{
  "id": 94375,
  "title": "Any other limitations expect the GPU time?",
  "url": "/competitions/imet-2019-fgvc6/discussion/94375",
  "author_name": "Yiheng Wang",
  "post_date": "2019-06-04T06:43:56.850000",
  "votes": 1,
  "comment_count": 7,
  "views": 0,
  "content": "<p>Like storages, CPU time? I cannot find other rules expect the 9 hours' GPU running time.</p>",
  "messages": [
    {
      "id": 542868,
      "postDate": "2019-06-04T06:43:56.850Z",
      "content": "<p>Like storages, CPU time? I cannot find other rules expect the 9 hours' GPU running time.</p>",
      "rawMarkdown": "Like storages, CPU time? I cannot find other rules expect the 9 hours' GPU running time.",
      "votes": 1
    },
    {
      "id": 542893,
      "postDate": "2019-06-04T07:06:06.780Z",
      "content": "<p><a href=\"https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/discussion/87719#latest-536719\">https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/discussion/87719#latest-536719</a>\nI just found this post, does it mean we can use as many trained models as possible?</p>",
      "rawMarkdown": "https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/discussion/87719#latest-536719\nI just found this post, does it mean we can use as many trained models as possible?",
      "replies": [
        {
          "id": 542957,
          "postDate": "2019-06-04T08:27:04.960Z",
          "content": "<p>I think so. You can build a huge ensemble as long as time limit allows.</p>",
          "rawMarkdown": "I think so. You can build a huge ensemble as long as time limit allows.",
          "votes": 1
        },
        {
          "id": 543163,
          "postDate": "2019-06-04T11:04:37.870Z",
          "content": "<p>What made you think the number of trained models might be limited?</p>",
          "rawMarkdown": "What made you think the number of trained models might be limited?"
        }
      ]
    },
    {
      "id": 542870,
      "postDate": "2019-06-04T06:45:09.493Z",
      "content": "<p>What if we meet memory error? </p>",
      "rawMarkdown": "What if we meet memory error? ",
      "replies": [
        {
          "id": 543142,
          "postDate": "2019-06-04T10:51:35.607Z",
          "content": "<p>Sure, there are memory restrictions as well.  When training deep networks in kernel, memory errors happen. When I trained even Resnet152 I had to decrease batch size because of memory errors. Kernel just dies in such cases.</p>",
          "rawMarkdown": "Sure, there are memory restrictions as well.  When training deep networks in kernel, memory errors happen. When I trained even Resnet152 I had to decrease batch size because of memory errors. Kernel just dies in such cases."
        },
        {
          "id": 543167,
          "postDate": "2019-06-04T11:07:04.273Z",
          "content": "<p>A terrible thing is that in stage 1, the restrictions are fine, but error happens in stage 2 : (</p>",
          "rawMarkdown": "A terrible thing is that in stage 1, the restrictions are fine, but error happens in stage 2 : (",
          "votes": 1
        }
      ]
    },
    {
      "id": 542935,
      "postDate": "2019-06-04T08:03:04.993Z",
      "rawMarkdown": "",
      "votes": 1,
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 542893,
      "author_name": "Yiheng Wang",
      "author_url": "",
      "post_date": "2019-06-04T07:06:06.780000",
      "content": "<p><a href=\"https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/discussion/87719#latest-536719\">https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/discussion/87719#latest-536719</a>\nI just found this post, does it mean we can use as many trained models as possible?</p>",
      "votes": 0,
      "replies": [
        {
          "id": 542957,
          "author_name": "Appian",
          "author_url": "",
          "post_date": "2019-06-04T08:27:04.960000",
          "content": "<p>I think so. You can build a huge ensemble as long as time limit allows.</p>",
          "votes": 1,
          "replies": []
        },
        {
          "id": 543163,
          "author_name": "Artyom Palvelev",
          "author_url": "",
          "post_date": "2019-06-04T11:04:37.870000",
          "content": "<p>What made you think the number of trained models might be limited?</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 542870,
      "author_name": "Yiheng Wang",
      "author_url": "",
      "post_date": "2019-06-04T06:45:09.493000",
      "content": "<p>What if we meet memory error? </p>",
      "votes": 0,
      "replies": [
        {
          "id": 543142,
          "author_name": "Anna Novikova",
          "author_url": "",
          "post_date": "2019-06-04T10:51:35.607000",
          "content": "<p>Sure, there are memory restrictions as well.  When training deep networks in kernel, memory errors happen. When I trained even Resnet152 I had to decrease batch size because of memory errors. Kernel just dies in such cases.</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 543167,
          "author_name": "Yiheng Wang",
          "author_url": "",
          "post_date": "2019-06-04T11:07:04.273000",
          "content": "<p>A terrible thing is that in stage 1, the restrictions are fine, but error happens in stage 2 : (</p>",
          "votes": 1,
          "replies": []
        }
      ]
    },
    {
      "id": 542935,
      "author_name": "",
      "author_url": "",
      "post_date": "2019-06-04T08:03:04.993000",
      "content": "",
      "votes": 1,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "542868": "Like storages, CPU time? I cannot find other rules expect the 9 hours' GPU running time.",
    "542893": "https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/discussion/87719#latest-536719\nI just found this post, does it mean we can use as many trained models as possible?",
    "542870": "What if we meet memory error? ",
    "542935": ""
  }
}