{
  "id": 392226,
  "title": "41st solution(2D-CNN, 1D-CNN, Stacking)",
  "url": "/competitions/nfl-player-contact-detection/writeups/taro-pan-41st-solution-2d-cnn-1d-cnn-stacking",
  "author_name": "",
  "post_date": "2023-03-04T09:09:21.490Z",
  "votes": 13,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Thank you to all the organizers and participants for this amazing competition!<br>\nI really enjoyed this competion!</p>\n<h1>Overview</h1>\n<ul>\n<li>2D-CNN prediction [zoom out, zoom in, masked]  <br>\n  input single frame, predict Endzone &amp; Sideline by same model</li>\n<li>1D-CNN prediction [tracking, helmet]<br>\n input tracking and helmet position data shift(-6~6)<br>\n (pos, speed, acc,distance, orientation, direction, sa, helmet position)</li>\n<li>Stacking<br>\npredcitions and 5 features by 2D-CNN (zoom out, zoom in, masked = 3 models)<br>\n3 tracking features by 1D-CNN <br>\ntabel features</li>\n<li>Moving average post processing<br>\nafter concat distance &gt; 2 data  </li>\n</ul>\n<h1>Score</h1>\n<ul>\n<li>CV(Group K fold by game_play) : 0.740</li>\n<li>Publie LB : 0.73699</li>\n<li>Private LB : 0.7302</li>\n</ul>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8492034%2Fa09fc34dbc377600c780166cae5bfc22%2FNFL_Solutions.png?generation=1677920953965547&amp;alt=media\" alt=\"\"></p>",
  "messages": [
    {
      "id": "2168457",
      "postDate": "03/04/2023 09:03:36",
      "content": "<p>Thank you to all the organizers and participants for this amazing competition!<br>\nI really enjoyed this competion!</p>\n<h1>Overview</h1>\n<ul>\n<li>2D-CNN prediction [zoom out, zoom in, masked]  <br>\n  input single frame, predict Endzone &amp; Sideline by same model</li>\n<li>1D-CNN prediction [tracking, helmet]<br>\n input tracking and helmet position data shift(-6~6)<br>\n (pos, speed, acc,distance, orientation, direction, sa, helmet position)</li>\n<li>Stacking<br>\npredcitions and 5 features by 2D-CNN (zoom out, zoom in, masked = 3 models)<br>\n3 tracking features by 1D-CNN <br>\ntabel features</li>\n<li>Moving average post processing<br>\nafter concat distance &gt; 2 data  </li>\n</ul>\n<h1>Score</h1>\n<ul>\n<li>CV(Group K fold by game_play) : 0.740</li>\n<li>Publie LB : 0.73699</li>\n<li>Private LB : 0.7302</li>\n</ul>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8492034%2Fa09fc34dbc377600c780166cae5bfc22%2FNFL_Solutions.png?generation=1677920953965547&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "Thank you to all the organizers and participants for this amazing competition!\nI really enjoyed this competion!\n\n# Overview\n- 2D-CNN prediction [zoom out, zoom in, masked]  \n      input single frame, predict Endzone & Sideline by same model\n-  1D-CNN prediction [tracking, helmet]\n     input tracking and helmet position data shift(-6~6)\n     (pos, speed, acc,distance, orientation, direction, sa, helmet position)\n- Stacking\n    predcitions and 5 features by 2D-CNN (zoom out, zoom in, masked = 3 models)\n    3 tracking features by 1D-CNN \n    tabel features\n- Moving average post processing\n    after concat distance > 2 data  \n\n# Score\n- CV(Group K fold by game_play) : 0.740\n- Publie LB : 0.73699\n- Private LB : 0.7302\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8492034%2Fa09fc34dbc377600c780166cae5bfc22%2FNFL_Solutions.png?generation=1677920953965547&alt=media)",
      "votes": null
    },
    {
      "id": "2169078",
      "postDate": "03/04/2023 19:00:23",
      "content": "<p>Congrats <a href=\"https://www.kaggle.com/stgkrtua\" target=\"_blank\">@stgkrtua</a>! Nice solution and write up!</p>",
      "rawMarkdown": "Congrats @stgkrtua! Nice solution and write up!",
      "votes": null
    },
    {
      "id": "2170046",
      "postDate": "03/05/2023 16:33:36",
      "content": "<p>very clear model!</p>",
      "rawMarkdown": "very clear model!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2169078,
      "author_name": "ravishah1",
      "author_url": "",
      "post_date": "03/04/2023 19:00:23",
      "content": "<p>Congrats <a href=\"https://www.kaggle.com/stgkrtua\" target=\"_blank\">@stgkrtua</a>! Nice solution and write up!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2170046,
      "author_name": "chg0901",
      "author_url": "",
      "post_date": "03/05/2023 16:33:36",
      "content": "<p>very clear model!</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2168457": "Thank you to all the organizers and participants for this amazing competition!\nI really enjoyed this competion!\n\n# Overview\n- 2D-CNN prediction [zoom out, zoom in, masked]  \n      input single frame, predict Endzone & Sideline by same model\n-  1D-CNN prediction [tracking, helmet]\n     input tracking and helmet position data shift(-6~6)\n     (pos, speed, acc,distance, orientation, direction, sa, helmet position)\n- Stacking\n    predcitions and 5 features by 2D-CNN (zoom out, zoom in, masked = 3 models)\n    3 tracking features by 1D-CNN \n    tabel features\n- Moving average post processing\n    after concat distance > 2 data  \n\n# Score\n- CV(Group K fold by game_play) : 0.740\n- Publie LB : 0.73699\n- Private LB : 0.7302\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8492034%2Fa09fc34dbc377600c780166cae5bfc22%2FNFL_Solutions.png?generation=1677920953965547&alt=media)",
    "2169078": "Congrats @stgkrtua! Nice solution and write up!",
    "2170046": "very clear model!"
  },
  "source": "meta"
}