{
  "id": 93542,
  "title": "Question Regarding Model Tuning for Stage 2",
  "url": "/competitions/imet-2019-fgvc6/discussion/93542",
  "author_name": "",
  "post_date": "2019-05-28T03:48:16.757419400Z",
  "votes": 1,
  "comment_count": 4,
  "views": 0,
  "content": "<p>I've come across rules like this in Kaggle contests and was wondering how it applies and whether this applies to this competition too:</p>\n\n<blockquote>\n  <p>In the second phase, they use the same models developed in the first phase to predict on an unseen test set.</p>\n  \n  <blockquote>\n    <p>What happens if I want to change something in my code in the second stage?</p>\n  </blockquote>\n  \n  <p>We expect you may need to make some \"non scientific\" alterations, such as changes to path &gt;names, in order to create your submissions for the second stage. You are allowed to re-train &gt;your model (including the stage one data), but your code should not change. You should not be &gt;doing any hyper parameter tuning in the second stage. Parameter tuning is permitted as long &gt;as it is fully automated.</p>\n</blockquote>\n\n<p>This is one of my first Kaggle contests so pardon me if this question is trivial. What kind of tuning is allowed? Are we allowed to train new architectures or ensembles after the first stage on the 28th? Appreciate any insight here. Thanks!</p>",
  "messages": [
    {
      "id": "538054",
      "postDate": "05/28/2019 03:48:16",
      "content": "<p>I've come across rules like this in Kaggle contests and was wondering how it applies and whether this applies to this competition too:</p>\n\n<blockquote>\n  <p>In the second phase, they use the same models developed in the first phase to predict on an unseen test set.</p>\n  \n  <blockquote>\n    <p>What happens if I want to change something in my code in the second stage?</p>\n  </blockquote>\n  \n  <p>We expect you may need to make some \"non scientific\" alterations, such as changes to path &gt;names, in order to create your submissions for the second stage. You are allowed to re-train &gt;your model (including the stage one data), but your code should not change. You should not be &gt;doing any hyper parameter tuning in the second stage. Parameter tuning is permitted as long &gt;as it is fully automated.</p>\n</blockquote>\n\n<p>This is one of my first Kaggle contests so pardon me if this question is trivial. What kind of tuning is allowed? Are we allowed to train new architectures or ensembles after the first stage on the 28th? Appreciate any insight here. Thanks!</p>",
      "rawMarkdown": "I've come across rules like this in Kaggle contests and was wondering how it applies and whether this applies to this competition too:\n&gt; In the second phase, they use the same models developed in the first phase to predict on an unseen test set.\n\n&gt;&gt;What happens if I want to change something in my code in the second stage?\n\n&gt;We expect you may need to make some \"non scientific\" alterations, such as changes to path &gt;names, in order to create your submissions for the second stage. You are allowed to re-train &gt;your model (including the stage one data), but your code should not change. You should not be &gt;doing any hyper parameter tuning in the second stage. Parameter tuning is permitted as long &gt;as it is fully automated.\n\nThis is one of my first Kaggle contests so pardon me if this question is trivial. What kind of tuning is allowed? Are we allowed to train new architectures or ensembles after the first stage on the 28th? Appreciate any insight here. Thanks!",
      "votes": null
    },
    {
      "id": "538305",
      "postDate": "05/28/2019 11:44:13",
      "content": "<p>No tuning is allowed for this one. You select your two kernels and wait for second stage re-run results. Test data path remains the same. Be aware of time and memory limits and don't hardcode any sample id.</p>",
      "rawMarkdown": "No tuning is allowed for this one. You select your two kernels and wait for second stage re-run results. Test data path remains the same. Be aware of time and memory limits and don't hardcode any sample id.",
      "votes": null
    },
    {
      "id": "538351",
      "postDate": "05/28/2019 12:57:43",
      "content": "<p>got it! Thanks :)</p>",
      "rawMarkdown": "got it! Thanks :)",
      "votes": null
    },
    {
      "id": "538386",
      "postDate": "05/28/2019 14:04:56",
      "content": "<p><a href=\"/yaroshevskiy\">@yaroshevskiy</a>  One more question. So is the 28th the day to freeze the kernels for submission for stage 2 or is it June 4th?</p>",
      "rawMarkdown": "yaroshevskiy  One more question. So is the 28th the day to freeze the kernels for submission for stage 2 or is it June 4th?",
      "votes": null
    },
    {
      "id": "538411",
      "postDate": "05/28/2019 14:37:34",
      "content": "<p>4th June</p>",
      "rawMarkdown": "4th June",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 538305,
      "author_name": "yaroshevskiy",
      "author_url": "",
      "post_date": "05/28/2019 11:44:13",
      "content": "<p>No tuning is allowed for this one. You select your two kernels and wait for second stage re-run results. Test data path remains the same. Be aware of time and memory limits and don't hardcode any sample id.</p>",
      "votes": null,
      "replies": [
        {
          "id": 538351,
          "author_name": "sairam6087",
          "author_url": "",
          "post_date": "05/28/2019 12:57:43",
          "content": "<p>got it! Thanks :)</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 538386,
          "author_name": "sairam6087",
          "author_url": "",
          "post_date": "05/28/2019 14:04:56",
          "content": "<p><a href=\"/yaroshevskiy\">@yaroshevskiy</a>  One more question. So is the 28th the day to freeze the kernels for submission for stage 2 or is it June 4th?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 538411,
          "author_name": "yaroshevskiy",
          "author_url": "",
          "post_date": "05/28/2019 14:37:34",
          "content": "<p>4th June</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "538054": "I've come across rules like this in Kaggle contests and was wondering how it applies and whether this applies to this competition too:\n&gt; In the second phase, they use the same models developed in the first phase to predict on an unseen test set.\n\n&gt;&gt;What happens if I want to change something in my code in the second stage?\n\n&gt;We expect you may need to make some \"non scientific\" alterations, such as changes to path &gt;names, in order to create your submissions for the second stage. You are allowed to re-train &gt;your model (including the stage one data), but your code should not change. You should not be &gt;doing any hyper parameter tuning in the second stage. Parameter tuning is permitted as long &gt;as it is fully automated.\n\nThis is one of my first Kaggle contests so pardon me if this question is trivial. What kind of tuning is allowed? Are we allowed to train new architectures or ensembles after the first stage on the 28th? Appreciate any insight here. Thanks!",
    "538305": "No tuning is allowed for this one. You select your two kernels and wait for second stage re-run results. Test data path remains the same. Be aware of time and memory limits and don't hardcode any sample id.",
    "538351": "got it! Thanks :)",
    "538386": "yaroshevskiy  One more question. So is the 28th the day to freeze the kernels for submission for stage 2 or is it June 4th?",
    "538411": "4th June"
  },
  "source": "meta"
}