color: DDR_DATA.modelColors[model] || '#888', width: 2 },
|
| 332 |
+
marker: { size: 6, color: DDR_DATA.modelColors[model] || '#888' },
|
| 333 |
+
hovertemplate: `<b>${model}</b><br>Turn: %{customdata}<br>Accuracy: %{y:.2f}%<extra></extra>`,
|
| 334 |
+
customdata: data[model].turns
|
| 335 |
+
};
|
| 336 |
+
});
|
| 337 |
+
|
| 338 |
+
// First, add lines with opacity 0
|
| 339 |
+
Plotly.react(containerId, tracesWithLines, graphDiv.layout, plotlyConfig).then(() => {
|
| 340 |
+
// Get all line paths
|
| 341 |
+
const paths = graphDiv.querySelectorAll('.scatterlayer .trace .lines path');
|
| 342 |
+
|
| 343 |
+
// Set initial state: lines hidden via stroke-dashoffset
|
| 344 |
+
paths.forEach((path) => {
|
| 345 |
+
const len = path.getTotalLength();
|
| 346 |
+
if (len > 0) {
|
| 347 |
+
path.style.transition = 'none';
|
| 348 |
+
path.style.strokeDasharray = len + ' ' + len;
|
| 349 |
+
path.style.strokeDashoffset = len;
|
| 350 |
+
}
|
| 351 |
+
});
|
| 352 |
+
|
| 353 |
+
// Force reflow
|
| 354 |
+
graphDiv.getBoundingClientRect();
|
| 355 |
+
|
| 356 |
+
// Animate the lines drawing in with staggered delay
|
| 357 |
+
requestAnimationFrame(() => {
|
| 358 |
+
paths.forEach((path, index) => {
|
| 359 |
+
const len = path.getTotalLength();
|
| 360 |
+
if (len > 0) {
|
| 361 |
+
// Stagger the animation for each line
|
| 362 |
+
const delay = index * 80; // 80ms delay between each line
|
| 363 |
+
path.style.transition = `stroke-dashoffset 0.8s ease-out ${delay}ms`;
|
| 364 |
+
path.style.strokeDashoffset = '0';
|
| 365 |
+
}
|
| 366 |
+
});
|
| 367 |
+
});
|
| 368 |
+
});
|
| 369 |
}
|
| 370 |
|
| 371 |
function updateScalingCharts(dimension) {
|
|
|
|
| 1292 |
// ============================================================================
|
| 1293 |
document.addEventListener('DOMContentLoaded', () => {
|
| 1294 |
// Register all sections for lazy loading
|
| 1295 |
+
// Charts will only be initialized when they become visible
|
| 1296 |
const sections = document.querySelectorAll('section.section');
|
| 1297 |
sections.forEach(section => {
|
| 1298 |
lazyLoadObserver.observe(section);
|
| 1299 |
});
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1300 |
});
|
| 1301 |
|
| 1302 |
// Handle window resize with longer debounce for better performance
|
index.html
CHANGED
|
@@ -137,8 +137,12 @@
|
|
| 137 |
</div>
|
| 138 |
<div class="framework-grid">
|
| 139 |
<div class="framework-card">
|
| 140 |
-
<
|
| 141 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 142 |
<h3>Task Formulation</h3>
|
| 143 |
<p class="framework-description">A case of Claude Sonnet 4.5's trajectory and evaluation checklist
|
| 144 |
in the MIMIC scenario of DDR-Bench. Verified fact and supporting insights are
|
|
@@ -147,8 +151,12 @@
|
|
| 147 |
</p>
|
| 148 |
</div>
|
| 149 |
<div class="framework-card">
|
| 150 |
-
<
|
| 151 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 152 |
<h3>Evaluation Pipeline</h3>
|
| 153 |
<p class="framework-description"><b>Left</b>: Compared with previous tasks, <i>DDR</i> maximises
|
| 154 |
exploration openness and agency, focusing on the direct evaluation of insight quality.
|
|
@@ -597,6 +605,7 @@
|
|
| 597 |
const cards = Array.from(track.querySelectorAll('.carousel-card'));
|
| 598 |
const cardCount = cards.length;
|
| 599 |
let currentIndex = 0;
|
|
|
|
| 600 |
|
| 601 |
// Create dots
|
| 602 |
for (let i = 0; i < cardCount; i++) {
|
|
@@ -689,12 +698,34 @@
|
|
| 689 |
const carouselWrapper = document.querySelector('.carousel-wrapper');
|
| 690 |
if (carouselWrapper) {
|
| 691 |
carouselWrapper.addEventListener('mouseenter', stopAutoPlay);
|
| 692 |
-
carouselWrapper.addEventListener('mouseleave',
|
|
|
|
|
|
|
| 693 |
}
|
| 694 |
|
| 695 |
-
// Initialize
|
| 696 |
updateCarousel();
|
| 697 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 698 |
});
|
| 699 |
</script>
|
| 700 |
|
|
|
|
| 137 |
</div>
|
| 138 |
<div class="framework-grid">
|
| 139 |
<div class="framework-card">
|
| 140 |
+
<div class="framework-img-wrapper">
|
| 141 |
+
<div class="skeleton-loader"></div>
|
| 142 |
+
<img src="assets/framework_task.png" alt="Task Formulation Framework" class="framework-img"
|
| 143 |
+
loading="lazy"
|
| 144 |
+
onload="this.classList.add('loaded'); this.previousElementSibling.style.display='none';">
|
| 145 |
+
</div>
|
| 146 |
<h3>Task Formulation</h3>
|
| 147 |
<p class="framework-description">A case of Claude Sonnet 4.5's trajectory and evaluation checklist
|
| 148 |
in the MIMIC scenario of DDR-Bench. Verified fact and supporting insights are
|
|
|
|
| 151 |
</p>
|
| 152 |
</div>
|
| 153 |
<div class="framework-card">
|
| 154 |
+
<div class="framework-img-wrapper">
|
| 155 |
+
<div class="skeleton-loader"></div>
|
| 156 |
+
<img src="assets/framework_pipeline.png" alt="Evaluation Pipeline Framework"
|
| 157 |
+
class="framework-img" loading="lazy"
|
| 158 |
+
onload="this.classList.add('loaded'); this.previousElementSibling.style.display='none';">
|
| 159 |
+
</div>
|
| 160 |
<h3>Evaluation Pipeline</h3>
|
| 161 |
<p class="framework-description"><b>Left</b>: Compared with previous tasks, <i>DDR</i> maximises
|
| 162 |
exploration openness and agency, focusing on the direct evaluation of insight quality.
|
|
|
|
| 605 |
const cards = Array.from(track.querySelectorAll('.carousel-card'));
|
| 606 |
const cardCount = cards.length;
|
| 607 |
let currentIndex = 0;
|
| 608 |
+
let hasStartedAutoPlay = false;
|
| 609 |
|
| 610 |
// Create dots
|
| 611 |
for (let i = 0; i < cardCount; i++) {
|
|
|
|
| 698 |
const carouselWrapper = document.querySelector('.carousel-wrapper');
|
| 699 |
if (carouselWrapper) {
|
| 700 |
carouselWrapper.addEventListener('mouseenter', stopAutoPlay);
|
| 701 |
+
carouselWrapper.addEventListener('mouseleave', () => {
|
| 702 |
+
if (hasStartedAutoPlay) startAutoPlay();
|
| 703 |
+
});
|
| 704 |
}
|
| 705 |
|
| 706 |
+
// Initialize carousel position (but don't auto-play yet)
|
| 707 |
updateCarousel();
|
| 708 |
+
|
| 709 |
+
// Use Intersection Observer to start auto-play only when visible
|
| 710 |
+
const resultsSection = document.getElementById('results');
|
| 711 |
+
if (resultsSection) {
|
| 712 |
+
const carouselObserver = new IntersectionObserver((entries) => {
|
| 713 |
+
entries.forEach(entry => {
|
| 714 |
+
if (entry.isIntersecting) {
|
| 715 |
+
if (!hasStartedAutoPlay) {
|
| 716 |
+
hasStartedAutoPlay = true;
|
| 717 |
+
}
|
| 718 |
+
startAutoPlay();
|
| 719 |
+
} else {
|
| 720 |
+
stopAutoPlay();
|
| 721 |
+
}
|
| 722 |
+
});
|
| 723 |
+
}, {
|
| 724 |
+
threshold: 0.3 // Start when 30% visible
|
| 725 |
+
});
|
| 726 |
+
|
| 727 |
+
carouselObserver.observe(resultsSection);
|
| 728 |
+
}
|
| 729 |
});
|
| 730 |
</script>
|
| 731 |
|
styles.css
CHANGED
|
@@ -495,6 +495,12 @@ img {
|
|
| 495 |
width: 100%;
|
| 496 |
}
|
| 497 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 498 |
/* ========================= */
|
| 499 |
/* FRAMEWORK SECTION */
|
| 500 |
/* ========================= */
|
|
@@ -558,6 +564,51 @@ img {
|
|
| 558 |
backface-visibility: hidden;
|
| 559 |
}
|
| 560 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 561 |
.framework-description {
|
| 562 |
font-size: 16px;
|
| 563 |
line-height: 1.6;
|
|
|
|
| 495 |
width: 100%;
|
| 496 |
}
|
| 497 |
|
| 498 |
+
.chart-container-benchmarking {
|
| 499 |
+
height: 850px;
|
| 500 |
+
width: 100%;
|
| 501 |
+
}
|
| 502 |
+
|
| 503 |
+
|
| 504 |
/* ========================= */
|
| 505 |
/* FRAMEWORK SECTION */
|
| 506 |
/* ========================= */
|
|
|
|
| 564 |
backface-visibility: hidden;
|
| 565 |
}
|
| 566 |
|
| 567 |
+
/* Framework Image Loading Effect */
|
| 568 |
+
.framework-img-wrapper {
|
| 569 |
+
position: relative;
|
| 570 |
+
width: 100%;
|
| 571 |
+
min-height: 400px;
|
| 572 |
+
border-radius: var(--radius-md);
|
| 573 |
+
overflow: hidden;
|
| 574 |
+
}
|
| 575 |
+
|
| 576 |
+
.framework-img {
|
| 577 |
+
opacity: 0;
|
| 578 |
+
transition: opacity 0.6s ease-in-out;
|
| 579 |
+
}
|
| 580 |
+
|
| 581 |
+
.framework-img.loaded {
|
| 582 |
+
opacity: 1;
|
| 583 |
+
}
|
| 584 |
+
|
| 585 |
+
/* Skeleton Loader Animation */
|
| 586 |
+
.skeleton-loader {
|
| 587 |
+
position: absolute;
|
| 588 |
+
top: 0;
|
| 589 |
+
left: 0;
|
| 590 |
+
right: 0;
|
| 591 |
+
bottom: 0;
|
| 592 |
+
background: linear-gradient(90deg,
|
| 593 |
+
#f0f0f2 0%,
|
| 594 |
+
#e8e8ed 20%,
|
| 595 |
+
#f0f0f2 40%,
|
| 596 |
+
#f0f0f2 100%);
|
| 597 |
+
background-size: 200% 100%;
|
| 598 |
+
animation: skeleton-shimmer 1.5s ease-in-out infinite;
|
| 599 |
+
border-radius: var(--radius-md);
|
| 600 |
+
}
|
| 601 |
+
|
| 602 |
+
@keyframes skeleton-shimmer {
|
| 603 |
+
0% {
|
| 604 |
+
background-position: 200% 0;
|
| 605 |
+
}
|
| 606 |
+
|
| 607 |
+
100% {
|
| 608 |
+
background-position: -200% 0;
|
| 609 |
+
}
|
| 610 |
+
}
|
| 611 |
+
|
| 612 |
.framework-description {
|
| 613 |
font-size: 16px;
|
| 614 |
line-height: 1.6;
|