{
  "id": 323153,
  "title": "how many Candidates are \"enough\" for reranker or ranker?",
  "url": "/competitions/h-and-m-personalized-fashion-recommendations/discussion/323153",
  "author_name": "",
  "post_date": "2022-05-05T03:15:32.821714200Z",
  "votes": 6,
  "comment_count": 3,
  "views": 0,
  "content": "<p>when recall items from different recall method(like nn, itemcf, pop), how many Candidates are enough for train a reranker or ranker(like LightGBMRanker)?</p>\n<p>the answer to me is 24, what's yours?</p>",
  "messages": [
    {
      "id": "1778013",
      "postDate": "05/05/2022 03:15:32",
      "content": "<p>when recall items from different recall method(like nn, itemcf, pop), how many Candidates are enough for train a reranker or ranker(like LightGBMRanker)?</p>\n<p>the answer to me is 24, what's yours?</p>",
      "rawMarkdown": "when recall items from different recall method(like nn, itemcf, pop), how many Candidates are enough for train a reranker or ranker(like LightGBMRanker)?\n\nthe answer to me is 24, what's yours?",
      "votes": null
    },
    {
      "id": "1778038",
      "postDate": "05/05/2022 03:53:33",
      "content": "<p>In my case, on average the number is 200, when I keep incresing the number, the performance didn't increase any more.</p>",
      "rawMarkdown": "In my case, on average the number is 200, when I keep incresing the number, the performance didn't increase any more.",
      "votes": null
    },
    {
      "id": "1778065",
      "postDate": "05/05/2022 04:36:35",
      "content": "<p>thanks for your sharing. by the way, </p>\n<ol>\n<li>so you test table is about 1m(user)* 200 * N(recall_ways)? it's about 1billion samples?</li>\n<li>is the number heavily influence the performance?</li>\n</ol>",
      "rawMarkdown": "thanks for your sharing. by the way, \n1. so you test table is about 1m(user)* 200 * N(recall_ways)? it's about 1billion samples?\n2. is the number heavily influence the performance?",
      "votes": null
    },
    {
      "id": "1778069",
      "postDate": "05/05/2022 04:48:46",
      "content": "<ol>\n<li>200 is the overall number from all recall ways. So my test table is actually about 400 million samples.</li>\n<li>I would not call it's \"heavily\" influence the performance.</li>\n</ol>",
      "rawMarkdown": "1. 200 is the overall number from all recall ways. So my test table is actually about 400 million samples.\n2. I would not call it's \"heavily\" influence the performance.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1778038,
      "author_name": "lihaorocky",
      "author_url": "",
      "post_date": "05/05/2022 03:53:33",
      "content": "<p>In my case, on average the number is 200, when I keep incresing the number, the performance didn't increase any more.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1778065,
          "author_name": "biubiug",
          "author_url": "",
          "post_date": "05/05/2022 04:36:35",
          "content": "<p>thanks for your sharing. by the way, </p>\n<ol>\n<li>so you test table is about 1m(user)* 200 * N(recall_ways)? it's about 1billion samples?</li>\n<li>is the number heavily influence the performance?</li>\n</ol>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1778069,
          "author_name": "lihaorocky",
          "author_url": "",
          "post_date": "05/05/2022 04:48:46",
          "content": "<ol>\n<li>200 is the overall number from all recall ways. So my test table is actually about 400 million samples.</li>\n<li>I would not call it's \"heavily\" influence the performance.</li>\n</ol>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1778013": "when recall items from different recall method(like nn, itemcf, pop), how many Candidates are enough for train a reranker or ranker(like LightGBMRanker)?\n\nthe answer to me is 24, what's yours?",
    "1778038": "In my case, on average the number is 200, when I keep incresing the number, the performance didn't increase any more.",
    "1778065": "thanks for your sharing. by the way, \n1. so you test table is about 1m(user)* 200 * N(recall_ways)? it's about 1billion samples?\n2. is the number heavily influence the performance?",
    "1778069": "1. 200 is the overall number from all recall ways. So my test table is actually about 400 million samples.\n2. I would not call it's \"heavily\" influence the performance."
  },
  "source": "meta"
}