{
  "id": 580235,
  "title": "What's your CV strategy and score?",
  "url": "/competitions/drw-crypto-market-prediction/discussion/580235",
  "author_name": "yunsuxiaozi",
  "post_date": "2025-05-23T09:11:15.817000",
  "votes": 7,
  "comment_count": 4,
  "views": 0,
  "content": "<pre><code>train[]=train.index//\n</code></pre>\n<p>CV:0.047,LB:0.056</p>",
  "messages": [
    {
      "id": 3207814,
      "postDate": "2025-05-23T09:11:15.817Z",
      "content": "<pre><code>train[]=train.index//\n</code></pre>\n<p>CV:0.047,LB:0.056</p>",
      "rawMarkdown": "```python\ntrain['fold']=train.index//110000\n```\n\nCV:0.047,LB:0.056",
      "votes": 7
    },
    {
      "id": 3211291,
      "postDate": "2025-05-28T09:17:42.010Z",
      "content": "<p>train.index is simply np.arange(len(train))? If done as this, there will be cases where one day gets split between folds and leaks information. </p>",
      "rawMarkdown": "train.index is simply np.arange(len(train))? If done as this, there will be cases where one day gets split between folds and leaks information. "
    },
    {
      "id": 3210748,
      "postDate": "2025-05-27T16:38:20.243Z",
      "content": "<p>My current idea is a classic time series split with shuffling afterwards, it seems to work pretty good.<br>\nCV: 0.04419 LB: 0.04812</p>",
      "rawMarkdown": "My current idea is a classic time series split with shuffling afterwards, it seems to work pretty good.\nCV: 0.04419 LB: 0.04812"
    },
    {
      "id": 3207925,
      "postDate": "2025-05-23T12:39:21.953Z",
      "content": "<p>I'm a sophomore majoring in Artificial Intelligence, and today is my first time participating in a Kaggle competition. I noticed that your model's CV and LB scores are very close, which is really impressive. Could you please share how you set up your model evaluation and validation strategy to achieve such similar CV and LB results? Also, I’m curious about your journey in machine learning and Kaggle—how did you get started and what has your experience been like?</p>",
      "rawMarkdown": "I'm a sophomore majoring in Artificial Intelligence, and today is my first time participating in a Kaggle competition. I noticed that your model's CV and LB scores are very close, which is really impressive. Could you please share how you set up your model evaluation and validation strategy to achieve such similar CV and LB results? Also, I’m curious about your journey in machine learning and Kaggle—how did you get started and what has your experience been like?"
    },
    {
      "id": 3208042,
      "postDate": "2025-05-23T15:44:46.730Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 3211291,
      "author_name": "RandomWalker",
      "author_url": "",
      "post_date": "2025-05-28T09:17:42.010000",
      "content": "<p>train.index is simply np.arange(len(train))? If done as this, there will be cases where one day gets split between folds and leaks information. </p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 3210748,
      "author_name": "MaxUhl98",
      "author_url": "",
      "post_date": "2025-05-27T16:38:20.243000",
      "content": "<p>My current idea is a classic time series split with shuffling afterwards, it seems to work pretty good.<br>\nCV: 0.04419 LB: 0.04812</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 3207925,
      "author_name": "Jason_Qian",
      "author_url": "",
      "post_date": "2025-05-23T12:39:21.953000",
      "content": "<p>I'm a sophomore majoring in Artificial Intelligence, and today is my first time participating in a Kaggle competition. I noticed that your model's CV and LB scores are very close, which is really impressive. Could you please share how you set up your model evaluation and validation strategy to achieve such similar CV and LB results? Also, I’m curious about your journey in machine learning and Kaggle—how did you get started and what has your experience been like?</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 3208042,
      "author_name": "",
      "author_url": "",
      "post_date": "2025-05-23T15:44:46.730000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3207814": "```python\ntrain['fold']=train.index//110000\n```\n\nCV:0.047,LB:0.056",
    "3211291": "train.index is simply np.arange(len(train))? If done as this, there will be cases where one day gets split between folds and leaks information. ",
    "3210748": "My current idea is a classic time series split with shuffling afterwards, it seems to work pretty good.\nCV: 0.04419 LB: 0.04812",
    "3207925": "I'm a sophomore majoring in Artificial Intelligence, and today is my first time participating in a Kaggle competition. I noticed that your model's CV and LB scores are very close, which is really impressive. Could you please share how you set up your model evaluation and validation strategy to achieve such similar CV and LB results? Also, I’m curious about your journey in machine learning and Kaggle—how did you get started and what has your experience been like?",
    "3208042": ""
  }
}