{
  "id": 79323,
  "title": "Any deep learning models?",
  "url": "/competitions/elo-merchant-category-recommendation/discussion/79323",
  "author_name": "",
  "post_date": "2019-02-02T17:20:34.208801400Z",
  "votes": null,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Hello everyone,</p>\n\n<p>I hope you are enjoying this challenge so far. I was wondering if anyone of you have tried \na deep learning model and if so, do you get \"decent\" results? </p>\n\n<p>I know that deep learning models aren't best suited for this type of data (don't trust my word, check this reddit <a href=\"https://www.reddit.com/r/MachineLearning/comments/9826bt/d_why_is_deep_learning_so_bad_for_tabular_data/\">thread</a>). </p>\n\n<p>Nonetheless, I am asking this since I am starting to <a href=\"http://blog.kaggle.com/2016/12/27/a-kagglers-guide-to-model-stacking-in-practice/\">stack</a> some models and thought that a deep learning model might add something to the blend. Any suggestions on this topic are much appreciated. Thanks in advance. </p>",
  "messages": [
    {
      "id": "465242",
      "postDate": "02/02/2019 17:20:34",
      "content": "<p>Hello everyone,</p>\n\n<p>I hope you are enjoying this challenge so far. I was wondering if anyone of you have tried \na deep learning model and if so, do you get \"decent\" results? </p>\n\n<p>I know that deep learning models aren't best suited for this type of data (don't trust my word, check this reddit <a href=\"https://www.reddit.com/r/MachineLearning/comments/9826bt/d_why_is_deep_learning_so_bad_for_tabular_data/\">thread</a>). </p>\n\n<p>Nonetheless, I am asking this since I am starting to <a href=\"http://blog.kaggle.com/2016/12/27/a-kagglers-guide-to-model-stacking-in-practice/\">stack</a> some models and thought that a deep learning model might add something to the blend. Any suggestions on this topic are much appreciated. Thanks in advance. </p>",
      "rawMarkdown": "Hello everyone,\n\nI hope you are enjoying this challenge so far. I was wondering if anyone of you have tried \na deep learning model and if so, do you get \"decent\" results? \n\nI know that deep learning models aren't best suited for this type of data (don't trust my word, check this reddit [thread][1]). \n\nNonetheless, I am asking this since I am starting to [stack][2] some models and thought that a deep learning model might add something to the blend. Any suggestions on this topic are much appreciated. Thanks in advance. \n\n\n\n\n\n  [1]: https://www.reddit.com/r/MachineLearning/comments/9826bt/d_why_is_deep_learning_so_bad_for_tabular_data/\n  [2]: http://blog.kaggle.com/2016/12/27/a-kagglers-guide-to-model-stacking-in-practice/",
      "votes": null
    },
    {
      "id": "466310",
      "postDate": "02/05/2019 05:03:57",
      "content": "<p>Hi! I've been working since by 2 or 3 days on the idea that neural nets can be used to minimize the natural data noise in some way and improve the accuracy, got interesting results on that. </p>",
      "rawMarkdown": "Hi! I've been working since by 2 or 3 days on the idea that neural nets can be used to minimize the natural data noise in some way and improve the accuracy, got interesting results on that.",
      "votes": null
    },
    {
      "id": "466427",
      "postDate": "02/05/2019 11:09:00",
      "content": "<p>That's great. Do you mind sharing the type of the model? </p>",
      "rawMarkdown": "That's great. Do you mind sharing the type of the model?",
      "votes": null
    },
    {
      "id": "472598",
      "postDate": "02/16/2019 09:01:06",
      "content": "<p>i have used a NN with only train and test, got 3.952 vs. catboost 3.932. I'll try it in my ensamble</p>",
      "rawMarkdown": "i have used a NN with only train and test, got 3.952 vs. catboost 3.932. I'll try it in my ensamble",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 466310,
      "author_name": "camaradaancap",
      "author_url": "",
      "post_date": "02/05/2019 05:03:57",
      "content": "<p>Hi! I've been working since by 2 or 3 days on the idea that neural nets can be used to minimize the natural data noise in some way and improve the accuracy, got interesting results on that. </p>",
      "votes": null,
      "replies": [
        {
          "id": 466427,
          "author_name": "yassinealouini",
          "author_url": "",
          "post_date": "02/05/2019 11:09:00",
          "content": "<p>That's great. Do you mind sharing the type of the model? </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 472598,
      "author_name": "andreamar",
      "author_url": "",
      "post_date": "02/16/2019 09:01:06",
      "content": "<p>i have used a NN with only train and test, got 3.952 vs. catboost 3.932. I'll try it in my ensamble</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "465242": "Hello everyone,\n\nI hope you are enjoying this challenge so far. I was wondering if anyone of you have tried \na deep learning model and if so, do you get \"decent\" results? \n\nI know that deep learning models aren't best suited for this type of data (don't trust my word, check this reddit [thread][1]). \n\nNonetheless, I am asking this since I am starting to [stack][2] some models and thought that a deep learning model might add something to the blend. Any suggestions on this topic are much appreciated. Thanks in advance. \n\n\n\n\n\n  [1]: https://www.reddit.com/r/MachineLearning/comments/9826bt/d_why_is_deep_learning_so_bad_for_tabular_data/\n  [2]: http://blog.kaggle.com/2016/12/27/a-kagglers-guide-to-model-stacking-in-practice/",
    "466310": "Hi! I've been working since by 2 or 3 days on the idea that neural nets can be used to minimize the natural data noise in some way and improve the accuracy, got interesting results on that.",
    "466427": "That's great. Do you mind sharing the type of the model?",
    "472598": "i have used a NN with only train and test, got 3.952 vs. catboost 3.932. I'll try it in my ensamble"
  },
  "source": "meta"
}