{
  "id": 2407,
  "title": "Updated Training and Test Data",
  "url": "/competitions/MerckActivity/discussion/2407",
  "author_name": "",
  "post_date": "2012-08-22T01:25:53.267Z",
  "votes": 1,
  "comment_count": 6,
  "views": 4222,
  "content": "<p><span>New data files are now&nbsp;</span><a href=\"https://www.kaggle.com/c/MerckActivity/data\" rel=\"nofollow\" target=\"_blank\">available</a><span>. Molecule names and descriptors names have been simplified to no longer use special characters and thus avoid file\r\n reading problems. Also, files should contain only unique molecules and unique descriptors within each data set. As described previously, some molecules and descriptors will appear in multiple data sets.</span></p>\r\n<p>Please download the new data to continue participating in the competition. As the molecule names have changed, previously generated submission files will no longer work.</p>\r\n<p>We apologize for the inconvenience, however the data should be easier to work with now. Good luck to everyone!</p>",
  "messages": [
    {
      "id": "13301",
      "postDate": "08/22/2012 01:25:53",
      "content": "<p><span>New data files are now&nbsp;</span><a href=\"https://www.kaggle.com/c/MerckActivity/data\" rel=\"nofollow\" target=\"_blank\">available</a><span>. Molecule names and descriptors names have been simplified to no longer use special characters and thus avoid file\r\n reading problems. Also, files should contain only unique molecules and unique descriptors within each data set. As described previously, some molecules and descriptors will appear in multiple data sets.</span></p>\r\n<p>Please download the new data to continue participating in the competition. As the molecule names have changed, previously generated submission files will no longer work.</p>\r\n<p>We apologize for the inconvenience, however the data should be easier to work with now. Good luck to everyone!</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "13304",
      "postDate": "08/22/2012 02:08:27",
      "content": "<p>Will the public leaderboard be reset?</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "13309",
      "postDate": "08/22/2012 03:44:16",
      "content": "<p>No, unfortunately not. Previously entered submissions need to be resbumitted. Since the molecule IDs have changed, rescoring old submissions will fail since the identifiers have changed.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "13346",
      "postDate": "08/23/2012 01:45:55",
      "content": "<p>Even after the updates, I still find the data weird. There is a huge gap between my cv score and the public leaderboard score. The distributions of my predictions for the test sets look also suspect (particularly for Act 2, 5, 8 and 15).<br>\r\nHave you done any test yourself? My code is pretty simple with no risk of overfitting. Are you sure the test data are correct? Are the test samples generated randomly?</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "13352",
      "postDate": "08/23/2012 11:57:12",
      "content": "<p>[quote=Gxav (Xavier Conort);13346]</p>\r\n<p>Even after the updates, I still find the data weird. There is a huge gap between my cv score and the public leaderboard score. The distributions of my predictions for the test sets look also suspect (particularly for Act 2, 5, 8 and 15).<br>\r\nHave you done any test yourself? My code is pretty simple with no risk of overfitting. Are you sure the test data are correct? Are the test samples generated randomly?</p>\r\n<p>[/quote]</p>\r\n<p>Yup, I agree with Gxav. My CV scores doesn't match up with the leaderboard. I don't know why yet but I am of the same opinion as Gxav. May be worth checking the test data again?\r\n</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "13353",
      "postDate": "08/23/2012 12:27:39",
      "content": "<p>[quote=Gxav (Xavier Conort);13346]</p>\r\n<p>Even after the updates, I still find the data weird. There is a huge gap between my cv score and the public leaderboard score. The distributions of my predictions for the test sets look also suspect (particularly for Act 2, 5, 8 and 15).<br>\r\nHave you done any test yourself? My code is pretty simple with no risk of overfitting. Are you sure the test data are correct? Are the test samples generated randomly?</p>\r\n<p>[/quote]</p>\r\n<p>&nbsp;</p>\r\n<p>My CV scores are near sqrt(leaderboard). Could the problem be the metric implementation?</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "14289",
      "postDate": "09/13/2012 01:15:51",
      "content": "<p>[quote=Blind Ape;13353]</p>\r\n<p>[quote=Gxav (Xavier Conort);13346]</p>\r\n<p>Even after the updates, I still find the data weird. There is a huge gap between my cv score and the public leaderboard score. The distributions of my predictions for the test sets look also suspect (particularly for Act 2, 5, 8 and 15).<br>\r\nHave you done any test yourself? My code is pretty simple with no risk of overfitting. Are you sure the test data are correct? Are the test samples generated randomly?</p>\r\n<p>[/quote]</p>\r\n<p>&nbsp;</p>\r\n<p>My CV scores are near sqrt(leaderboard). Could the problem be the metric implementation?</p>\r\n<p>[/quote]</p>\r\n<p>&nbsp;</p>\r\n<p>My cv scores are similar to Gxav's.</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 13304,
      "author_name": "julianvillaruz0",
      "author_url": "",
      "post_date": "08/22/2012 02:08:27",
      "content": "<p>Will the public leaderboard be reset?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 13309,
      "author_name": "joycenv",
      "author_url": "",
      "post_date": "08/22/2012 03:44:16",
      "content": "<p>No, unfortunately not. Previously entered submissions need to be resbumitted. Since the molecule IDs have changed, rescoring old submissions will fail since the identifiers have changed.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 13346,
      "author_name": "xavierconort",
      "author_url": "",
      "post_date": "08/23/2012 01:45:55",
      "content": "<p>Even after the updates, I still find the data weird. There is a huge gap between my cv score and the public leaderboard score. The distributions of my predictions for the test sets look also suspect (particularly for Act 2, 5, 8 and 15).<br>\r\nHave you done any test yourself? My code is pretty simple with no risk of overfitting. Are you sure the test data are correct? Are the test samples generated randomly?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 13352,
      "author_name": "ejlok1",
      "author_url": "",
      "post_date": "08/23/2012 11:57:12",
      "content": "<p>[quote=Gxav (Xavier Conort);13346]</p>\r\n<p>Even after the updates, I still find the data weird. There is a huge gap between my cv score and the public leaderboard score. The distributions of my predictions for the test sets look also suspect (particularly for Act 2, 5, 8 and 15).<br>\r\nHave you done any test yourself? My code is pretty simple with no risk of overfitting. Are you sure the test data are correct? Are the test samples generated randomly?</p>\r\n<p>[/quote]</p>\r\n<p>Yup, I agree with Gxav. My CV scores doesn't match up with the leaderboard. I don't know why yet but I am of the same opinion as Gxav. May be worth checking the test data again?\r\n</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 13353,
      "author_name": "blindape",
      "author_url": "",
      "post_date": "08/23/2012 12:27:39",
      "content": "<p>[quote=Gxav (Xavier Conort);13346]</p>\r\n<p>Even after the updates, I still find the data weird. There is a huge gap between my cv score and the public leaderboard score. The distributions of my predictions for the test sets look also suspect (particularly for Act 2, 5, 8 and 15).<br>\r\nHave you done any test yourself? My code is pretty simple with no risk of overfitting. Are you sure the test data are correct? Are the test samples generated randomly?</p>\r\n<p>[/quote]</p>\r\n<p>&nbsp;</p>\r\n<p>My CV scores are near sqrt(leaderboard). Could the problem be the metric implementation?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 14289,
      "author_name": "fever53443",
      "author_url": "",
      "post_date": "09/13/2012 01:15:51",
      "content": "<p>[quote=Blind Ape;13353]</p>\r\n<p>[quote=Gxav (Xavier Conort);13346]</p>\r\n<p>Even after the updates, I still find the data weird. There is a huge gap between my cv score and the public leaderboard score. The distributions of my predictions for the test sets look also suspect (particularly for Act 2, 5, 8 and 15).<br>\r\nHave you done any test yourself? My code is pretty simple with no risk of overfitting. Are you sure the test data are correct? Are the test samples generated randomly?</p>\r\n<p>[/quote]</p>\r\n<p>&nbsp;</p>\r\n<p>My CV scores are near sqrt(leaderboard). Could the problem be the metric implementation?</p>\r\n<p>[/quote]</p>\r\n<p>&nbsp;</p>\r\n<p>My cv scores are similar to Gxav's.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "13301": "",
    "13304": "",
    "13309": "",
    "13346": "",
    "13352": "",
    "13353": "",
    "14289": ""
  },
  "source": "meta"
}