{
  "id": 590804,
  "title": "What's your r2_score of your prediction when your pearson coeff is below 0.15?",
  "url": "/competitions/drw-crypto-market-prediction/discussion/590804",
  "author_name": "",
  "post_date": "2025-07-23T05:48:49.742764500Z",
  "votes": null,
  "comment_count": 4,
  "views": 0,
  "content": "<p>I got negative r2_score when the pearson coeff is about 0.1, not sure what's the problem.</p>",
  "messages": [
    {
      "id": "3252637",
      "postDate": "07/23/2025 05:48:49",
      "content": "<p>I got negative r2_score when the pearson coeff is about 0.1, not sure what's the problem.</p>",
      "rawMarkdown": "I got negative r2_score when the pearson coeff is about 0.1, not sure what's the problem.",
      "votes": null
    },
    {
      "id": "3253378",
      "postDate": "07/24/2025 15:00:01",
      "content": "<p>I also found r2 and Pearson not correlated in the given dataset</p>",
      "rawMarkdown": "I also found r2 and Pearson not correlated in the given dataset",
      "votes": null
    },
    {
      "id": "3253397",
      "postDate": "07/24/2025 15:35:41",
      "content": "<p>You can expect an overestimation of R², as it is approximately equal to the square of the Pearson correlation. In your case, when your Pearson correlation is below 0.15, R²&lt;(0.15)²~0.0225</p>",
      "rawMarkdown": "You can expect an overestimation of R², as it is approximately equal to the square of the Pearson correlation. In your case, when your Pearson correlation is below 0.15, R²<(0.15)²~0.0225",
      "votes": null
    },
    {
      "id": "3253482",
      "postDate": "07/24/2025 19:01:34",
      "content": "<p>It depends on how you define r2. It can be either 1 - SSR/SST or (pearson)^2. In the former case  you can indeed have negative r2, and it is not an appropriate measure of fit anymore.</p>",
      "rawMarkdown": "It depends on how you define r2. It can be either 1 - SSR/SST or (pearson)^2. In the former case  you can indeed have negative r2, and it is not an appropriate measure of fit anymore.",
      "votes": null
    },
    {
      "id": "3253604",
      "postDate": "07/25/2025 01:50:18",
      "content": "<p>Yeah, I use 1 - SSR/SST. Why isn't it an appropriate measure of fit? Negative r2 means the prediction is worse than the average of the labels(which is about 0, actually I also use 0 instead of averaged label in 1-SSR/SST, it's still negative). Does it mean the prediction has some power in predicting sign (positive or negative), but it doesn't do good job in predicting amplitude?</p>",
      "rawMarkdown": "Yeah, I use 1 - SSR/SST. Why isn't it an appropriate measure of fit? Negative r2 means the prediction is worse than the average of the labels(which is about 0, actually I also use 0 instead of averaged label in 1-SSR/SST, it's still negative). Does it mean the prediction has some power in predicting sign (positive or negative), but it doesn't do good job in predicting amplitude?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3253378,
      "author_name": "alexzhongs",
      "author_url": "",
      "post_date": "07/24/2025 15:00:01",
      "content": "<p>I also found r2 and Pearson not correlated in the given dataset</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 3253397,
      "author_name": "youneseloiarm",
      "author_url": "",
      "post_date": "07/24/2025 15:35:41",
      "content": "<p>You can expect an overestimation of R², as it is approximately equal to the square of the Pearson correlation. In your case, when your Pearson correlation is below 0.15, R²&lt;(0.15)²~0.0225</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 3253482,
      "author_name": "yannfb",
      "author_url": "",
      "post_date": "07/24/2025 19:01:34",
      "content": "<p>It depends on how you define r2. It can be either 1 - SSR/SST or (pearson)^2. In the former case  you can indeed have negative r2, and it is not an appropriate measure of fit anymore.</p>",
      "votes": null,
      "replies": [
        {
          "id": 3253604,
          "author_name": "lyraaa",
          "author_url": "",
          "post_date": "07/25/2025 01:50:18",
          "content": "<p>Yeah, I use 1 - SSR/SST. Why isn't it an appropriate measure of fit? Negative r2 means the prediction is worse than the average of the labels(which is about 0, actually I also use 0 instead of averaged label in 1-SSR/SST, it's still negative). Does it mean the prediction has some power in predicting sign (positive or negative), but it doesn't do good job in predicting amplitude?</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3252637": "I got negative r2_score when the pearson coeff is about 0.1, not sure what's the problem.",
    "3253378": "I also found r2 and Pearson not correlated in the given dataset",
    "3253397": "You can expect an overestimation of R², as it is approximately equal to the square of the Pearson correlation. In your case, when your Pearson correlation is below 0.15, R²<(0.15)²~0.0225",
    "3253482": "It depends on how you define r2. It can be either 1 - SSR/SST or (pearson)^2. In the former case  you can indeed have negative r2, and it is not an appropriate measure of fit anymore.",
    "3253604": "Yeah, I use 1 - SSR/SST. Why isn't it an appropriate measure of fit? Negative r2 means the prediction is worse than the average of the labels(which is about 0, actually I also use 0 instead of averaged label in 1-SSR/SST, it's still negative). Does it mean the prediction has some power in predicting sign (positive or negative), but it doesn't do good job in predicting amplitude?"
  },
  "source": "meta"
}