{
  "id": 376905,
  "title": "Is it necessary for us to write an recall@20 metric for the competition",
  "url": "/competitions/otto-recommender-system/discussion/376905",
  "author_name": "",
  "post_date": "2023-01-09T08:14:28.211026400Z",
  "votes": null,
  "comment_count": 3,
  "views": 0,
  "content": "<p>The metric in the log of XGBoost is map, which is inconsistent with the metric used in the competition. Is it necessary for us to write an metric  for the competition. If not, how to judge whether the model should stop training according to the map?</p>",
  "messages": [
    {
      "id": "2092205",
      "postDate": "01/09/2023 08:14:28",
      "content": "<p>The metric in the log of XGBoost is map, which is inconsistent with the metric used in the competition. Is it necessary for us to write an metric  for the competition. If not, how to judge whether the model should stop training according to the map?</p>",
      "rawMarkdown": "The metric in the log of XGBoost is map, which is inconsistent with the metric used in the competition. Is it necessary for us to write an metric  for the competition. If not, how to judge whether the model should stop training according to the map?",
      "votes": null
    },
    {
      "id": "2092297",
      "postDate": "01/09/2023 09:23:47",
      "content": "<p>You don't have to compute competition metric every time after fitting a tree. It would be inefficient. You can compute your final external validation score after training ends.</p>",
      "rawMarkdown": "You don't have to compute competition metric every time after fitting a tree. It would be inefficient. You can compute your final external validation score after training ends.",
      "votes": null
    },
    {
      "id": "2093479",
      "postDate": "01/10/2023 03:21:55",
      "content": "<p>agree so    </p>",
      "rawMarkdown": "agree so",
      "votes": null
    },
    {
      "id": "2098025",
      "postDate": "01/13/2023 06:54:04",
      "content": "<p>Thanks, that's how I currently do it</p>",
      "rawMarkdown": "Thanks, that's how I currently do it",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2092297,
      "author_name": "gunesevitan",
      "author_url": "",
      "post_date": "01/09/2023 09:23:47",
      "content": "<p>You don't have to compute competition metric every time after fitting a tree. It would be inefficient. You can compute your final external validation score after training ends.</p>",
      "votes": null,
      "replies": [
        {
          "id": 2093479,
          "author_name": "earlee0412",
          "author_url": "",
          "post_date": "01/10/2023 03:21:55",
          "content": "<p>agree so    </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 2098025,
          "author_name": "brycexxx",
          "author_url": "",
          "post_date": "01/13/2023 06:54:04",
          "content": "<p>Thanks, that's how I currently do it</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2092205": "The metric in the log of XGBoost is map, which is inconsistent with the metric used in the competition. Is it necessary for us to write an metric  for the competition. If not, how to judge whether the model should stop training according to the map?",
    "2092297": "You don't have to compute competition metric every time after fitting a tree. It would be inefficient. You can compute your final external validation score after training ends.",
    "2093479": "agree so",
    "2098025": "Thanks, that's how I currently do it"
  },
  "source": "meta"
}