{
  "id": 266805,
  "title": "Simulated Target Generator",
  "url": "/competitions/seti-breakthrough-listen/discussion/266805",
  "author_name": "Giba",
  "post_date": "2021-08-20T14:39:04.593000",
  "votes": 11,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Some teams used a signal generator to build artificial targets in this competition. <a href=\"https://www.kaggle.com/titericz/simulated-target?scriptVersionId=72515923\" target=\"_blank\">HERE</a> is my implementation. </p>\n<p>I was able to score 0.77+ in Public/Private LB training a model using train+test control images and adding that simulated signals. The idea of using test control images to train was to diminish the effect of the different background patterns found only in test. Besides the fact that it didn't scores high it may be a good candidate for a blend since it presents low correlation with models trained on competition ground truth.</p>",
  "messages": [
    {
      "id": 1483267,
      "postDate": "2021-08-20T14:39:04.593Z",
      "content": "<p>Some teams used a signal generator to build artificial targets in this competition. <a href=\"https://www.kaggle.com/titericz/simulated-target?scriptVersionId=72515923\" target=\"_blank\">HERE</a> is my implementation. </p>\n<p>I was able to score 0.77+ in Public/Private LB training a model using train+test control images and adding that simulated signals. The idea of using test control images to train was to diminish the effect of the different background patterns found only in test. Besides the fact that it didn't scores high it may be a good candidate for a blend since it presents low correlation with models trained on competition ground truth.</p>",
      "rawMarkdown": "Some teams used a signal generator to build artificial targets in this competition. [HERE](https://www.kaggle.com/titericz/simulated-target?scriptVersionId=72515923) is my implementation. \n\nI was able to score 0.77+ in Public/Private LB training a model using train+test control images and adding that simulated signals. The idea of using test control images to train was to diminish the effect of the different background patterns found only in test. Besides the fact that it didn't scores high it may be a good candidate for a blend since it presents low correlation with models trained on competition ground truth.",
      "votes": 11
    },
    {
      "id": 1483286,
      "postDate": "2021-08-20T14:56:37.780Z",
      "content": "<p>Thanks for sharing this, it was really useful. Upvoting! ⬆️</p>",
      "rawMarkdown": "Thanks for sharing this, it was really useful. Upvoting! ⬆️",
      "votes": 1
    },
    {
      "id": 1483296,
      "postDate": "2021-08-20T15:03:46.927Z",
      "content": "<p>Wow, LB 0.77+ using generated data. That's impressive! Great idea</p>\n<p>I didn't realize there was a previous competition with previous competition data <a href=\"https://www.kaggle.com/tentotheminus9/seti-data\" target=\"_blank\">here</a>. And GitHub repositories showing code for generating signals like <a href=\"https://github.com/setiQuest/ML4SETI\" target=\"_blank\">here</a> and <a href=\"https://github.com/bbrzycki/setigen\" target=\"_blank\">here</a>. That's helpful for understanding this year's Kaggle SETI comp.</p>",
      "rawMarkdown": "Wow, LB 0.77+ using generated data. That's impressive! Great idea\n\nI didn't realize there was a previous competition with previous competition data [here][1]. And GitHub repositories showing code for generating signals like [here][3] and [here][2]. That's helpful for understanding this year's Kaggle SETI comp.\n\n[1]: https://www.kaggle.com/tentotheminus9/seti-data\n[2]: https://github.com/bbrzycki/setigen\n[3]: https://github.com/setiQuest/ML4SETI",
      "votes": 2
    }
  ],
  "comments": [
    {
      "id": 1483286,
      "author_name": "Rishiraj Acharya",
      "author_url": "",
      "post_date": "2021-08-20T14:56:37.780000",
      "content": "<p>Thanks for sharing this, it was really useful. Upvoting! ⬆️</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 1483296,
      "author_name": "Chris Deotte",
      "author_url": "",
      "post_date": "2021-08-20T15:03:46.927000",
      "content": "<p>Wow, LB 0.77+ using generated data. That's impressive! Great idea</p>\n<p>I didn't realize there was a previous competition with previous competition data <a href=\"https://www.kaggle.com/tentotheminus9/seti-data\" target=\"_blank\">here</a>. And GitHub repositories showing code for generating signals like <a href=\"https://github.com/setiQuest/ML4SETI\" target=\"_blank\">here</a> and <a href=\"https://github.com/bbrzycki/setigen\" target=\"_blank\">here</a>. That's helpful for understanding this year's Kaggle SETI comp.</p>",
      "votes": 2,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1483267": "Some teams used a signal generator to build artificial targets in this competition. [HERE](https://www.kaggle.com/titericz/simulated-target?scriptVersionId=72515923) is my implementation. \n\nI was able to score 0.77+ in Public/Private LB training a model using train+test control images and adding that simulated signals. The idea of using test control images to train was to diminish the effect of the different background patterns found only in test. Besides the fact that it didn't scores high it may be a good candidate for a blend since it presents low correlation with models trained on competition ground truth.",
    "1483286": "Thanks for sharing this, it was really useful. Upvoting! ⬆️",
    "1483296": "Wow, LB 0.77+ using generated data. That's impressive! Great idea\n\nI didn't realize there was a previous competition with previous competition data [here][1]. And GitHub repositories showing code for generating signals like [here][3] and [here][2]. That's helpful for understanding this year's Kaggle SETI comp.\n\n[1]: https://www.kaggle.com/tentotheminus9/seti-data\n[2]: https://github.com/bbrzycki/setigen\n[3]: https://github.com/setiQuest/ML4SETI"
  }
}