{
  "id": 609332,
  "title": "FYI: differentiable pytorch transit curve fitting - pylightcurve-torch",
  "url": "/competitions/ariel-data-challenge-2025/discussion/609332",
  "author_name": "hengck23",
  "post_date": "2025-09-26T00:06:56.290000",
  "votes": 8,
  "comment_count": 3,
  "views": 0,
  "content": "<p>i did't have much time in this competition and briefly explored it for a week. <br>\ni am trying to use pyshics aware deep net PINN to solve it end-to-end, so i experiment with is open source:  </p>\n<p><a href=\"https://github.com/ucl-exoplanets/pylightcurve-torch\" target=\"_blank\">https://github.com/ucl-exoplanets/pylightcurve-torch</a><br>\npaper: <a href=\"https://arxiv.org/abs/2011.02030\" target=\"_blank\">https://arxiv.org/abs/2011.02030</a></p>\n<p>I verify that it pylightcurve-torch has exact the same results of batman transit curve. <br>\nI can recover/fit from noiseless flux curve or batman curve + my injected version spike noise + systematic trend.<br>\nBut I failed miserably in kaggle dataset. Maybe my modeling is wrong and i hope to learn more from the 2025 solutions.  </p>\n<hr>\n<p>pylightcurve-torch maybe useful for future competition. <br>\nSo I put it in this post as a record ….  see you next year in Ariel 2026?</p>",
  "messages": [
    {
      "id": 3294352,
      "postDate": "2025-09-26T00:06:56.290Z",
      "content": "<p>i did't have much time in this competition and briefly explored it for a week. <br>\ni am trying to use pyshics aware deep net PINN to solve it end-to-end, so i experiment with is open source:  </p>\n<p><a href=\"https://github.com/ucl-exoplanets/pylightcurve-torch\" target=\"_blank\">https://github.com/ucl-exoplanets/pylightcurve-torch</a><br>\npaper: <a href=\"https://arxiv.org/abs/2011.02030\" target=\"_blank\">https://arxiv.org/abs/2011.02030</a></p>\n<p>I verify that it pylightcurve-torch has exact the same results of batman transit curve. <br>\nI can recover/fit from noiseless flux curve or batman curve + my injected version spike noise + systematic trend.<br>\nBut I failed miserably in kaggle dataset. Maybe my modeling is wrong and i hope to learn more from the 2025 solutions.  </p>\n<hr>\n<p>pylightcurve-torch maybe useful for future competition. <br>\nSo I put it in this post as a record ….  see you next year in Ariel 2026?</p>",
      "rawMarkdown": "i did't have much time in this competition and briefly explored it for a week. \ni am trying to use pyshics aware deep net PINN to solve it end-to-end, so i experiment with is open source:  \n\nhttps://github.com/ucl-exoplanets/pylightcurve-torch\npaper: https://arxiv.org/abs/2011.02030\n\nI verify that it pylightcurve-torch has exact the same results of batman transit curve. \nI can recover/fit from noiseless flux curve or batman curve + my injected version spike noise + systematic trend.\nBut I failed miserably in kaggle dataset. Maybe my modeling is wrong and i hope to learn more from the 2025 solutions.  \n\n---\n\npylightcurve-torch maybe useful for future competition. \nSo I put it in this post as a record ....  see you next year in Ariel 2026?",
      "votes": 8
    },
    {
      "id": 3294353,
      "postDate": "2025-09-26T00:13:41.007Z",
      "content": "<p>We used it.</p>",
      "rawMarkdown": "We used it.",
      "votes": 1,
      "replies": [
        {
          "id": 3294354,
          "postDate": "2025-09-26T00:18:46.553Z",
          "content": "<p>oh. love to hear from your solution! 😍</p>",
          "rawMarkdown": "oh. love to hear from your solution! 😍",
          "replies": [
            {
              "id": 3294443,
              "postDate": "2025-09-26T07:29:01.243Z",
              "content": "<p>But we found out about it only a week before end. It has been a mad sprint.</p>",
              "rawMarkdown": "But we found out about it only a week before end. It has been a mad sprint.\n"
            }
          ]
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 3294353,
      "author_name": "CPMP",
      "author_url": "",
      "post_date": "2025-09-26T00:13:41.007000",
      "content": "<p>We used it.</p>",
      "votes": 1,
      "replies": [
        {
          "id": 3294354,
          "author_name": "hengck23",
          "author_url": "",
          "post_date": "2025-09-26T00:18:46.553000",
          "content": "<p>oh. love to hear from your solution! 😍</p>",
          "votes": 0,
          "replies": [
            {
              "id": 3294443,
              "author_name": "CPMP",
              "author_url": "",
              "post_date": "2025-09-26T07:29:01.243000",
              "content": "<p>But we found out about it only a week before end. It has been a mad sprint.</p>",
              "votes": 0,
              "replies": []
            }
          ]
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3294352": "i did't have much time in this competition and briefly explored it for a week. \ni am trying to use pyshics aware deep net PINN to solve it end-to-end, so i experiment with is open source:  \n\nhttps://github.com/ucl-exoplanets/pylightcurve-torch\npaper: https://arxiv.org/abs/2011.02030\n\nI verify that it pylightcurve-torch has exact the same results of batman transit curve. \nI can recover/fit from noiseless flux curve or batman curve + my injected version spike noise + systematic trend.\nBut I failed miserably in kaggle dataset. Maybe my modeling is wrong and i hope to learn more from the 2025 solutions.  \n\n---\n\npylightcurve-torch maybe useful for future competition. \nSo I put it in this post as a record ....  see you next year in Ariel 2026?",
    "3294353": "We used it."
  }
}