{
  "id": 347610,
  "title": "last insights",
  "url": "/competitions/amex-default-prediction/discussion/347610",
  "author_name": "",
  "post_date": "2022-08-24T19:53:53.142671100Z",
  "votes": 2,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I joined the competitoin, to late, <br>\nand got about 0.79</p>\n<p>only Today, i came to a conclustion, that everybody, just took the last record for every customer,<br>\nand all the high scores, are based on that concept.</p>\n<p>I think, that if i take, all records, or ever 2-3 last ones, it will improve the score, but i just didnt have the time to do it.</p>\n<p>I will try it later on, and will be happy to try to submot, just for curiousity.</p>",
  "messages": [
    {
      "id": "1912543",
      "postDate": "08/24/2022 19:53:53",
      "content": "<p>I joined the competitoin, to late, <br>\nand got about 0.79</p>\n<p>only Today, i came to a conclustion, that everybody, just took the last record for every customer,<br>\nand all the high scores, are based on that concept.</p>\n<p>I think, that if i take, all records, or ever 2-3 last ones, it will improve the score, but i just didnt have the time to do it.</p>\n<p>I will try it later on, and will be happy to try to submot, just for curiousity.</p>",
      "rawMarkdown": "I joined the competitoin, to late, \nand got about 0.79\n\nonly Today, i came to a conclustion, that everybody, just took the last record for every customer,\nand all the high scores, are based on that concept.\n\nI think, that if i take, all records, or ever 2-3 last ones, it will improve the score, but i just didnt have the time to do it.\n\nI will try it later on, and will be happy to try to submot, just for curiousity.",
      "votes": null
    },
    {
      "id": "1912548",
      "postDate": "08/24/2022 19:56:21",
      "content": "<p>You can get a decent score with only the last entry, that is true. But I don't think you can even get a medal with the last entry alone…</p>",
      "rawMarkdown": "You can get a decent score with only the last entry, that is true. But I don't think you can even get a medal with the last entry alone...",
      "votes": null
    },
    {
      "id": "1912568",
      "postDate": "08/24/2022 20:17:47",
      "content": "<p>Great part of my features are from variation like last vs first, last vs avg, last vs max/min, also I tried to create new categorical features joining first and last original categorical, some features have great impact but I failed tuning my models.</p>",
      "rawMarkdown": "Great part of my features are from variation like last vs first, last vs avg, last vs max/min, also I tried to create new categorical features joining first and last original categorical, some features have great impact but I failed tuning my models.",
      "votes": null
    },
    {
      "id": "1912658",
      "postDate": "08/24/2022 22:47:26",
      "content": "<p>The same. Did .800 after fifth submission. But having no time to deep inside dataset, just tried simple ideas and blended it. Well, it's all about the time series feature engineering. Table competitions is often about fe and less about models and architectures. So, good luck! </p>",
      "rawMarkdown": "The same. Did .800 after fifth submission. But having no time to deep inside dataset, just tried simple ideas and blended it. Well, it's all about the time series feature engineering. Table competitions is often about fe and less about models and architectures. So, good luck!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1912548,
      "author_name": "fritzcremer",
      "author_url": "",
      "post_date": "08/24/2022 19:56:21",
      "content": "<p>You can get a decent score with only the last entry, that is true. But I don't think you can even get a medal with the last entry alone…</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1912568,
      "author_name": "alexandreg1998",
      "author_url": "",
      "post_date": "08/24/2022 20:17:47",
      "content": "<p>Great part of my features are from variation like last vs first, last vs avg, last vs max/min, also I tried to create new categorical features joining first and last original categorical, some features have great impact but I failed tuning my models.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1912658,
      "author_name": "sggpls",
      "author_url": "",
      "post_date": "08/24/2022 22:47:26",
      "content": "<p>The same. Did .800 after fifth submission. But having no time to deep inside dataset, just tried simple ideas and blended it. Well, it's all about the time series feature engineering. Table competitions is often about fe and less about models and architectures. So, good luck! </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1912543": "I joined the competitoin, to late, \nand got about 0.79\n\nonly Today, i came to a conclustion, that everybody, just took the last record for every customer,\nand all the high scores, are based on that concept.\n\nI think, that if i take, all records, or ever 2-3 last ones, it will improve the score, but i just didnt have the time to do it.\n\nI will try it later on, and will be happy to try to submot, just for curiousity.",
    "1912548": "You can get a decent score with only the last entry, that is true. But I don't think you can even get a medal with the last entry alone...",
    "1912568": "Great part of my features are from variation like last vs first, last vs avg, last vs max/min, also I tried to create new categorical features joining first and last original categorical, some features have great impact but I failed tuning my models.",
    "1912658": "The same. Did .800 after fifth submission. But having no time to deep inside dataset, just tried simple ideas and blended it. Well, it's all about the time series feature engineering. Table competitions is often about fe and less about models and architectures. So, good luck!"
  },
  "source": "meta"
}