{
  "id": 80328,
  "title": "Any use of merchants?",
  "url": "/competitions/elo-merchant-category-recommendation/discussion/80328",
  "author_name": "",
  "post_date": "2019-02-12T17:36:52.452329500Z",
  "votes": 4,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Have you got any gain from merchants? </p>\n\n<p>I've tried four different approaches and they all don't affect target rmse: \n- bag of words and tf-idf for all merchants the customer interacted with\n- merged all categorical features from merchants.csv to history and then bow/tf-idf for filtering important correlations\n- aggregated statistics from merchants.csv numerical features\n- purchase_amount ratio month to month for top 50-100 merchants</p>\n\n<p>I've tried boosting, Ridge, LR and nothing. Ideas? Give a hint if you've found something  </p>",
  "messages": [
    {
      "id": "470285",
      "postDate": "02/12/2019 17:36:52",
      "content": "<p>Have you got any gain from merchants? </p>\n\n<p>I've tried four different approaches and they all don't affect target rmse: \n- bag of words and tf-idf for all merchants the customer interacted with\n- merged all categorical features from merchants.csv to history and then bow/tf-idf for filtering important correlations\n- aggregated statistics from merchants.csv numerical features\n- purchase_amount ratio month to month for top 50-100 merchants</p>\n\n<p>I've tried boosting, Ridge, LR and nothing. Ideas? Give a hint if you've found something  </p>",
      "rawMarkdown": "Have you got any gain from merchants? \n\nI've tried four different approaches and they all don't affect target rmse: \n- bag of words and tf-idf for all merchants the customer interacted with\n- merged all categorical features from merchants.csv to history and then bow/tf-idf for filtering important correlations\n- aggregated statistics from merchants.csv numerical features\n- purchase_amount ratio month to month for top 50-100 merchants\n\nI've tried boosting, Ridge, LR and nothing. Ideas? Give a hint if you've found something",
      "votes": null
    },
    {
      "id": "470326",
      "postDate": "02/12/2019 19:03:55",
      "content": "<p>there is a thread\n<a href=\"https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/77987\">https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/77987</a></p>",
      "rawMarkdown": "there is a thread\nhttps://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/77987",
      "votes": null
    },
    {
      "id": "470339",
      "postDate": "02/12/2019 19:35:38",
      "content": "<p>merchants.csv has been useless so far, in any way</p>",
      "rawMarkdown": "merchants.csv has been useless so far, in any way",
      "votes": null
    },
    {
      "id": "470356",
      "postDate": "02/12/2019 20:04:20",
      "content": "<p><a href=\"/raddar\">@raddar</a>\nI found many places where categorical features like subsector_id, are different between merchants.csv and historical.csv.\nwhich is weird, haven't yet tried it, but maybe aggregated features by one or the other give better results.</p>",
      "rawMarkdown": "raddar\nI found many places where categorical features like subsector_id, are different between merchants.csv and historical.csv.\nwhich is weird, haven't yet tried it, but maybe aggregated features by one or the other give better results.",
      "votes": null
    },
    {
      "id": "470358",
      "postDate": "02/12/2019 20:12:38",
      "content": "<p><a href=\"/raddar\">@raddar</a> yes, and that is odd. I thought there are some merchants in \"loyalty program\" and some not included or they have some graduation but so far nothing - we can only treat them as equal</p>",
      "rawMarkdown": "raddar yes, and that is odd. I thought there are some merchants in \"loyalty program\" and some not included or they have some graduation but so far nothing - we can only treat them as equal",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 470326,
      "author_name": "mks2192",
      "author_url": "",
      "post_date": "02/12/2019 19:03:55",
      "content": "<p>there is a thread\n<a href=\"https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/77987\">https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/77987</a></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 470339,
      "author_name": "raddar",
      "author_url": "",
      "post_date": "02/12/2019 19:35:38",
      "content": "<p>merchants.csv has been useless so far, in any way</p>",
      "votes": null,
      "replies": [
        {
          "id": 470356,
          "author_name": "tpthegreat",
          "author_url": "",
          "post_date": "02/12/2019 20:04:20",
          "content": "<p><a href=\"/raddar\">@raddar</a>\nI found many places where categorical features like subsector_id, are different between merchants.csv and historical.csv.\nwhich is weird, haven't yet tried it, but maybe aggregated features by one or the other give better results.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 470358,
          "author_name": "yaroshevskiy",
          "author_url": "",
          "post_date": "02/12/2019 20:12:38",
          "content": "<p><a href=\"/raddar\">@raddar</a> yes, and that is odd. I thought there are some merchants in \"loyalty program\" and some not included or they have some graduation but so far nothing - we can only treat them as equal</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "470285": "Have you got any gain from merchants? \n\nI've tried four different approaches and they all don't affect target rmse: \n- bag of words and tf-idf for all merchants the customer interacted with\n- merged all categorical features from merchants.csv to history and then bow/tf-idf for filtering important correlations\n- aggregated statistics from merchants.csv numerical features\n- purchase_amount ratio month to month for top 50-100 merchants\n\nI've tried boosting, Ridge, LR and nothing. Ideas? Give a hint if you've found something",
    "470326": "there is a thread\nhttps://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/77987",
    "470339": "merchants.csv has been useless so far, in any way",
    "470356": "raddar\nI found many places where categorical features like subsector_id, are different between merchants.csv and historical.csv.\nwhich is weird, haven't yet tried it, but maybe aggregated features by one or the other give better results.",
    "470358": "raddar yes, and that is odd. I thought there are some merchants in \"loyalty program\" and some not included or they have some graduation but so far nothing - we can only treat them as equal"
  },
  "source": "meta"
}