NSchuetz commited on
Commit
86c948b
·
verified ·
1 Parent(s): 41be75e

Fix Track 1 description: not 168-hour sensor embeddings

Browse files
Files changed (1) hide show
  1. app.py +4 -3
app.py CHANGED
@@ -133,8 +133,9 @@ _FORECASTING_NOTE = (
133
  "Source: <code>MyHeartCounts/OpenMHC-leaderboard-data</code>."
134
  )
135
  _DOWNSTREAM_NOTE = (
136
- "<b>Metric legend</b> — Track 1 predicts weekly health outcomes from 168-hour "
137
- "sensor embeddings; scores are computed vs the <b>Linear</b> baseline.<br>"
 
138
  "<b>R</b> (Average Rank) — mean cross-method rank across the outcome tasks; "
139
  "<b>1 = best</b> (lower is better).<br>"
140
  "<b>S</b> (Skill Score) — category-balanced % improvement over Linear across tasks "
@@ -371,7 +372,7 @@ thead th[data-tip]{position:relative;}
371
  <div class="hero">
372
  <p class="eyebrow">OpenMHC Benchmark</p>
373
  <h1>OpenMHC Leaderboard</h1>
374
- <p class="lede">Wearable &amp; mobile health benchmark on MyHeartCounts. Track 1 (predictive tasks) predicts weekly health outcomes &mdash; demographics, medical risk, vitals, mental well-being, and lifestyle &mdash; from 168-hour sensor embeddings. The generative tasks operate on the raw signals: Track 2a (imputation) reconstructs masked daily, minute-level signals, and Track 2b (forecasting) predicts future hourly signals. Each method is ranked by skill score vs a track baseline, computed live from the per-user evaluation substrate.</p>
375
  <div class="pills">
376
  <a class="pill primary" href="%%SUBMIT%%" target="_blank" rel="noopener">&#128228; Submit a model</a>
377
  <a class="pill" href="%%CODE%%" target="_blank" rel="noopener">&#9881;&#65039; Code</a>
 
133
  "Source: <code>MyHeartCounts/OpenMHC-leaderboard-data</code>."
134
  )
135
  _DOWNSTREAM_NOTE = (
136
+ "<b>Metric legend</b> — Track 1 predicts one self-reported health outcome per "
137
+ "participant from their wearable history; scores are computed vs the <b>Linear</b> "
138
+ "baseline.<br>"
139
  "<b>R</b> (Average Rank) — mean cross-method rank across the outcome tasks; "
140
  "<b>1 = best</b> (lower is better).<br>"
141
  "<b>S</b> (Skill Score) — category-balanced % improvement over Linear across tasks "
 
372
  <div class="hero">
373
  <p class="eyebrow">OpenMHC Benchmark</p>
374
  <h1>OpenMHC Leaderboard</h1>
375
+ <p class="lede">Wearable &amp; mobile health benchmark on MyHeartCounts. Track 1 (predictive tasks) predicts self-reported health outcomes &mdash; demographics, medical risk, vitals, mental well-being, and lifestyle &mdash; from a participant's wearable history, with each model free to choose its own temporal resolution and context length. The generative tasks: Track 2a (imputation) reconstructs masked daily, minute-level signals, and Track 2b (forecasting) predicts future hourly signals. Each method is ranked by skill score vs a track baseline, computed live from the per-user evaluation substrate.</p>
376
  <div class="pills">
377
  <a class="pill primary" href="%%SUBMIT%%" target="_blank" rel="noopener">&#128228; Submit a model</a>
378
  <a class="pill" href="%%CODE%%" target="_blank" rel="noopener">&#9881;&#65039; Code</a>