{
  "id": 82035,
  "title": "Congrats to winners!",
  "url": "/competitions/elo-merchant-category-recommendation/discussion/82035",
  "author_name": "",
  "post_date": "2019-02-27T00:07:01.411104100Z",
  "votes": 16,
  "comment_count": 13,
  "views": 0,
  "content": "<p>I'm sure I will learn a lot from their writeups, as I have only scratched the surface of this competition.   </p>\n\n<p>With Giba we only had lgb models built in 5 days, and one NN ran during the last couple of hours.  This probably saved us from overfitting :)</p>",
  "messages": [
    {
      "id": "479035",
      "postDate": "02/27/2019 00:07:01",
      "content": "<p>I'm sure I will learn a lot from their writeups, as I have only scratched the surface of this competition.   </p>\n\n<p>With Giba we only had lgb models built in 5 days, and one NN ran during the last couple of hours.  This probably saved us from overfitting :)</p>",
      "rawMarkdown": "I'm sure I will learn a lot from their writeups, as I have only scratched the surface of this competition.   \n\nWith Giba we only had lgb models built in 5 days, and one NN ran during the last couple of hours.  This probably saved us from overfitting :)",
      "votes": null
    },
    {
      "id": "479053",
      "postDate": "02/27/2019 00:23:54",
      "content": "<p>Congrats. I wish you entered competition earlier, your comments  are really helpful to us.</p>",
      "rawMarkdown": "Congrats. I wish you entered competition earlier, your comments  are really helpful to us.",
      "votes": null
    },
    {
      "id": "479059",
      "postDate": "02/27/2019 00:28:31",
      "content": "<p>Thanks, Malware was eating all my time unfortunately.  I also wish I entered this one earlier!</p>",
      "rawMarkdown": "Thanks, Malware was eating all my time unfortunately.  I also wish I entered this one earlier!",
      "votes": null
    },
    {
      "id": "479100",
      "postDate": "02/27/2019 01:04:07",
      "content": "<p>Congratulations to the winners.</p>\n\n<p>Congrats <a href=\"/cpmpml\">@cpmpml</a>. Not bad for a week work on this interesting and challenging data. </p>\n\n<p>Most of the progress I made is in the last 3 days as well. Although I did not do any LB probing, it was all that feature selection techniques that saved moi. I jumped from public LB #311 to #239 in the last couple of days.</p>",
      "rawMarkdown": "Congratulations to the winners.\n\nCongrats @cpmpml. Not bad for a week work on this interesting and challenging data. \n\nMost of the progress I made is in the last 3 days as well. Although I did not do any LB probing, it was all that feature selection techniques that saved moi. I jumped from public LB #311 to #239 in the last couple of days.",
      "votes": null
    },
    {
      "id": "479101",
      "postDate": "02/27/2019 01:05:43",
      "content": "<p>Your result is quite good, congrats!</p>",
      "rawMarkdown": "Your result is quite good, congrats!",
      "votes": null
    },
    {
      "id": "479104",
      "postDate": "02/27/2019 01:07:25",
      "content": "<p>Less than a week actually, my first sub is from 4 days ago!</p>\n\n<p>This doe snot mean we would have got better result with more time, handling outliers the right way was key, and we have not found it.</p>",
      "rawMarkdown": "Less than a week actually, my first sub is from 4 days ago!\n\nThis doe snot mean we would have got better result with more time, handling outliers the right way was key, and we have not found it.",
      "votes": null
    },
    {
      "id": "479127",
      "postDate": "02/27/2019 01:27:34",
      "content": "<p>Congratulation CPMP !  Your topic about reverse engineering the purchase amount was really cool but I still don't know how to use this kind of information to improve my score. Thanks a lot for sharing this info.</p>",
      "rawMarkdown": "Congratulation CPMP !  Your topic about reverse engineering the purchase amount was really cool but I still don't know how to use this kind of information to improve my score. Thanks a lot for sharing this info.",
      "votes": null
    },
    {
      "id": "479176",
      "postDate": "02/27/2019 02:11:29",
      "content": "<p>Thanks @cpmpl. I did not find the right way to handle the outliers either.</p>\n\n<p>Here is my post on feature selection techniques <a href=\"https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/82052\">https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/82052</a></p>",
      "rawMarkdown": "Thanks @cpmpl. I did not find the right way to handle the outliers either.\n\nHere is my post on feature selection techniques https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/82052",
      "votes": null
    },
    {
      "id": "479282",
      "postDate": "02/27/2019 03:37:30",
      "content": "<p>Congrats, and could you share your solution? I am very curious about your features, models, etc..</p>",
      "rawMarkdown": "Congrats, and could you share your solution? I am very curious about your features, models, etc..",
      "votes": null
    },
    {
      "id": "479428",
      "postDate": "02/27/2019 06:03:38",
      "content": "<p><a href=\"/recruta42\">@recruta42</a>, just using <a href=\"/raddar\">@raddar</a>'s formula for purchase amount and changing the observation date improved my CV/LB significantly. I recreated all my features with those changes. Then I did more feature engineering to improve further.</p>",
      "rawMarkdown": "recruta42, just using @raddar's formula for purchase amount and changing the observation date improved my CV/LB significantly. I recreated all my features with those changes. Then I did more feature engineering to improve further.",
      "votes": null
    },
    {
      "id": "479458",
      "postDate": "02/27/2019 06:43:45",
      "content": "<p>Congratulations <a href=\"/cpmpml\">@cpmpml</a>, as a person full of sharing spirit, your comments can always bring great inspiration, and look forward to your continued activity on the stage of kaggle.</p>",
      "rawMarkdown": "Congratulations @cpmpml, as a person full of sharing spirit, your comments can always bring great inspiration, and look forward to your continued activity on the stage of kaggle.",
      "votes": null
    },
    {
      "id": "479481",
      "postDate": "02/27/2019 07:21:36",
      "content": "<p>Congratulations <a href=\"/cpmpml\">@cpmpml</a>. Did blending with ANN solved the overfitting?</p>",
      "rawMarkdown": "Congratulations @cpmpml. Did blending with ANN solved the overfitting?",
      "votes": null
    },
    {
      "id": "479598",
      "postDate": "02/27/2019 09:16:30",
      "content": "<p>Actually, the NN didn't make it to the final blend.</p>\n\n<p>WHat saved us from overfititng was that we did not had time to work on outlier detection!</p>",
      "rawMarkdown": "Actually, the NN didn't make it to the final blend.\n\nWHat saved us from overfititng was that we did not had time to work on outlier detection!",
      "votes": null
    },
    {
      "id": "479660",
      "postDate": "02/27/2019 10:20:28",
      "content": "<p>There isn't much to share, just proper use of lgb. Closest to what we did is probably one of the public kernels with lgb.  Things that are different, if any:\n - cleaned some data for isntance only keeping authorized ones, used rescaled purchase amount (see @Raddar kernel and mine), using CV to validate feature selection, etc.</p>",
      "rawMarkdown": "There isn't much to share, just proper use of lgb. Closest to what we did is probably one of the public kernels with lgb.  Things that are different, if any:\n - cleaned some data for isntance only keeping authorized ones, used rescaled purchase amount (see @Raddar kernel and mine), using CV to validate feature selection, etc.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 479053,
      "author_name": "fatall",
      "author_url": "",
      "post_date": "02/27/2019 00:23:54",
      "content": "<p>Congrats. I wish you entered competition earlier, your comments  are really helpful to us.</p>",
      "votes": null,
      "replies": [
        {
          "id": 479059,
          "author_name": "cpmpml",
          "author_url": "",
          "post_date": "02/27/2019 00:28:31",
          "content": "<p>Thanks, Malware was eating all my time unfortunately.  I also wish I entered this one earlier!</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 479282,
          "author_name": "liewmanchoi",
          "author_url": "",
          "post_date": "02/27/2019 03:37:30",
          "content": "<p>Congrats, and could you share your solution? I am very curious about your features, models, etc..</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 479660,
          "author_name": "cpmpml",
          "author_url": "",
          "post_date": "02/27/2019 10:20:28",
          "content": "<p>There isn't much to share, just proper use of lgb. Closest to what we did is probably one of the public kernels with lgb.  Things that are different, if any:\n - cleaned some data for isntance only keeping authorized ones, used rescaled purchase amount (see @Raddar kernel and mine), using CV to validate feature selection, etc.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 479100,
      "author_name": "sheriytm",
      "author_url": "",
      "post_date": "02/27/2019 01:04:07",
      "content": "<p>Congratulations to the winners.</p>\n\n<p>Congrats <a href=\"/cpmpml\">@cpmpml</a>. Not bad for a week work on this interesting and challenging data. </p>\n\n<p>Most of the progress I made is in the last 3 days as well. Although I did not do any LB probing, it was all that feature selection techniques that saved moi. I jumped from public LB #311 to #239 in the last couple of days.</p>",
      "votes": null,
      "replies": [
        {
          "id": 479101,
          "author_name": "cpmpml",
          "author_url": "",
          "post_date": "02/27/2019 01:05:43",
          "content": "<p>Your result is quite good, congrats!</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 479104,
          "author_name": "cpmpml",
          "author_url": "",
          "post_date": "02/27/2019 01:07:25",
          "content": "<p>Less than a week actually, my first sub is from 4 days ago!</p>\n\n<p>This doe snot mean we would have got better result with more time, handling outliers the right way was key, and we have not found it.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 479176,
          "author_name": "sheriytm",
          "author_url": "",
          "post_date": "02/27/2019 02:11:29",
          "content": "<p>Thanks @cpmpl. I did not find the right way to handle the outliers either.</p>\n\n<p>Here is my post on feature selection techniques <a href=\"https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/82052\">https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/82052</a></p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 479127,
      "author_name": "recruta42",
      "author_url": "",
      "post_date": "02/27/2019 01:27:34",
      "content": "<p>Congratulation CPMP !  Your topic about reverse engineering the purchase amount was really cool but I still don't know how to use this kind of information to improve my score. Thanks a lot for sharing this info.</p>",
      "votes": null,
      "replies": [
        {
          "id": 479428,
          "author_name": "sheriytm",
          "author_url": "",
          "post_date": "02/27/2019 06:03:38",
          "content": "<p><a href=\"/recruta42\">@recruta42</a>, just using <a href=\"/raddar\">@raddar</a>'s formula for purchase amount and changing the observation date improved my CV/LB significantly. I recreated all my features with those changes. Then I did more feature engineering to improve further.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 479458,
      "author_name": "liewmanchoi",
      "author_url": "",
      "post_date": "02/27/2019 06:43:45",
      "content": "<p>Congratulations <a href=\"/cpmpml\">@cpmpml</a>, as a person full of sharing spirit, your comments can always bring great inspiration, and look forward to your continued activity on the stage of kaggle.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 479481,
      "author_name": "karthik7395",
      "author_url": "",
      "post_date": "02/27/2019 07:21:36",
      "content": "<p>Congratulations <a href=\"/cpmpml\">@cpmpml</a>. Did blending with ANN solved the overfitting?</p>",
      "votes": null,
      "replies": [
        {
          "id": 479598,
          "author_name": "cpmpml",
          "author_url": "",
          "post_date": "02/27/2019 09:16:30",
          "content": "<p>Actually, the NN didn't make it to the final blend.</p>\n\n<p>WHat saved us from overfititng was that we did not had time to work on outlier detection!</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "479035": "I'm sure I will learn a lot from their writeups, as I have only scratched the surface of this competition.   \n\nWith Giba we only had lgb models built in 5 days, and one NN ran during the last couple of hours.  This probably saved us from overfitting :)",
    "479053": "Congrats. I wish you entered competition earlier, your comments  are really helpful to us.",
    "479059": "Thanks, Malware was eating all my time unfortunately.  I also wish I entered this one earlier!",
    "479100": "Congratulations to the winners.\n\nCongrats @cpmpml. Not bad for a week work on this interesting and challenging data. \n\nMost of the progress I made is in the last 3 days as well. Although I did not do any LB probing, it was all that feature selection techniques that saved moi. I jumped from public LB #311 to #239 in the last couple of days.",
    "479101": "Your result is quite good, congrats!",
    "479104": "Less than a week actually, my first sub is from 4 days ago!\n\nThis doe snot mean we would have got better result with more time, handling outliers the right way was key, and we have not found it.",
    "479127": "Congratulation CPMP !  Your topic about reverse engineering the purchase amount was really cool but I still don't know how to use this kind of information to improve my score. Thanks a lot for sharing this info.",
    "479176": "Thanks @cpmpl. I did not find the right way to handle the outliers either.\n\nHere is my post on feature selection techniques https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/82052",
    "479282": "Congrats, and could you share your solution? I am very curious about your features, models, etc..",
    "479428": "recruta42, just using @raddar's formula for purchase amount and changing the observation date improved my CV/LB significantly. I recreated all my features with those changes. Then I did more feature engineering to improve further.",
    "479458": "Congratulations @cpmpml, as a person full of sharing spirit, your comments can always bring great inspiration, and look forward to your continued activity on the stage of kaggle.",
    "479481": "Congratulations @cpmpml. Did blending with ANN solved the overfitting?",
    "479598": "Actually, the NN didn't make it to the final blend.\n\nWHat saved us from overfititng was that we did not had time to work on outlier detection!",
    "479660": "There isn't much to share, just proper use of lgb. Closest to what we did is probably one of the public kernels with lgb.  Things that are different, if any:\n - cleaned some data for isntance only keeping authorized ones, used rescaled purchase amount (see @Raddar kernel and mine), using CV to validate feature selection, etc."
  },
  "source": "meta"
}