{
  "id": 115599,
  "title": "How many epochs are you training?",
  "url": "/competitions/understanding_cloud_organization/discussion/115599",
  "author_name": "",
  "post_date": "2019-11-04T01:47:56.143489100Z",
  "votes": 3,
  "comment_count": 13,
  "views": 0,
  "content": "<p>In my case, it's 20 with Early stopping; training more epochs overfits in my experiments. How about yours?</p>",
  "messages": [
    {
      "id": "664627",
      "postDate": "11/04/2019 01:47:56",
      "content": "<p>In my case, it's 20 with Early stopping; training more epochs overfits in my experiments. How about yours?</p>",
      "rawMarkdown": "In my case, it's 20 with Early stopping; training more epochs overfits in my experiments. How about yours?",
      "votes": null
    },
    {
      "id": "664628",
      "postDate": "11/04/2019 01:53:38",
      "content": "<p>I have more than yours, about 30 </p>",
      "rawMarkdown": "I have more than yours, about 30",
      "votes": null
    },
    {
      "id": "664640",
      "postDate": "11/04/2019 02:16:06",
      "content": "<p>I train my model with 20</p>",
      "rawMarkdown": "I train my model with 20",
      "votes": null
    },
    {
      "id": "664689",
      "postDate": "11/04/2019 03:58:45",
      "content": "<p>In my case, around 25.</p>",
      "rawMarkdown": "In my case, around 25.",
      "votes": null
    },
    {
      "id": "664704",
      "postDate": "11/04/2019 05:04:30",
      "content": "<p>In about 25-30 epochs I can get an idea whether the model is giving decent results or getting overfitted. But in all my experiments after 30 epochs model overfits :(</p>",
      "rawMarkdown": "In about 25-30 epochs I can get an idea whether the model is giving decent results or getting overfitted. But in all my experiments after 30 epochs model overfits :(",
      "votes": null
    },
    {
      "id": "664734",
      "postDate": "11/04/2019 06:31:56",
      "content": "<p>I think around 30 should be fine. But you can save the epoch weight which has the best validation scores. This will allow you to try more epochs.  </p>",
      "rawMarkdown": "I think around 30 should be fine. But you can save the epoch weight which has the best validation scores. This will allow you to try more epochs.",
      "votes": null
    },
    {
      "id": "664736",
      "postDate": "11/04/2019 06:33:12",
      "content": "<p>It depends on model and fold. </p>",
      "rawMarkdown": "It depends on model and fold.",
      "votes": null
    },
    {
      "id": "664739",
      "postDate": "11/04/2019 06:35:36",
      "content": "<p>Oh my, I just notice that you become rank 1.\nThat was fast!</p>",
      "rawMarkdown": "Oh my, I just notice that you become rank 1.\nThat was fast!",
      "votes": null
    },
    {
      "id": "664911",
      "postDate": "11/04/2019 11:55:04",
      "content": "<p>How long is a single epoch taking for everyone?\nMine is ~15 minutes including both training and validation</p>",
      "rawMarkdown": "How long is a single epoch taking for everyone?\nMine is ~15 minutes including both training and validation",
      "votes": null
    },
    {
      "id": "665069",
      "postDate": "11/04/2019 15:48:20",
      "content": "<p>I trained mine with 2 epochs</p>",
      "rawMarkdown": "I trained mine with 2 epochs",
      "votes": null
    },
    {
      "id": "665179",
      "postDate": "11/04/2019 17:39:42",
      "content": "<p>It depends on lots of things. The image size, augmentation, learning rate, network structure and hardware. </p>",
      "rawMarkdown": "It depends on lots of things. The image size, augmentation, learning rate, network structure and hardware.",
      "votes": null
    },
    {
      "id": "665200",
      "postDate": "11/04/2019 18:23:39",
      "content": "<p>resnet unet 100 epochs...</p>",
      "rawMarkdown": "resnet unet 100 epochs...",
      "votes": null
    },
    {
      "id": "665204",
      "postDate": "11/04/2019 18:31:19",
      "content": "<p>wow!! without overfitting? your loss function should be amazing(as always)</p>",
      "rawMarkdown": "wow!! without overfitting? your loss function should be amazing(as always)",
      "votes": null
    },
    {
      "id": "665326",
      "postDate": "11/04/2019 21:58:53",
      "content": "<p>About 40 epochs converge faster than steels</p>",
      "rawMarkdown": "About 40 epochs converge faster than steels",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 664628,
      "author_name": "zhangeng",
      "author_url": "",
      "post_date": "11/04/2019 01:53:38",
      "content": "<p>I have more than yours, about 30 </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 664640,
      "author_name": "gustavo219",
      "author_url": "",
      "post_date": "11/04/2019 02:16:06",
      "content": "<p>I train my model with 20</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 664689,
      "author_name": "xiejialun",
      "author_url": "",
      "post_date": "11/04/2019 03:58:45",
      "content": "<p>In my case, around 25.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 664704,
      "author_name": "axel81",
      "author_url": "",
      "post_date": "11/04/2019 05:04:30",
      "content": "<p>In about 25-30 epochs I can get an idea whether the model is giving decent results or getting overfitted. But in all my experiments after 30 epochs model overfits :(</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 664734,
      "author_name": "mykttu",
      "author_url": "",
      "post_date": "11/04/2019 06:31:56",
      "content": "<p>I think around 30 should be fine. But you can save the epoch weight which has the best validation scores. This will allow you to try more epochs.  </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 664736,
      "author_name": "naivelamb",
      "author_url": "",
      "post_date": "11/04/2019 06:33:12",
      "content": "<p>It depends on model and fold. </p>",
      "votes": null,
      "replies": [
        {
          "id": 664739,
          "author_name": "xiejialun",
          "author_url": "",
          "post_date": "11/04/2019 06:35:36",
          "content": "<p>Oh my, I just notice that you become rank 1.\nThat was fast!</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 664911,
      "author_name": "timetraveller98",
      "author_url": "",
      "post_date": "11/04/2019 11:55:04",
      "content": "<p>How long is a single epoch taking for everyone?\nMine is ~15 minutes including both training and validation</p>",
      "votes": null,
      "replies": [
        {
          "id": 665179,
          "author_name": "naivelamb",
          "author_url": "",
          "post_date": "11/04/2019 17:39:42",
          "content": "<p>It depends on lots of things. The image size, augmentation, learning rate, network structure and hardware. </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 665069,
      "author_name": "polmonroig",
      "author_url": "",
      "post_date": "11/04/2019 15:48:20",
      "content": "<p>I trained mine with 2 epochs</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 665200,
      "author_name": "tugstugi",
      "author_url": "",
      "post_date": "11/04/2019 18:23:39",
      "content": "<p>resnet unet 100 epochs...</p>",
      "votes": null,
      "replies": [
        {
          "id": 665204,
          "author_name": "bibek777",
          "author_url": "",
          "post_date": "11/04/2019 18:31:19",
          "content": "<p>wow!! without overfitting? your loss function should be amazing(as always)</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 665326,
      "author_name": "strideradu",
      "author_url": "",
      "post_date": "11/04/2019 21:58:53",
      "content": "<p>About 40 epochs converge faster than steels</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "664627": "In my case, it's 20 with Early stopping; training more epochs overfits in my experiments. How about yours?",
    "664628": "I have more than yours, about 30",
    "664640": "I train my model with 20",
    "664689": "In my case, around 25.",
    "664704": "In about 25-30 epochs I can get an idea whether the model is giving decent results or getting overfitted. But in all my experiments after 30 epochs model overfits :(",
    "664734": "I think around 30 should be fine. But you can save the epoch weight which has the best validation scores. This will allow you to try more epochs.",
    "664736": "It depends on model and fold.",
    "664739": "Oh my, I just notice that you become rank 1.\nThat was fast!",
    "664911": "How long is a single epoch taking for everyone?\nMine is ~15 minutes including both training and validation",
    "665069": "I trained mine with 2 epochs",
    "665179": "It depends on lots of things. The image size, augmentation, learning rate, network structure and hardware.",
    "665200": "resnet unet 100 epochs...",
    "665204": "wow!! without overfitting? your loss function should be amazing(as always)",
    "665326": "About 40 epochs converge faster than steels"
  },
  "source": "meta"
}