{
  "id": 296171,
  "title": "Anyone encountered mAP&F2-score drop problem?",
  "url": "/competitions/tensorflow-great-barrier-reef/discussion/296171",
  "author_name": "",
  "post_date": "2021-12-20T07:55:35.313274900Z",
  "votes": 3,
  "comment_count": 9,
  "views": 0,
  "content": "<p>My model drops after only 7 or 8 epochs, does anyone has the same problem?<br>\nI used the dataset split from <a href=\"url\" target=\"_blank\">https://www.kaggle.com/julian3833/reef-a-cv-strategy-subsequences</a></p>",
  "messages": [
    {
      "id": "1623752",
      "postDate": "12/20/2021 07:55:35",
      "content": "<p>My model drops after only 7 or 8 epochs, does anyone has the same problem?<br>\nI used the dataset split from <a href=\"url\" target=\"_blank\">https://www.kaggle.com/julian3833/reef-a-cv-strategy-subsequences</a></p>",
      "rawMarkdown": "My model drops after only 7 or 8 epochs, does anyone has the same problem?\nI used the dataset split from [https://www.kaggle.com/julian3833/reef-a-cv-strategy-subsequences](url)",
      "votes": null
    },
    {
      "id": "1623754",
      "postDate": "12/20/2021 07:59:22",
      "content": "<p>We have the same …. when we use CV5. What CV strategy do you use?</p>",
      "rawMarkdown": "We have the same .... when we use CV5. What CV strategy do you use?",
      "votes": null
    },
    {
      "id": "1623759",
      "postDate": "12/20/2021 08:05:51",
      "content": "<p>I used train0.8 val0.2</p>",
      "rawMarkdown": "I used train0.8 val0.2",
      "votes": null
    },
    {
      "id": "1623762",
      "postDate": "12/20/2021 08:07:05",
      "content": "<p>We used 0.8/0.2 and CV5 files …. the same effect. Looking for solution.</p>",
      "rawMarkdown": "We used 0.8/0.2 and CV5 files .... the same effect. Looking for solution.",
      "votes": null
    },
    {
      "id": "1624240",
      "postDate": "12/20/2021 16:59:40",
      "content": "<p>does cv5 mean u r dividing whole data into 5 parts? 4 part for training and 1 for validation? kind of new here so.</p>",
      "rawMarkdown": "does cv5 mean u r dividing whole data into 5 parts? 4 part for training and 1 for validation? kind of new here so.",
      "votes": null
    },
    {
      "id": "1624803",
      "postDate": "12/21/2021 08:48:36",
      "content": "<p>Yes my model drops after 9-10 epochs. batch-size 4, 5 fold groupkfold</p>",
      "rawMarkdown": "Yes my model drops after 9-10 epochs. batch-size 4, 5 fold groupkfold",
      "votes": null
    },
    {
      "id": "1626526",
      "postDate": "12/22/2021 23:48:16",
      "content": "<p>Same problem here. The score recovers back up after the drop tho</p>",
      "rawMarkdown": "Same problem here. The score recovers back up after the drop tho",
      "votes": null
    },
    {
      "id": "1626730",
      "postDate": "12/23/2021 06:53:55",
      "content": "<p>Yes. Right.</p>",
      "rawMarkdown": "Yes. Right.",
      "votes": null
    },
    {
      "id": "1639236",
      "postDate": "01/05/2022 14:02:20",
      "content": "<p><a href=\"https://www.kaggle.com/ptran1203\" target=\"_blank\">@ptran1203</a> because you are now in the top 50, how many epochs do you train your model now?</p>",
      "rawMarkdown": "ptran1203 because you are now in the top 50, how many epochs do you train your model now?",
      "votes": null
    },
    {
      "id": "1639908",
      "postDate": "01/06/2022 03:22:02",
      "content": "<p>The same with you. Do you random split the dataset now?</p>",
      "rawMarkdown": "The same with you. Do you random split the dataset now?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1623754,
      "author_name": "remekkinas",
      "author_url": "",
      "post_date": "12/20/2021 07:59:22",
      "content": "<p>We have the same …. when we use CV5. What CV strategy do you use?</p>",
      "votes": null,
      "replies": [
        {
          "id": 1623759,
          "author_name": "klawensliu",
          "author_url": "",
          "post_date": "12/20/2021 08:05:51",
          "content": "<p>I used train0.8 val0.2</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1623762,
          "author_name": "remekkinas",
          "author_url": "",
          "post_date": "12/20/2021 08:07:05",
          "content": "<p>We used 0.8/0.2 and CV5 files …. the same effect. Looking for solution.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1624240,
          "author_name": "keshhere",
          "author_url": "",
          "post_date": "12/20/2021 16:59:40",
          "content": "<p>does cv5 mean u r dividing whole data into 5 parts? 4 part for training and 1 for validation? kind of new here so.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1626730,
          "author_name": "remekkinas",
          "author_url": "",
          "post_date": "12/23/2021 06:53:55",
          "content": "<p>Yes. Right.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1624803,
      "author_name": "ptran1203",
      "author_url": "",
      "post_date": "12/21/2021 08:48:36",
      "content": "<p>Yes my model drops after 9-10 epochs. batch-size 4, 5 fold groupkfold</p>",
      "votes": null,
      "replies": [
        {
          "id": 1639236,
          "author_name": "meesz9",
          "author_url": "",
          "post_date": "01/05/2022 14:02:20",
          "content": "<p><a href=\"https://www.kaggle.com/ptran1203\" target=\"_blank\">@ptran1203</a> because you are now in the top 50, how many epochs do you train your model now?</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1626526,
      "author_name": "joon9502",
      "author_url": "",
      "post_date": "12/22/2021 23:48:16",
      "content": "<p>Same problem here. The score recovers back up after the drop tho</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1639908,
      "author_name": "jarviskevin",
      "author_url": "",
      "post_date": "01/06/2022 03:22:02",
      "content": "<p>The same with you. Do you random split the dataset now?</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1623752": "My model drops after only 7 or 8 epochs, does anyone has the same problem?\nI used the dataset split from [https://www.kaggle.com/julian3833/reef-a-cv-strategy-subsequences](url)",
    "1623754": "We have the same .... when we use CV5. What CV strategy do you use?",
    "1623759": "I used train0.8 val0.2",
    "1623762": "We used 0.8/0.2 and CV5 files .... the same effect. Looking for solution.",
    "1624240": "does cv5 mean u r dividing whole data into 5 parts? 4 part for training and 1 for validation? kind of new here so.",
    "1624803": "Yes my model drops after 9-10 epochs. batch-size 4, 5 fold groupkfold",
    "1626526": "Same problem here. The score recovers back up after the drop tho",
    "1626730": "Yes. Right.",
    "1639236": "ptran1203 because you are now in the top 50, how many epochs do you train your model now?",
    "1639908": "The same with you. Do you random split the dataset now?"
  },
  "source": "meta"
}