{
  "id": 89580,
  "title": "Is it allowed in this competition?",
  "url": "/competitions/freesound-audio-tagging-2019/discussion/89580",
  "author_name": "",
  "post_date": "2019-04-16T01:28:52.804535500Z",
  "votes": null,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I'm somewhat confused because it is my first kernel-only competition.</p>\n\n<p>Please verify my plan, Thanks!</p>\n\n<p>My strategy  here.</p>\n\n<p>(1). Develop my own model(not any resnet series) </p>\n\n<p>(2) Train the model with competition datasets (not using any external data) </p>\n\n<ul>\n<li>I will train the model by using my GPU (offline)</li>\n</ul>\n\n<p>(3).  Upload weights on the kernel in this competition</p>\n\n<p>(4)  Not train in kernel. Just predict the labels of testset using uploaded weight.</p>\n\n<p>Is this whole process(1) -  (4) OK for this competition?</p>\n\n<p>Thanks!</p>",
  "messages": [
    {
      "id": "517389",
      "postDate": "04/16/2019 01:28:52",
      "content": "<p>I'm somewhat confused because it is my first kernel-only competition.</p>\n\n<p>Please verify my plan, Thanks!</p>\n\n<p>My strategy  here.</p>\n\n<p>(1). Develop my own model(not any resnet series) </p>\n\n<p>(2) Train the model with competition datasets (not using any external data) </p>\n\n<ul>\n<li>I will train the model by using my GPU (offline)</li>\n</ul>\n\n<p>(3).  Upload weights on the kernel in this competition</p>\n\n<p>(4)  Not train in kernel. Just predict the labels of testset using uploaded weight.</p>\n\n<p>Is this whole process(1) -  (4) OK for this competition?</p>\n\n<p>Thanks!</p>",
      "rawMarkdown": "I'm somewhat confused because it is my first kernel-only competition.\n\nPlease verify my plan, Thanks!\n\nMy strategy  here.\n\n(1). Develop my own model(not any resnet series) \n\n(2) Train the model with competition datasets (not using any external data) \n\n- I will train the model by using my GPU (offline)\n\n(3).  Upload weights on the kernel in this competition\n\n(4)  Not train in kernel. Just predict the labels of testset using uploaded weight.\n\n Is this whole process(1) -  (4) OK for this competition?\n\nThanks!",
      "votes": null
    },
    {
      "id": "517412",
      "postDate": "04/16/2019 01:59:44",
      "content": "<p>That's how it should be done. Note that in the second stage the test set will be swapped with another unseen set, so you must include any preprocessing (e.g generating melspectrograms) on the test data in your submission kernel.</p>",
      "rawMarkdown": "That's how it should be done. Note that in the second stage the test set will be swapped with another unseen set, so you must include any preprocessing (e.g generating melspectrograms) on the test data in your submission kernel.",
      "votes": null
    },
    {
      "id": "517422",
      "postDate": "04/16/2019 02:20:50",
      "content": "<p>Ok, Thanks. I read the fact that new test set is larger than test set(now) by 3 times. \nAs you said, I need to be careful to deal execution time. Thanks! \nHope you get a good score!</p>",
      "rawMarkdown": "Ok, Thanks. I read the fact that new test set is larger than test set(now) by 3 times. \nAs you said, I need to be careful to deal execution time. Thanks! \nHope you get a good score!",
      "votes": null
    },
    {
      "id": "517758",
      "postDate": "04/16/2019 13:31:43",
      "content": "<p>This is also my first kernel-only competition. But by training offline then uploading the weights in a kernel, how can we be sure that nobody is using external data ?</p>\n\n<p>For now I'm doing the training and predictions in a single kernel and in &lt;1h.</p>",
      "rawMarkdown": "This is also my first kernel-only competition. But by training offline then uploading the weights in a kernel, how can we be sure that nobody is using external data ?\n\nFor now I'm doing the training and predictions in a single kernel and in &lt;1h.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 517412,
      "author_name": "suicaokhoailang",
      "author_url": "",
      "post_date": "04/16/2019 01:59:44",
      "content": "<p>That's how it should be done. Note that in the second stage the test set will be swapped with another unseen set, so you must include any preprocessing (e.g generating melspectrograms) on the test data in your submission kernel.</p>",
      "votes": null,
      "replies": [
        {
          "id": 517422,
          "author_name": "youhanlee",
          "author_url": "",
          "post_date": "04/16/2019 02:20:50",
          "content": "<p>Ok, Thanks. I read the fact that new test set is larger than test set(now) by 3 times. \nAs you said, I need to be careful to deal execution time. Thanks! \nHope you get a good score!</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 517758,
      "author_name": "nathanh12",
      "author_url": "",
      "post_date": "04/16/2019 13:31:43",
      "content": "<p>This is also my first kernel-only competition. But by training offline then uploading the weights in a kernel, how can we be sure that nobody is using external data ?</p>\n\n<p>For now I'm doing the training and predictions in a single kernel and in &lt;1h.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "517389": "I'm somewhat confused because it is my first kernel-only competition.\n\nPlease verify my plan, Thanks!\n\nMy strategy  here.\n\n(1). Develop my own model(not any resnet series) \n\n(2) Train the model with competition datasets (not using any external data) \n\n- I will train the model by using my GPU (offline)\n\n(3).  Upload weights on the kernel in this competition\n\n(4)  Not train in kernel. Just predict the labels of testset using uploaded weight.\n\n Is this whole process(1) -  (4) OK for this competition?\n\nThanks!",
    "517412": "That's how it should be done. Note that in the second stage the test set will be swapped with another unseen set, so you must include any preprocessing (e.g generating melspectrograms) on the test data in your submission kernel.",
    "517422": "Ok, Thanks. I read the fact that new test set is larger than test set(now) by 3 times. \nAs you said, I need to be careful to deal execution time. Thanks! \nHope you get a good score!",
    "517758": "This is also my first kernel-only competition. But by training offline then uploading the weights in a kernel, how can we be sure that nobody is using external data ?\n\nFor now I'm doing the training and predictions in a single kernel and in &lt;1h."
  },
  "source": "meta"
}