{
  "id": 441122,
  "title": "Train data reactivity columns are null?",
  "url": "/competitions/stanford-ribonanza-rna-folding/discussion/441122",
  "author_name": "Deniz Can Elci",
  "post_date": "2023-09-17T16:45:46.839000",
  "votes": 1,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Hi all,<br>\nWhen I check the data I see Train data reactivite columns are all null(%100),<br>\nIs it normal, should they be with values so that we can train our model(since they are outcome features)?<br>\nBest regards.</p>",
  "messages": [
    {
      "id": 2443959,
      "postDate": "2023-09-18T05:24:23.237Z",
      "content": "<p>Thanks for asking the question!</p>\n<p>There are indeed some columns that are all null in <code>train_data.csv</code>.</p>\n<p>It is because our current experimental method cannot capture the reactivity at the beginning and end of each RNA sequence. </p>\n<p>However, we may be able to make measurements that capture beginning and end of the RNA for some of the test sequences. </p>\n<p>So you might still try to make predictions for the corresponding locations in the test sequences. Your model should hopefully be able to generalize from training on internal positions to these more exterior positions in the sequence.</p>",
      "rawMarkdown": "Thanks for asking the question!\n\nThere are indeed some columns that are all null in `train_data.csv`.\n\nIt is because our current experimental method cannot capture the reactivity at the beginning and end of each RNA sequence. \n\nHowever, we may be able to make measurements that capture beginning and end of the RNA for some of the test sequences. \n\nSo you might still try to make predictions for the corresponding locations in the test sequences. Your model should hopefully be able to generalize from training on internal positions to these more exterior positions in the sequence.",
      "votes": 3,
      "replies": [
        {
          "id": 2488003,
          "postDate": "2023-10-19T00:20:20.437Z",
          "rawMarkdown": "",
          "isDeleted": true
        }
      ]
    },
    {
      "id": 2443321,
      "postDate": "2023-09-17T16:45:46.840Z",
      "content": "<p>Hi all,<br>\nWhen I check the data I see Train data reactivite columns are all null(%100),<br>\nIs it normal, should they be with values so that we can train our model(since they are outcome features)?<br>\nBest regards.</p>",
      "rawMarkdown": "Hi all,\nWhen I check the data I see Train data reactivite columns are all null(%100),\nIs it normal, should they be with values so that we can train our model(since they are outcome features)?\nBest regards.",
      "votes": 1
    },
    {
      "id": 2488005,
      "postDate": "2023-10-19T00:20:48.927Z",
      "content": "<p>Following up on this, do we get penalized if we predict a reactivity value for a sequence position towards the beginning or end, but your experimental method gives a null value for that position? How is this situation treated?</p>",
      "rawMarkdown": "Following up on this, do we get penalized if we predict a reactivity value for a sequence position towards the beginning or end, but your experimental method gives a null value for that position? How is this situation treated?",
      "replies": [
        {
          "id": 2489094,
          "postDate": "2023-10-19T17:16:55Z",
          "content": "<p>No penalty in that case. That is, for any sequences or sequences positions where we can't get data, we will not penalize the model score. </p>",
          "rawMarkdown": "No penalty in that case. That is, for any sequences or sequences positions where we can't get data, we will not penalize the model score. "
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 2443959,
      "author_name": "Rhiju Das",
      "author_url": "",
      "post_date": "2023-09-18T05:24:23.237000",
      "content": "<p>Thanks for asking the question!</p>\n<p>There are indeed some columns that are all null in <code>train_data.csv</code>.</p>\n<p>It is because our current experimental method cannot capture the reactivity at the beginning and end of each RNA sequence. </p>\n<p>However, we may be able to make measurements that capture beginning and end of the RNA for some of the test sequences. </p>\n<p>So you might still try to make predictions for the corresponding locations in the test sequences. Your model should hopefully be able to generalize from training on internal positions to these more exterior positions in the sequence.</p>",
      "votes": 3,
      "replies": [
        {
          "id": 2488003,
          "author_name": "",
          "author_url": "",
          "post_date": "2023-10-19T00:20:20.437000",
          "content": "",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 2488005,
      "author_name": "ammobeast",
      "author_url": "",
      "post_date": "2023-10-19T00:20:48.927000",
      "content": "<p>Following up on this, do we get penalized if we predict a reactivity value for a sequence position towards the beginning or end, but your experimental method gives a null value for that position? How is this situation treated?</p>",
      "votes": 0,
      "replies": [
        {
          "id": 2489094,
          "author_name": "Rhiju Das",
          "author_url": "",
          "post_date": "2023-10-19T17:16:55",
          "content": "<p>No penalty in that case. That is, for any sequences or sequences positions where we can't get data, we will not penalize the model score. </p>",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2443959": "Thanks for asking the question!\n\nThere are indeed some columns that are all null in `train_data.csv`.\n\nIt is because our current experimental method cannot capture the reactivity at the beginning and end of each RNA sequence. \n\nHowever, we may be able to make measurements that capture beginning and end of the RNA for some of the test sequences. \n\nSo you might still try to make predictions for the corresponding locations in the test sequences. Your model should hopefully be able to generalize from training on internal positions to these more exterior positions in the sequence.",
    "2443321": "Hi all,\nWhen I check the data I see Train data reactivite columns are all null(%100),\nIs it normal, should they be with values so that we can train our model(since they are outcome features)?\nBest regards.",
    "2488005": "Following up on this, do we get penalized if we predict a reactivity value for a sequence position towards the beginning or end, but your experimental method gives a null value for that position? How is this situation treated?"
  }
}