{
  "id": 518583,
  "title": "Late but intereasting thing about sEH positives",
  "url": "/competitions/leash-BELKA/discussion/518583",
  "author_name": "",
  "post_date": "2024-07-07T09:04:01.964685900Z",
  "votes": 5,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hi. I was dividing data by bulding block 1 just for a last experiment and I've found something intereasting about sEH positives that was unbalancing my random selection:<br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8722753%2Ffde836be63de53b261ea17295704f346%2FSin%20ttulo.png?generation=1720343008355231&amp;alt=media\"></p>\n<p>I suppose most of people knew this already. But just in case. Good luck.</p>",
  "messages": [
    {
      "id": "2909734",
      "postDate": "07/07/2024 09:04:01",
      "content": "<p>Hi. I was dividing data by bulding block 1 just for a last experiment and I've found something intereasting about sEH positives that was unbalancing my random selection:<br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8722753%2Ffde836be63de53b261ea17295704f346%2FSin%20ttulo.png?generation=1720343008355231&amp;alt=media\"></p>\n<p>I suppose most of people knew this already. But just in case. Good luck.</p>",
      "rawMarkdown": "Hi. I was dividing data by bulding block 1 just for a last experiment and I've found something intereasting about sEH positives that was unbalancing my random selection:\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8722753%2Ffde836be63de53b261ea17295704f346%2FSin%20ttulo.png?generation=1720343008355231&alt=media)\n\nI suppose most of people knew this already. But just in case. Good luck.",
      "votes": null
    },
    {
      "id": "2911190",
      "postDate": "07/08/2024 06:17:29",
      "content": "<p>The submission test set also includes new blocks not in the training set.  </p>",
      "rawMarkdown": "The submission test set also includes new blocks not in the training set.",
      "votes": null
    },
    {
      "id": "2911579",
      "postDate": "07/08/2024 12:02:34",
      "content": "<p>The responsables are BB 75 and 76. After some last minut considerations I've decided to reduce the CV to 4 and use BB 75, 76 and both half of the remaining. I'm waiting till the lasts folds finish of training but I've noticed validation on 75 folds are just terrible (everything binds) so I've decided to use only folds where 75 have been included to predict those on test. I think 76 will have similar results. The thing is that probably would been better just make different CV models for those BB, dividing then by second BB.</p>",
      "rawMarkdown": "The responsables are BB 75 and 76. After some last minut considerations I've decided to reduce the CV to 4 and use BB 75, 76 and both half of the remaining. I'm waiting till the lasts folds finish of training but I've noticed validation on 75 folds are just terrible (everything binds) so I've decided to use only folds where 75 have been included to predict those on test. I think 76 will have similar results. The thing is that probably would been better just make different CV models for those BB, dividing then by second BB.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2911190,
      "author_name": "joejeo1",
      "author_url": "",
      "post_date": "07/08/2024 06:17:29",
      "content": "<p>The submission test set also includes new blocks not in the training set.  </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2911579,
      "author_name": "sacuscreed",
      "author_url": "",
      "post_date": "07/08/2024 12:02:34",
      "content": "<p>The responsables are BB 75 and 76. After some last minut considerations I've decided to reduce the CV to 4 and use BB 75, 76 and both half of the remaining. I'm waiting till the lasts folds finish of training but I've noticed validation on 75 folds are just terrible (everything binds) so I've decided to use only folds where 75 have been included to predict those on test. I think 76 will have similar results. The thing is that probably would been better just make different CV models for those BB, dividing then by second BB.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2909734": "Hi. I was dividing data by bulding block 1 just for a last experiment and I've found something intereasting about sEH positives that was unbalancing my random selection:\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8722753%2Ffde836be63de53b261ea17295704f346%2FSin%20ttulo.png?generation=1720343008355231&alt=media)\n\nI suppose most of people knew this already. But just in case. Good luck.",
    "2911190": "The submission test set also includes new blocks not in the training set.",
    "2911579": "The responsables are BB 75 and 76. After some last minut considerations I've decided to reduce the CV to 4 and use BB 75, 76 and both half of the remaining. I'm waiting till the lasts folds finish of training but I've noticed validation on 75 folds are just terrible (everything binds) so I've decided to use only folds where 75 have been included to predict those on test. I think 76 will have similar results. The thing is that probably would been better just make different CV models for those BB, dividing then by second BB."
  },
  "source": "meta"
}