{
  "id": 589216,
  "title": "Still Poor CV - LB Correlation",
  "url": "/competitions/drw-crypto-market-prediction/discussion/589216",
  "author_name": "",
  "post_date": "2025-07-11T03:22:27.890929100Z",
  "votes": 5,
  "comment_count": 6,
  "views": 0,
  "content": "<p>Does anyone else also experience that the correlation still didn't get any better with the introduction of a new test.csv data? </p>",
  "messages": [
    {
      "id": "3246496",
      "postDate": "07/11/2025 03:22:27",
      "content": "<p>Does anyone else also experience that the correlation still didn't get any better with the introduction of a new test.csv data? </p>",
      "rawMarkdown": "Does anyone else also experience that the correlation still didn't get any better with the introduction of a new test.csv data?",
      "votes": null
    },
    {
      "id": "3246687",
      "postDate": "07/11/2025 11:01:54",
      "content": "<p>my CV score is okay, but the LB score is negative …. anyone same issue with me? </p>",
      "rawMarkdown": "my CV score is okay, but the LB score is negative .... anyone same issue with me?",
      "votes": null
    },
    {
      "id": "3246910",
      "postDate": "07/11/2025 18:58:39",
      "content": "<p>Yes - the same thing happened to me!</p>",
      "rawMarkdown": "Yes - the same thing happened to me!",
      "votes": null
    },
    {
      "id": "3246940",
      "postDate": "07/11/2025 20:10:24",
      "content": "<p>The discrepancies between Train -&gt; CV and CV -&gt; LB are at the heart of this challenge. The data has a very low signal to noise ratio, which is actually very realistic for financial markets.</p>\n<p>With the new dataset I am trying new strategies, and I have recently found that training an adversarial model to counter balance suspected over-fitting is showing promise. I hope that it will continue to perform well when combined with regularization, drop out, and noise injection.</p>",
      "rawMarkdown": "The discrepancies between Train -> CV and CV -> LB are at the heart of this challenge. The data has a very low signal to noise ratio, which is actually very realistic for financial markets.\n\nWith the new dataset I am trying new strategies, and I have recently found that training an adversarial model to counter balance suspected over-fitting is showing promise. I hope that it will continue to perform well when combined with regularization, drop out, and noise injection.",
      "votes": null
    },
    {
      "id": "3247304",
      "postDate": "07/12/2025 14:06:35",
      "content": "<p>What is your CV score?</p>",
      "rawMarkdown": "What is your CV score?",
      "votes": null
    },
    {
      "id": "3249478",
      "postDate": "07/16/2025 14:59:12",
      "content": "<p>do you find adversarial useful, I used to find the data too noisy for it to tell</p>",
      "rawMarkdown": "do you find adversarial useful, I used to find the data too noisy for it to tell",
      "votes": null
    },
    {
      "id": "3249481",
      "postDate": "07/16/2025 15:07:13",
      "content": "<p>I found it slightly more useful than typical strategies like regularization.</p>",
      "rawMarkdown": "I found it slightly more useful than typical strategies like regularization.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3246687,
      "author_name": "weizhengchoo",
      "author_url": "",
      "post_date": "07/11/2025 11:01:54",
      "content": "<p>my CV score is okay, but the LB score is negative …. anyone same issue with me? </p>",
      "votes": null,
      "replies": [
        {
          "id": 3246910,
          "author_name": "jwangexe",
          "author_url": "",
          "post_date": "07/11/2025 18:58:39",
          "content": "<p>Yes - the same thing happened to me!</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 3246940,
      "author_name": "taylorsamarel",
      "author_url": "",
      "post_date": "07/11/2025 20:10:24",
      "content": "<p>The discrepancies between Train -&gt; CV and CV -&gt; LB are at the heart of this challenge. The data has a very low signal to noise ratio, which is actually very realistic for financial markets.</p>\n<p>With the new dataset I am trying new strategies, and I have recently found that training an adversarial model to counter balance suspected over-fitting is showing promise. I hope that it will continue to perform well when combined with regularization, drop out, and noise injection.</p>",
      "votes": null,
      "replies": [
        {
          "id": 3249478,
          "author_name": "alexzhongs",
          "author_url": "",
          "post_date": "07/16/2025 14:59:12",
          "content": "<p>do you find adversarial useful, I used to find the data too noisy for it to tell</p>",
          "votes": null,
          "replies": [
            {
              "id": 3249481,
              "author_name": "taylorsamarel",
              "author_url": "",
              "post_date": "07/16/2025 15:07:13",
              "content": "<p>I found it slightly more useful than typical strategies like regularization.</p>",
              "votes": null,
              "replies": []
            }
          ]
        }
      ]
    },
    {
      "id": 3247304,
      "author_name": "jankowalski2000",
      "author_url": "",
      "post_date": "07/12/2025 14:06:35",
      "content": "<p>What is your CV score?</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3246496": "Does anyone else also experience that the correlation still didn't get any better with the introduction of a new test.csv data?",
    "3246687": "my CV score is okay, but the LB score is negative .... anyone same issue with me?",
    "3246910": "Yes - the same thing happened to me!",
    "3246940": "The discrepancies between Train -> CV and CV -> LB are at the heart of this challenge. The data has a very low signal to noise ratio, which is actually very realistic for financial markets.\n\nWith the new dataset I am trying new strategies, and I have recently found that training an adversarial model to counter balance suspected over-fitting is showing promise. I hope that it will continue to perform well when combined with regularization, drop out, and noise injection.",
    "3247304": "What is your CV score?",
    "3249478": "do you find adversarial useful, I used to find the data too noisy for it to tell",
    "3249481": "I found it slightly more useful than typical strategies like regularization."
  },
  "source": "meta"
}