{
  "id": 23971,
  "title": "Rule Clarification",
  "url": "/competitions/bosch-production-line-performance/discussion/23971",
  "author_name": "",
  "post_date": "2016-09-29T10:32:22.440Z",
  "votes": 3,
  "comment_count": 7,
  "views": 1015,
  "content": "<p>When I see ids 1,2,4,6 in train and <strong>3</strong>,7,8,9 in test for example I get a little worried with respect to data leaks.</p>\n\n<p>Can someone clarify that a model using id directly or via interpolation is invalid in this competition.</p>\n\n<p>Thanks in advance</p>",
  "messages": [
    {
      "id": "137219",
      "postDate": "09/29/2016 10:32:22",
      "content": "<p>When I see ids 1,2,4,6 in train and <strong>3</strong>,7,8,9 in test for example I get a little worried with respect to data leaks.</p>\n\n<p>Can someone clarify that a model using id directly or via interpolation is invalid in this competition.</p>\n\n<p>Thanks in advance</p>",
      "rawMarkdown": "When I see ids 1,2,4,6 in train and **3**,7,8,9 in test for example I get a little worried with respect to data leaks.\r\n\r\nCan someone clarify that a model using id directly or via interpolation is invalid in this competition.\r\n\r\nThanks in advance",
      "votes": null
    },
    {
      "id": "137220",
      "postDate": "09/29/2016 10:37:33",
      "content": "<p>It's invalid. Clever ID interpolation can get around 0.53-0.54 AUC in my tests. However, the dataset is designed so that the train and test sets are randomly sampled from the same time period, so I don't think this would be a leak even if it did exist.</p>",
      "rawMarkdown": "It's invalid. Clever ID interpolation can get around 0.53-0.54 AUC in my tests. However, the dataset is designed so that the train and test sets are randomly sampled from the same time period, so I don't think this would be a leak even if it did exist.",
      "votes": null
    },
    {
      "id": "137221",
      "postDate": "09/29/2016 10:39:40",
      "content": "<p>Thanks Anokas I shall use [1:-1] in my arrays then!</p>",
      "rawMarkdown": "Thanks Anokas I shall use [1:-1] in my arrays then!",
      "votes": null
    },
    {
      "id": "137222",
      "postDate": "09/29/2016 10:49:07",
      "content": "<p>A clarification for me: what do you mean by ID interpolation?</p>",
      "rawMarkdown": "A clarification for me: what do you mean by ID interpolation?",
      "votes": null
    },
    {
      "id": "137224",
      "postDate": "09/29/2016 11:15:14",
      "content": "<p>If train ids: 1,3 have response values of 1.  Which value would you guess for test with an id=2?</p>",
      "rawMarkdown": "If train ids: 1,3 have response values of 1.  Which value would you guess for test with an id=2?",
      "votes": null
    },
    {
      "id": "137229",
      "postDate": "09/29/2016 11:40:34",
      "content": "<p>OK, that figures. So when @anokas says this is 'invalid', I assume he means that it is not effective on this data set and not that it is against the rules?</p>",
      "rawMarkdown": "OK, that figures. So when @anokas says this is 'invalid', I assume he means that it is not effective on this data set and not that it is against the rules?",
      "votes": null
    },
    {
      "id": "137232",
      "postDate": "09/29/2016 12:32:44",
      "content": "<p>Anokas has done the investigation and has stated that it doesn't help.</p>\n\n<p>However I want it ruled out explicitly so that the models that win are useful - if you need to know the id then the models will be hopeless in terms of predictive power - but this is a personal point of view</p>",
      "rawMarkdown": "Anokas has done the investigation and has stated that it doesn't help.\r\n\r\nHowever I want it ruled out explicitly so that the models that win are useful - if you need to know the id then the models will be hopeless in terms of predictive power - but this is a personal point of view",
      "votes": null
    },
    {
      "id": "137236",
      "postDate": "09/29/2016 13:18:21",
      "content": "<p>[quote=joconnor;137229]</p>\n\n<p>OK, that figures. So when @anokas says this is 'invalid', I assume he means that it is not effective on this data set and not that it is against the rules?</p>\n\n<p>[/quote]</p>\n\n<p>Yeah, that's what I think. As far as I can see it's not effective, however there is no reason why it would be against the rules.</p>",
      "rawMarkdown": "[quote=joconnor;137229]\r\n\r\nOK, that figures. So when @anokas says this is 'invalid', I assume he means that it is not effective on this data set and not that it is against the rules?\r\n\r\n[/quote]\r\n\r\nYeah, that's what I think. As far as I can see it's not effective, however there is no reason why it would be against the rules.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 137220,
      "author_name": "anokas",
      "author_url": "",
      "post_date": "09/29/2016 10:37:33",
      "content": "<p>It's invalid. Clever ID interpolation can get around 0.53-0.54 AUC in my tests. However, the dataset is designed so that the train and test sets are randomly sampled from the same time period, so I don't think this would be a leak even if it did exist.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 137221,
      "author_name": "scirpus",
      "author_url": "",
      "post_date": "09/29/2016 10:39:40",
      "content": "<p>Thanks Anokas I shall use [1:-1] in my arrays then!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 137222,
      "author_name": "joconnor",
      "author_url": "",
      "post_date": "09/29/2016 10:49:07",
      "content": "<p>A clarification for me: what do you mean by ID interpolation?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 137224,
      "author_name": "scirpus",
      "author_url": "",
      "post_date": "09/29/2016 11:15:14",
      "content": "<p>If train ids: 1,3 have response values of 1.  Which value would you guess for test with an id=2?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 137229,
      "author_name": "joconnor",
      "author_url": "",
      "post_date": "09/29/2016 11:40:34",
      "content": "<p>OK, that figures. So when @anokas says this is 'invalid', I assume he means that it is not effective on this data set and not that it is against the rules?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 137232,
      "author_name": "scirpus",
      "author_url": "",
      "post_date": "09/29/2016 12:32:44",
      "content": "<p>Anokas has done the investigation and has stated that it doesn't help.</p>\n\n<p>However I want it ruled out explicitly so that the models that win are useful - if you need to know the id then the models will be hopeless in terms of predictive power - but this is a personal point of view</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 137236,
      "author_name": "anokas",
      "author_url": "",
      "post_date": "09/29/2016 13:18:21",
      "content": "<p>[quote=joconnor;137229]</p>\n\n<p>OK, that figures. So when @anokas says this is 'invalid', I assume he means that it is not effective on this data set and not that it is against the rules?</p>\n\n<p>[/quote]</p>\n\n<p>Yeah, that's what I think. As far as I can see it's not effective, however there is no reason why it would be against the rules.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "137219": "When I see ids 1,2,4,6 in train and **3**,7,8,9 in test for example I get a little worried with respect to data leaks.\r\n\r\nCan someone clarify that a model using id directly or via interpolation is invalid in this competition.\r\n\r\nThanks in advance",
    "137220": "It's invalid. Clever ID interpolation can get around 0.53-0.54 AUC in my tests. However, the dataset is designed so that the train and test sets are randomly sampled from the same time period, so I don't think this would be a leak even if it did exist.",
    "137221": "Thanks Anokas I shall use [1:-1] in my arrays then!",
    "137222": "A clarification for me: what do you mean by ID interpolation?",
    "137224": "If train ids: 1,3 have response values of 1.  Which value would you guess for test with an id=2?",
    "137229": "OK, that figures. So when @anokas says this is 'invalid', I assume he means that it is not effective on this data set and not that it is against the rules?",
    "137232": "Anokas has done the investigation and has stated that it doesn't help.\r\n\r\nHowever I want it ruled out explicitly so that the models that win are useful - if you need to know the id then the models will be hopeless in terms of predictive power - but this is a personal point of view",
    "137236": "[quote=joconnor;137229]\r\n\r\nOK, that figures. So when @anokas says this is 'invalid', I assume he means that it is not effective on this data set and not that it is against the rules?\r\n\r\n[/quote]\r\n\r\nYeah, that's what I think. As far as I can see it's not effective, however there is no reason why it would be against the rules."
  },
  "source": "meta"
}