{
  "id": 19950,
  "title": "external data?",
  "url": "/competitions/state-farm-distracted-driver-detection/discussion/19950",
  "author_name": "",
  "post_date": "2016-04-05T18:07:55.953Z",
  "votes": 6,
  "comment_count": 7,
  "views": 2296,
  "content": "<p>Hi,</p>\n\n<p>Is external data like pretrained nets allowed? ;)</p>",
  "messages": [
    {
      "id": "113857",
      "postDate": "04/05/2016 18:07:55",
      "content": "<p>Hi,</p>\n\n<p>Is external data like pretrained nets allowed? ;)</p>",
      "rawMarkdown": "Hi,\r\n\r\nIs external data like pretrained nets allowed? ;)",
      "votes": null
    },
    {
      "id": "113859",
      "postDate": "04/05/2016 18:12:52",
      "content": "<p>No. </p>",
      "rawMarkdown": "No.",
      "votes": null
    },
    {
      "id": "113876",
      "postDate": "04/05/2016 21:39:12",
      "content": "<p>why not? all the approaches that might work for this competition generally use pre-trained networks. Read the literature.    </p>",
      "rawMarkdown": "why not? all the approaches that might work for this competition generally use pre-trained networks. Read the literature.",
      "votes": null
    },
    {
      "id": "113896",
      "postDate": "04/06/2016 01:25:34",
      "content": "<p>@DavidGbodiOdaibo,</p>\n\n<p>In <a href=\"https://www.kaggle.com/c/yelp-restaurant-photo-classification\">some competitions</a> we allow pre-trained nets, so you're welcome to use your pre-trained nets in those competitions. In <a href=\"https://www.kaggle.com/c/second-annual-data-science-bowl/forums/t/17899/pre-training-on-non-mri-medical-data\">some others,</a> external data (including pre-trained models) are not allowed. These decisions are made based on the nature of the dataset. </p>",
      "rawMarkdown": "DavidGbodiOdaibo,\r\n\r\nIn [some competitions][1] we allow pre-trained nets, so you're welcome to use your pre-trained nets in those competitions. In [some others,][2] external data (including pre-trained models) are not allowed. These decisions are made based on the nature of the dataset. \r\n\r\n\r\n  [1]: https://www.kaggle.com/c/yelp-restaurant-photo-classification\r\n  [2]: https://www.kaggle.com/c/second-annual-data-science-bowl/forums/t/17899/pre-training-on-non-mri-medical-data",
      "votes": null
    },
    {
      "id": "114020",
      "postDate": "04/06/2016 23:38:43",
      "content": "<p>@Wendy Kan: It seems unfair in some sense, because lets say someone who wants to use a Transfer Learning based approach (and has access to the whole ImageNet dataset, and a decent-ish GPU cluster) can still train a standard architecture for a week or more to come up with their own version of pre trained net as  the <em>first step</em> of their solution !! </p>",
      "rawMarkdown": "Wendy Kan: It seems unfair in some sense, because lets say someone who wants to use a Transfer Learning based approach (and has access to the whole ImageNet dataset, and a decent-ish GPU cluster) can still train a standard architecture for a week or more to come up with their own version of pre trained net as  the *first step* of their solution !!",
      "votes": null
    },
    {
      "id": "114027",
      "postDate": "04/07/2016 03:40:37",
      "content": "<p>[quote=SP Mohanty;114020]</p>\n\n<p>@Wendy Kan: It seems unfair in some sense, because lets say someone who wants to use a Transfer Learning based approach (and has access to the whole ImageNet dataset, and a decent-ish GPU cluster) can still train a standard architecture for a week or more to come up with their own version of pre trained net as  the <em>first step</em> of their solution !! </p>\n\n<p>[/quote]</p>\n\n<p>Technically, what you're describing would still be a violation of rules (use of external dataset: ImageNet).  Whether copied or homebrew, any pre-trained net based on external data would be a no-no.</p>\n\n<p>I think the sponsors/Kaggle want this one to be from scratch, using the given dataset only.  </p>",
      "rawMarkdown": "[quote=SP Mohanty;114020]\r\n\r\n@Wendy Kan: It seems unfair in some sense, because lets say someone who wants to use a Transfer Learning based approach (and has access to the whole ImageNet dataset, and a decent-ish GPU cluster) can still train a standard architecture for a week or more to come up with their own version of pre trained net as  the *first step* of their solution !! \r\n\r\n[/quote]\r\n\r\nTechnically, what you're describing would still be a violation of rules (use of external dataset: ImageNet).  Whether copied or homebrew, any pre-trained net based on external data would be a no-no.\r\n\r\nI think the sponsors/Kaggle want this one to be from scratch, using the given dataset only.",
      "votes": null
    },
    {
      "id": "117986",
      "postDate": "05/02/2016 09:45:05",
      "content": "<p>Hi , can I use MATLAB for this assignment?</p>",
      "rawMarkdown": "Hi , can I use MATLAB for this assignment?",
      "votes": null
    },
    {
      "id": "118006",
      "postDate": "05/02/2016 13:25:35",
      "content": "<p>[quote=Wendy Kan;113859]</p>\n\n<p>No. </p>\n\n<p>[/quote]</p>\n\n<p>Hi Wendy, does this thread <a href=\"https://www.kaggle.com/c/state-farm-distracted-driver-detection/forums/t/20141/official-pre-trained-models-and-external-data-thread\">https://www.kaggle.com/c/state-farm-distracted-driver-detection/forums/t/20141/official-pre-trained-models-and-external-data-thread</a> mean that the answer is now &quot;Yes, but you should refer to this external data in the form&quot;?</p>\n\n<p>Ah, I see, the rules are updated to allow external data. Thanks.</p>",
      "rawMarkdown": "[quote=Wendy Kan;113859]\r\n\r\nNo. \r\n\r\n[/quote]\r\n\r\nHi Wendy, does this thread https://www.kaggle.com/c/state-farm-distracted-driver-detection/forums/t/20141/official-pre-trained-models-and-external-data-thread mean that the answer is now \"Yes, but you should refer to this external data in the form\"?\r\n\r\nAh, I see, the rules are updated to allow external data. Thanks.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 113859,
      "author_name": "wendykan",
      "author_url": "",
      "post_date": "04/05/2016 18:12:52",
      "content": "<p>No. </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 113876,
      "author_name": "godaibo",
      "author_url": "",
      "post_date": "04/05/2016 21:39:12",
      "content": "<p>why not? all the approaches that might work for this competition generally use pre-trained networks. Read the literature.    </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 113896,
      "author_name": "wendykan",
      "author_url": "",
      "post_date": "04/06/2016 01:25:34",
      "content": "<p>@DavidGbodiOdaibo,</p>\n\n<p>In <a href=\"https://www.kaggle.com/c/yelp-restaurant-photo-classification\">some competitions</a> we allow pre-trained nets, so you're welcome to use your pre-trained nets in those competitions. In <a href=\"https://www.kaggle.com/c/second-annual-data-science-bowl/forums/t/17899/pre-training-on-non-mri-medical-data\">some others,</a> external data (including pre-trained models) are not allowed. These decisions are made based on the nature of the dataset. </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 114020,
      "author_name": "spmohanty",
      "author_url": "",
      "post_date": "04/06/2016 23:38:43",
      "content": "<p>@Wendy Kan: It seems unfair in some sense, because lets say someone who wants to use a Transfer Learning based approach (and has access to the whole ImageNet dataset, and a decent-ish GPU cluster) can still train a standard architecture for a week or more to come up with their own version of pre trained net as  the <em>first step</em> of their solution !! </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 114027,
      "author_name": "taffeylewis",
      "author_url": "",
      "post_date": "04/07/2016 03:40:37",
      "content": "<p>[quote=SP Mohanty;114020]</p>\n\n<p>@Wendy Kan: It seems unfair in some sense, because lets say someone who wants to use a Transfer Learning based approach (and has access to the whole ImageNet dataset, and a decent-ish GPU cluster) can still train a standard architecture for a week or more to come up with their own version of pre trained net as  the <em>first step</em> of their solution !! </p>\n\n<p>[/quote]</p>\n\n<p>Technically, what you're describing would still be a violation of rules (use of external dataset: ImageNet).  Whether copied or homebrew, any pre-trained net based on external data would be a no-no.</p>\n\n<p>I think the sponsors/Kaggle want this one to be from scratch, using the given dataset only.  </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 117986,
      "author_name": "chetanksreddy",
      "author_url": "",
      "post_date": "05/02/2016 09:45:05",
      "content": "<p>Hi , can I use MATLAB for this assignment?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 118006,
      "author_name": "milakov",
      "author_url": "",
      "post_date": "05/02/2016 13:25:35",
      "content": "<p>[quote=Wendy Kan;113859]</p>\n\n<p>No. </p>\n\n<p>[/quote]</p>\n\n<p>Hi Wendy, does this thread <a href=\"https://www.kaggle.com/c/state-farm-distracted-driver-detection/forums/t/20141/official-pre-trained-models-and-external-data-thread\">https://www.kaggle.com/c/state-farm-distracted-driver-detection/forums/t/20141/official-pre-trained-models-and-external-data-thread</a> mean that the answer is now &quot;Yes, but you should refer to this external data in the form&quot;?</p>\n\n<p>Ah, I see, the rules are updated to allow external data. Thanks.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "113857": "Hi,\r\n\r\nIs external data like pretrained nets allowed? ;)",
    "113859": "No.",
    "113876": "why not? all the approaches that might work for this competition generally use pre-trained networks. Read the literature.",
    "113896": "DavidGbodiOdaibo,\r\n\r\nIn [some competitions][1] we allow pre-trained nets, so you're welcome to use your pre-trained nets in those competitions. In [some others,][2] external data (including pre-trained models) are not allowed. These decisions are made based on the nature of the dataset. \r\n\r\n\r\n  [1]: https://www.kaggle.com/c/yelp-restaurant-photo-classification\r\n  [2]: https://www.kaggle.com/c/second-annual-data-science-bowl/forums/t/17899/pre-training-on-non-mri-medical-data",
    "114020": "Wendy Kan: It seems unfair in some sense, because lets say someone who wants to use a Transfer Learning based approach (and has access to the whole ImageNet dataset, and a decent-ish GPU cluster) can still train a standard architecture for a week or more to come up with their own version of pre trained net as  the *first step* of their solution !!",
    "114027": "[quote=SP Mohanty;114020]\r\n\r\n@Wendy Kan: It seems unfair in some sense, because lets say someone who wants to use a Transfer Learning based approach (and has access to the whole ImageNet dataset, and a decent-ish GPU cluster) can still train a standard architecture for a week or more to come up with their own version of pre trained net as  the *first step* of their solution !! \r\n\r\n[/quote]\r\n\r\nTechnically, what you're describing would still be a violation of rules (use of external dataset: ImageNet).  Whether copied or homebrew, any pre-trained net based on external data would be a no-no.\r\n\r\nI think the sponsors/Kaggle want this one to be from scratch, using the given dataset only.",
    "117986": "Hi , can I use MATLAB for this assignment?",
    "118006": "[quote=Wendy Kan;113859]\r\n\r\nNo. \r\n\r\n[/quote]\r\n\r\nHi Wendy, does this thread https://www.kaggle.com/c/state-farm-distracted-driver-detection/forums/t/20141/official-pre-trained-models-and-external-data-thread mean that the answer is now \"Yes, but you should refer to this external data in the form\"?\r\n\r\nAh, I see, the rules are updated to allow external data. Thanks."
  },
  "source": "meta"
}