{
  "id": 35230,
  "title": "Suggestion for Two Stage Competitions",
  "url": "/competitions/intel-mobileodt-cervical-cancer-screening/discussion/35230",
  "author_name": "Ryan Munion",
  "post_date": "2017-06-24T18:32:31.965000",
  "votes": 2,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Please release stage 1 test labels 2-4 weeks before start of stage 2.  </p>\n\n<p>Then disallow retraining of models during stage 2. </p>\n\n<p>If I had mined the leaderboard and tuned my models accordingly,  After the release of the test labels I discovered that I needed to halve the learning rate on my models in order to converge to a decent training loss.  Unfortunately, even such a trivial change was disallowed under the current rules.  In the last few days I retrained all my models with the new learning rate and I scored 0.813</p>\n\n<p>I dislike the thought of mining the leaderboard for test labels.  Of course, even more so, I dislike having to retrain my models and evaluate the stage 2 test set in a single week.  It is needlessly stressful.</p>",
  "messages": [
    {
      "id": 195727,
      "postDate": "2017-06-24T18:32:31.967Z",
      "content": "<p>Please release stage 1 test labels 2-4 weeks before start of stage 2.  </p>\n\n<p>Then disallow retraining of models during stage 2. </p>\n\n<p>If I had mined the leaderboard and tuned my models accordingly,  After the release of the test labels I discovered that I needed to halve the learning rate on my models in order to converge to a decent training loss.  Unfortunately, even such a trivial change was disallowed under the current rules.  In the last few days I retrained all my models with the new learning rate and I scored 0.813</p>\n\n<p>I dislike the thought of mining the leaderboard for test labels.  Of course, even more so, I dislike having to retrain my models and evaluate the stage 2 test set in a single week.  It is needlessly stressful.</p>",
      "rawMarkdown": "Please release stage 1 test labels 2-4 weeks before start of stage 2.  \n\nThen disallow retraining of models during stage 2. \n\nIf I had mined the leaderboard and tuned my models accordingly,  After the release of the test labels I discovered that I needed to halve the learning rate on my models in order to converge to a decent training loss.  Unfortunately, even such a trivial change was disallowed under the current rules.  In the last few days I retrained all my models with the new learning rate and I scored 0.813\n\nI dislike the thought of mining the leaderboard for test labels.  Of course, even more so, I dislike having to retrain my models and evaluate the stage 2 test set in a single week.  It is needlessly stressful.",
      "votes": 2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "195727": "Please release stage 1 test labels 2-4 weeks before start of stage 2.  \n\nThen disallow retraining of models during stage 2. \n\nIf I had mined the leaderboard and tuned my models accordingly,  After the release of the test labels I discovered that I needed to halve the learning rate on my models in order to converge to a decent training loss.  Unfortunately, even such a trivial change was disallowed under the current rules.  In the last few days I retrained all my models with the new learning rate and I scored 0.813\n\nI dislike the thought of mining the leaderboard for test labels.  Of course, even more so, I dislike having to retrain my models and evaluate the stage 2 test set in a single week.  It is needlessly stressful."
  }
}