{
  "id": 379265,
  "title": "How long does your submission take?",
  "url": "/competitions/nfl-player-contact-detection/discussion/379265",
  "author_name": "",
  "post_date": "2023-01-18T22:10:14.218857500Z",
  "votes": 18,
  "comment_count": 8,
  "views": 0,
  "content": "<p>Efficiency matters; the faster it is, the more practical.</p>\n<p>My 0.73 sub right now takes about 1.5 hours since it uses some CNN.<br>\nJust wondering how short we can get.</p>",
  "messages": [
    {
      "id": "2106013",
      "postDate": "01/18/2023 22:10:14",
      "content": "<p>Efficiency matters; the faster it is, the more practical.</p>\n<p>My 0.73 sub right now takes about 1.5 hours since it uses some CNN.<br>\nJust wondering how short we can get.</p>",
      "rawMarkdown": "Efficiency matters; the faster it is, the more practical.\n\nMy 0.73 sub right now takes about 1.5 hours since it uses some CNN.\nJust wondering how short we can get.",
      "votes": null
    },
    {
      "id": "2106028",
      "postDate": "01/18/2023 22:38:19",
      "content": "<p>Mine took around 2.5 hours. I have a feeling that the kernels run slower in background compared to actively running them.</p>",
      "rawMarkdown": "Mine took around 2.5 hours. I have a feeling that the kernels run slower in background compared to actively running them.",
      "votes": null
    },
    {
      "id": "2106119",
      "postDate": "01/19/2023 01:27:26",
      "content": "<p>Similar to mine in terms of runtime.</p>\n<p>However, I feel my commits (I assume what you call background) runs much faster than interactive mode, but I am relatively new to kernels so, not much idea :D</p>",
      "rawMarkdown": "Similar to mine in terms of runtime.\n\nHowever, I feel my commits (I assume what you call background) runs much faster than interactive mode, but I am relatively new to kernels so, not much idea :D",
      "votes": null
    },
    {
      "id": "2106541",
      "postDate": "01/19/2023 08:08:58",
      "content": "<p>Currently 30 minutes, but I expect my submission times to increase to 1 - 1.5 hour once more ideas are implemented.</p>",
      "rawMarkdown": "Currently 30 minutes, but I expect my submission times to increase to 1 - 1.5 hour once more ideas are implemented.",
      "votes": null
    },
    {
      "id": "2106559",
      "postDate": "01/19/2023 08:14:59",
      "content": "<p>I got 0.754 public LB. It took 1.5 hours.<br>\nI used 2DCNN.</p>",
      "rawMarkdown": "I got 0.754 public LB. It took 1.5 hours.\nI used 2DCNN.",
      "votes": null
    },
    {
      "id": "2118747",
      "postDate": "01/28/2023 08:38:47",
      "content": "<p>Could you tell me your solution? Classification or detection? I use 2.5D model to classify images and get 0.69, but it's too slow.</p>",
      "rawMarkdown": "Could you tell me your solution? Classification or detection? I use 2.5D model to classify images and get 0.69, but it's too slow.",
      "votes": null
    },
    {
      "id": "2139513",
      "postDate": "02/10/2023 20:10:55",
      "content": "<p>I'm getting timeout on submission after running for 9hrs. It takes only around 30 min when I run it on the ~50K rows of test set and save and run all. Can anyone tell me what's the difference between your \"save and run all (commit)\" run time vs submission run time, so I could get some idea about how big the hidden test set is expected to be. </p>",
      "rawMarkdown": "I'm getting timeout on submission after running for 9hrs. It takes only around 30 min when I run it on the ~50K rows of test set and save and run all. Can anyone tell me what's the difference between your \"save and run all (commit)\" run time vs submission run time, so I could get some idea about how big the hidden test set is expected to be.",
      "votes": null
    },
    {
      "id": "2139528",
      "postDate": "02/10/2023 20:34:41",
      "content": "<p>The data tab says there are 61 unseen plays, and the given test set is 2 plays, so I'd expect (and have seen) the hidden set to take about ~30x longer than however long it takes to run on the 2 sample plays. </p>\n<p>So a 30min sample test runtime would be &gt; 9 hours (probably about 15 hours) on the unseen test set, which is why you're seeing the timeout.</p>",
      "rawMarkdown": "The data tab says there are 61 unseen plays, and the given test set is 2 plays, so I'd expect (and have seen) the hidden set to take about ~30x longer than however long it takes to run on the 2 sample plays. \n\nSo a 30min sample test runtime would be > 9 hours (probably about 15 hours) on the unseen test set, which is why you're seeing the timeout.",
      "votes": null
    },
    {
      "id": "2139535",
      "postDate": "02/10/2023 20:41:51",
      "content": "<p>Thanks. It makes sense now. I'll have to optimize the speed. I tried to look for information about hidden set on data tab but somehow missed it.</p>",
      "rawMarkdown": "Thanks. It makes sense now. I'll have to optimize the speed. I tried to look for information about hidden set on data tab but somehow missed it.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2106028,
      "author_name": "surajdengale",
      "author_url": "",
      "post_date": "01/18/2023 22:38:19",
      "content": "<p>Mine took around 2.5 hours. I have a feeling that the kernels run slower in background compared to actively running them.</p>",
      "votes": null,
      "replies": [
        {
          "id": 2106119,
          "author_name": "carloshuertas",
          "author_url": "",
          "post_date": "01/19/2023 01:27:26",
          "content": "<p>Similar to mine in terms of runtime.</p>\n<p>However, I feel my commits (I assume what you call background) runs much faster than interactive mode, but I am relatively new to kernels so, not much idea :D</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 2106541,
      "author_name": "samfc10",
      "author_url": "",
      "post_date": "01/19/2023 08:08:58",
      "content": "<p>Currently 30 minutes, but I expect my submission times to increase to 1 - 1.5 hour once more ideas are implemented.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2106559,
      "author_name": "shinmurashinmura",
      "author_url": "",
      "post_date": "01/19/2023 08:14:59",
      "content": "<p>I got 0.754 public LB. It took 1.5 hours.<br>\nI used 2DCNN.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2118747,
      "author_name": "behaunted",
      "author_url": "",
      "post_date": "01/28/2023 08:38:47",
      "content": "<p>Could you tell me your solution? Classification or detection? I use 2.5D model to classify images and get 0.69, but it's too slow.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2139513,
      "author_name": "imroze",
      "author_url": "",
      "post_date": "02/10/2023 20:10:55",
      "content": "<p>I'm getting timeout on submission after running for 9hrs. It takes only around 30 min when I run it on the ~50K rows of test set and save and run all. Can anyone tell me what's the difference between your \"save and run all (commit)\" run time vs submission run time, so I could get some idea about how big the hidden test set is expected to be. </p>",
      "votes": null,
      "replies": [
        {
          "id": 2139528,
          "author_name": "chris62",
          "author_url": "",
          "post_date": "02/10/2023 20:34:41",
          "content": "<p>The data tab says there are 61 unseen plays, and the given test set is 2 plays, so I'd expect (and have seen) the hidden set to take about ~30x longer than however long it takes to run on the 2 sample plays. </p>\n<p>So a 30min sample test runtime would be &gt; 9 hours (probably about 15 hours) on the unseen test set, which is why you're seeing the timeout.</p>",
          "votes": null,
          "replies": [
            {
              "id": 2139535,
              "author_name": "imroze",
              "author_url": "",
              "post_date": "02/10/2023 20:41:51",
              "content": "<p>Thanks. It makes sense now. I'll have to optimize the speed. I tried to look for information about hidden set on data tab but somehow missed it.</p>",
              "votes": null,
              "replies": []
            }
          ]
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2106013": "Efficiency matters; the faster it is, the more practical.\n\nMy 0.73 sub right now takes about 1.5 hours since it uses some CNN.\nJust wondering how short we can get.",
    "2106028": "Mine took around 2.5 hours. I have a feeling that the kernels run slower in background compared to actively running them.",
    "2106119": "Similar to mine in terms of runtime.\n\nHowever, I feel my commits (I assume what you call background) runs much faster than interactive mode, but I am relatively new to kernels so, not much idea :D",
    "2106541": "Currently 30 minutes, but I expect my submission times to increase to 1 - 1.5 hour once more ideas are implemented.",
    "2106559": "I got 0.754 public LB. It took 1.5 hours.\nI used 2DCNN.",
    "2118747": "Could you tell me your solution? Classification or detection? I use 2.5D model to classify images and get 0.69, but it's too slow.",
    "2139513": "I'm getting timeout on submission after running for 9hrs. It takes only around 30 min when I run it on the ~50K rows of test set and save and run all. Can anyone tell me what's the difference between your \"save and run all (commit)\" run time vs submission run time, so I could get some idea about how big the hidden test set is expected to be.",
    "2139528": "The data tab says there are 61 unseen plays, and the given test set is 2 plays, so I'd expect (and have seen) the hidden set to take about ~30x longer than however long it takes to run on the 2 sample plays. \n\nSo a 30min sample test runtime would be > 9 hours (probably about 15 hours) on the unseen test set, which is why you're seeing the timeout.",
    "2139535": "Thanks. It makes sense now. I'll have to optimize the speed. I tried to look for information about hidden set on data tab but somehow missed it."
  },
  "source": "meta"
}