{
  "id": 413349,
  "title": "GO Term count analysis for helping class sampling",
  "url": "/competitions/cafa-5-protein-function-prediction/discussion/413349",
  "author_name": "",
  "post_date": "2023-05-28T07:11:46.237090Z",
  "votes": 13,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I analyzed GO Term counting and found something interesting :)<br>\nThere is something like pareto rule. I summraize below.<br>\nA simple insight is that focusing on MFO is better for evaluation.<br>\nThe number of MFO's unique term is less than others (this imply to be easy for prediction), but The mean &amp; max IA weight of it is larger than others.<br>\n<a href=\"\" target=\"_blank\">Result is from this kernel</a><br>\n<br></p>\n<h3>Result1</h3>\n<ul>\n<li>For example, You can interprete the data at first row of 'Cumulative Count Threshold = 0.9' is 90% of total samples are from 9.83% of total GO Term <br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F6610354%2F577c7a6b34d043fbd544cc527981d38a%2Fpic1.png?generation=1685292788641595&amp;alt=media\" alt=\"\"></li>\n</ul>\n<h3>Result2</h3>\n<ul>\n<li>How about IA weight is larger than zero?<br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F6610354%2Fa9f9dcf79b06b673b6751c0ed027f069%2Fpic2.png?generation=1685292799492538&amp;alt=media\" alt=\"\"></li>\n</ul>",
  "messages": [
    {
      "id": "2277914",
      "postDate": "05/28/2023 07:11:46",
      "content": "<p>I analyzed GO Term counting and found something interesting :)<br>\nThere is something like pareto rule. I summraize below.<br>\nA simple insight is that focusing on MFO is better for evaluation.<br>\nThe number of MFO's unique term is less than others (this imply to be easy for prediction), but The mean &amp; max IA weight of it is larger than others.<br>\n<a href=\"\" target=\"_blank\">Result is from this kernel</a><br>\n<br></p>\n<h3>Result1</h3>\n<ul>\n<li>For example, You can interprete the data at first row of 'Cumulative Count Threshold = 0.9' is 90% of total samples are from 9.83% of total GO Term <br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F6610354%2F577c7a6b34d043fbd544cc527981d38a%2Fpic1.png?generation=1685292788641595&amp;alt=media\" alt=\"\"></li>\n</ul>\n<h3>Result2</h3>\n<ul>\n<li>How about IA weight is larger than zero?<br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F6610354%2Fa9f9dcf79b06b673b6751c0ed027f069%2Fpic2.png?generation=1685292799492538&amp;alt=media\" alt=\"\"></li>\n</ul>",
      "rawMarkdown": "I analyzed GO Term counting and found something interesting :)\nThere is something like pareto rule. I summraize below.\nA simple insight is that focusing on MFO is better for evaluation.\nThe number of MFO's unique term is less than others (this imply to be easy for prediction), but The mean & max IA weight of it is larger than others.\n[Result is from this kernel](\"https://www.kaggle.com/code/cafelatte1/cafa-eda-target-v1\")\n<br>\n\n### Result1\n* For example, You can interprete the data at first row of 'Cumulative Count Threshold = 0.9' is 90% of total samples are from 9.83% of total GO Term \n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F6610354%2F577c7a6b34d043fbd544cc527981d38a%2Fpic1.png?generation=1685292788641595&alt=media)\n\n### Result2\n* How about IA weight is larger than zero?\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F6610354%2Fa9f9dcf79b06b673b6751c0ed027f069%2Fpic2.png?generation=1685292799492538&alt=media)",
      "votes": null
    },
    {
      "id": "2279700",
      "postDate": "05/29/2023 14:34:25",
      "content": "<p>That’s a very interesting analysis. I think instead of taking the top K most represented GO term classes among all samples, it’s smarter to consider top K/3 terms of each of the 3 generic function classes. </p>",
      "rawMarkdown": "That’s a very interesting analysis. I think instead of taking the top K most represented GO term classes among all samples, it’s smarter to consider top K/3 terms of each of the 3 generic function classes.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2279700,
      "author_name": "henriupton",
      "author_url": "",
      "post_date": "05/29/2023 14:34:25",
      "content": "<p>That’s a very interesting analysis. I think instead of taking the top K most represented GO term classes among all samples, it’s smarter to consider top K/3 terms of each of the 3 generic function classes. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2277914": "I analyzed GO Term counting and found something interesting :)\nThere is something like pareto rule. I summraize below.\nA simple insight is that focusing on MFO is better for evaluation.\nThe number of MFO's unique term is less than others (this imply to be easy for prediction), but The mean & max IA weight of it is larger than others.\n[Result is from this kernel](\"https://www.kaggle.com/code/cafelatte1/cafa-eda-target-v1\")\n<br>\n\n### Result1\n* For example, You can interprete the data at first row of 'Cumulative Count Threshold = 0.9' is 90% of total samples are from 9.83% of total GO Term \n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F6610354%2F577c7a6b34d043fbd544cc527981d38a%2Fpic1.png?generation=1685292788641595&alt=media)\n\n### Result2\n* How about IA weight is larger than zero?\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F6610354%2Fa9f9dcf79b06b673b6751c0ed027f069%2Fpic2.png?generation=1685292799492538&alt=media)",
    "2279700": "That’s a very interesting analysis. I think instead of taking the top K most represented GO term classes among all samples, it’s smarter to consider top K/3 terms of each of the 3 generic function classes."
  },
  "source": "meta"
}