{
  "id": 655643,
  "title": "[CV vs LB] - Single Model",
  "url": "/competitions/MABe-mouse-behavior-detection/discussion/655643",
  "author_name": "",
  "post_date": "2025-12-08T15:23:52.733534100Z",
  "votes": 2,
  "comment_count": 4,
  "views": 0,
  "content": "<p>How are your correlations looking so far?\nThese are my two best runs till now (lab-wise 5-fold CatBoost):\nCV: 0.521 → LB: 0.486\nCV: 0.511 → LB: 0.484</p>",
  "messages": [
    {
      "id": "3367605",
      "postDate": "12/08/2025 15:23:52",
      "content": "<p>How are your correlations looking so far?\nThese are my two best runs till now (lab-wise 5-fold CatBoost):\nCV: 0.521 → LB: 0.486\nCV: 0.511 → LB: 0.484</p>",
      "rawMarkdown": "How are your correlations looking so far?\nThese are my two best runs till now (lab-wise 5-fold CatBoost):\nCV: 0.521 → LB: 0.486\nCV: 0.511 → LB: 0.484",
      "votes": null
    },
    {
      "id": "3367935",
      "postDate": "12/08/2025 20:27:52",
      "content": "<p>Single model GRU\nvalidation=0.453, LB=0.47</p>",
      "rawMarkdown": "Single model GRU\nvalidation=0.453, LB=0.47",
      "votes": null
    },
    {
      "id": "3368117",
      "postDate": "12/09/2025 01:42:29",
      "content": "<p>You use GRU? what feature do you use? My LSTM only perform better with some actions in section 8</p>",
      "rawMarkdown": "You use GRU? what feature do you use? My LSTM only perform better with some actions in section 8",
      "votes": null
    },
    {
      "id": "3368181",
      "postDate": "12/09/2025 03:34:03",
      "content": "<p>I was sure the meta for this competition would obviously be a heavy<br>\nOOF-based stacking ensemble with XGBoost, CatBoost, and LightGBM… </p>\n<p>Since the problem is a binary classification problem for each action,<br>\nand the labs where the data were collected aren’t continuous,<br>\nI thought GBDTs would be efficient for learning a coarse decision boundary.</p>\n<p>If we were to use a GRU,<br>\nI was thinking it wouldn’t model the entire action sequence,<br>\nbut rather serve as an auxiliary module to help the ensemble with actions like<br>\nattacks or position-dependent behaviors.</p>\n<p>To be honest,<br>\nI’m having a hard time even just opening the files for this competition.</p>",
      "rawMarkdown": "I was sure the meta for this competition would obviously be a heavy   \nOOF-based stacking ensemble with XGBoost, CatBoost, and LightGBM… \n\nSince the problem is a binary classification problem for each action,  \nand the labs where the data were collected aren’t continuous,   \nI thought GBDTs would be efficient for learning a coarse decision boundary.\n\nIf we were to use a GRU,  \nI was thinking it wouldn’t model the entire action sequence,   \nbut rather serve as an auxiliary module to help the ensemble with actions like   \nattacks or position-dependent behaviors.\n\nTo be honest,   \nI’m having a hard time even just opening the files for this competition.",
      "votes": null
    },
    {
      "id": "3368465",
      "postDate": "12/09/2025 08:41:08",
      "content": "<p>Yes I’m using a GRU. Not sure I understand, what do you mean by section 8?</p>",
      "rawMarkdown": "Yes I’m using a GRU. Not sure I understand, what do you mean by section 8?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3367935,
      "author_name": "amitaharoni",
      "author_url": "",
      "post_date": "12/08/2025 20:27:52",
      "content": "<p>Single model GRU\nvalidation=0.453, LB=0.47</p>",
      "votes": null,
      "replies": [
        {
          "id": 3368117,
          "author_name": "menghaozhao",
          "author_url": "",
          "post_date": "12/09/2025 01:42:29",
          "content": "<p>You use GRU? what feature do you use? My LSTM only perform better with some actions in section 8</p>",
          "votes": null,
          "replies": [
            {
              "id": 3368465,
              "author_name": "amitaharoni",
              "author_url": "",
              "post_date": "12/09/2025 08:41:08",
              "content": "<p>Yes I’m using a GRU. Not sure I understand, what do you mean by section 8?</p>",
              "votes": null,
              "replies": []
            }
          ]
        },
        {
          "id": 3368181,
          "author_name": "woosungyoon",
          "author_url": "",
          "post_date": "12/09/2025 03:34:03",
          "content": "<p>I was sure the meta for this competition would obviously be a heavy<br>\nOOF-based stacking ensemble with XGBoost, CatBoost, and LightGBM… </p>\n<p>Since the problem is a binary classification problem for each action,<br>\nand the labs where the data were collected aren’t continuous,<br>\nI thought GBDTs would be efficient for learning a coarse decision boundary.</p>\n<p>If we were to use a GRU,<br>\nI was thinking it wouldn’t model the entire action sequence,<br>\nbut rather serve as an auxiliary module to help the ensemble with actions like<br>\nattacks or position-dependent behaviors.</p>\n<p>To be honest,<br>\nI’m having a hard time even just opening the files for this competition.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3367605": "How are your correlations looking so far?\nThese are my two best runs till now (lab-wise 5-fold CatBoost):\nCV: 0.521 → LB: 0.486\nCV: 0.511 → LB: 0.484",
    "3367935": "Single model GRU\nvalidation=0.453, LB=0.47",
    "3368117": "You use GRU? what feature do you use? My LSTM only perform better with some actions in section 8",
    "3368181": "I was sure the meta for this competition would obviously be a heavy   \nOOF-based stacking ensemble with XGBoost, CatBoost, and LightGBM… \n\nSince the problem is a binary classification problem for each action,  \nand the labs where the data were collected aren’t continuous,   \nI thought GBDTs would be efficient for learning a coarse decision boundary.\n\nIf we were to use a GRU,  \nI was thinking it wouldn’t model the entire action sequence,   \nbut rather serve as an auxiliary module to help the ensemble with actions like   \nattacks or position-dependent behaviors.\n\nTo be honest,   \nI’m having a hard time even just opening the files for this competition.",
    "3368465": "Yes I’m using a GRU. Not sure I understand, what do you mean by section 8?"
  },
  "source": "meta"
}