{
  "id": 97680,
  "title": "Although  I  released the memory at all times, not enough, the kernel always die...sigh~",
  "url": "/competitions/champs-scalar-coupling/discussion/97680",
  "author_name": "",
  "post_date": "2019-06-28T10:41:40.853238400Z",
  "votes": 2,
  "comment_count": 6,
  "views": 0,
  "content": "",
  "messages": [
    {
      "id": "563502",
      "postDate": "06/28/2019 10:41:40",
      "content": "",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "563514",
      "postDate": "06/28/2019 10:58:50",
      "content": "<p>You can split data into 8 separate datasets by type and train models in separate kernels.</p>",
      "rawMarkdown": "You can split data into 8 separate datasets by type and train models in separate kernels.",
      "votes": null
    },
    {
      "id": "563534",
      "postDate": "06/28/2019 11:47:06",
      "content": "<p>That's really a good idea, I just realize that kernels can share the data files,thanks a lot</p>",
      "rawMarkdown": "That's really a good idea, I just realize that kernels can share the data files,thanks a lot",
      "votes": null
    },
    {
      "id": "566199",
      "postDate": "07/01/2019 22:20:25",
      "content": "<p>Hi! <a href=\"/luyouyou\">@luyouyou</a>, have you tried to optimize the type of variables that your data frame is using, also depending on the kind of model that you are training you can use generators to read and feed the data.</p>",
      "rawMarkdown": "Hi! @luyouyou, have you tried to optimize the type of variables that your data frame is using, also depending on the kind of model that you are training you can use generators to read and feed the data.",
      "votes": null
    },
    {
      "id": "580857",
      "postDate": "07/20/2019 22:53:27",
      "content": "<p>Thanks for the hint!</p>\n\n<p>But what will happen with the private dataset? I'd guess they'll provide an extended structures file?!? And the model might include features of test and train (e.g. mean distance per type). \nWill the 8 kernels be executed before a 9th kernels predicts and submits the results? Is there something like a kernel pipeline?</p>",
      "rawMarkdown": "Thanks for the hint!\n\nBut what will happen with the private dataset? I'd guess they'll provide an extended structures file?!? And the model might include features of test and train (e.g. mean distance per type). \nWill the 8 kernels be executed before a 9th kernels predicts and submits the results? Is there something like a kernel pipeline?",
      "votes": null
    },
    {
      "id": "580904",
      "postDate": "07/21/2019 03:28:27",
      "content": "<p>There is no stage 2 in this competition, all the data is already provided.\nAlso this isn't kernel-only competition, so you can submit a file from your computer.</p>\n\n<p>As fast as I know there is no such think as a kernel pipeline, if you train in 8 separate kernels, you'll need to combine predictions by yourself.</p>",
      "rawMarkdown": "There is no stage 2 in this competition, all the data is already provided.\nAlso this isn't kernel-only competition, so you can submit a file from your computer.\n\nAs fast as I know there is no such think as a kernel pipeline, if you train in 8 separate kernels, you'll need to combine predictions by yourself.",
      "votes": null
    },
    {
      "id": "581280",
      "postDate": "07/21/2019 17:43:44",
      "content": "<p>That's good news. I think I got this part wrong too in my first two competitions 🤕 . Always thought they'd run a not published test set over the selected kernels for the private score.\nNow I can train with more than 100 features :D. Thanks a lot!</p>",
      "rawMarkdown": "That's good news. I think I got this part wrong too in my first two competitions 🤕 . Always thought they'd run a not published test set over the selected kernels for the private score.\nNow I can train with more than 100 features :D. Thanks a lot!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 563514,
      "author_name": "artgor",
      "author_url": "",
      "post_date": "06/28/2019 10:58:50",
      "content": "<p>You can split data into 8 separate datasets by type and train models in separate kernels.</p>",
      "votes": null,
      "replies": [
        {
          "id": 580857,
          "author_name": "joatom",
          "author_url": "",
          "post_date": "07/20/2019 22:53:27",
          "content": "<p>Thanks for the hint!</p>\n\n<p>But what will happen with the private dataset? I'd guess they'll provide an extended structures file?!? And the model might include features of test and train (e.g. mean distance per type). \nWill the 8 kernels be executed before a 9th kernels predicts and submits the results? Is there something like a kernel pipeline?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 580904,
          "author_name": "artgor",
          "author_url": "",
          "post_date": "07/21/2019 03:28:27",
          "content": "<p>There is no stage 2 in this competition, all the data is already provided.\nAlso this isn't kernel-only competition, so you can submit a file from your computer.</p>\n\n<p>As fast as I know there is no such think as a kernel pipeline, if you train in 8 separate kernels, you'll need to combine predictions by yourself.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 581280,
          "author_name": "joatom",
          "author_url": "",
          "post_date": "07/21/2019 17:43:44",
          "content": "<p>That's good news. I think I got this part wrong too in my first two competitions 🤕 . Always thought they'd run a not published test set over the selected kernels for the private score.\nNow I can train with more than 100 features :D. Thanks a lot!</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 563534,
      "author_name": "luyouyou",
      "author_url": "",
      "post_date": "06/28/2019 11:47:06",
      "content": "<p>That's really a good idea, I just realize that kernels can share the data files,thanks a lot</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 566199,
      "author_name": "cv13j0",
      "author_url": "",
      "post_date": "07/01/2019 22:20:25",
      "content": "<p>Hi! <a href=\"/luyouyou\">@luyouyou</a>, have you tried to optimize the type of variables that your data frame is using, also depending on the kind of model that you are training you can use generators to read and feed the data.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "563502": "",
    "563514": "You can split data into 8 separate datasets by type and train models in separate kernels.",
    "563534": "That's really a good idea, I just realize that kernels can share the data files,thanks a lot",
    "566199": "Hi! @luyouyou, have you tried to optimize the type of variables that your data frame is using, also depending on the kind of model that you are training you can use generators to read and feed the data.",
    "580857": "Thanks for the hint!\n\nBut what will happen with the private dataset? I'd guess they'll provide an extended structures file?!? And the model might include features of test and train (e.g. mean distance per type). \nWill the 8 kernels be executed before a 9th kernels predicts and submits the results? Is there something like a kernel pipeline?",
    "580904": "There is no stage 2 in this competition, all the data is already provided.\nAlso this isn't kernel-only competition, so you can submit a file from your computer.\n\nAs fast as I know there is no such think as a kernel pipeline, if you train in 8 separate kernels, you'll need to combine predictions by yourself.",
    "581280": "That's good news. I think I got this part wrong too in my first two competitions 🤕 . Always thought they'd run a not published test set over the selected kernels for the private score.\nNow I can train with more than 100 features :D. Thanks a lot!"
  },
  "source": "meta"
}