{
  "id": 175504,
  "title": "1st Place Solution Code (Small ver.) Released!",
  "url": "/competitions/siim-isic-melanoma-classification/discussion/175504",
  "author_name": "",
  "post_date": "2020-08-18T11:38:36.186082300Z",
  "votes": 50,
  "comment_count": 16,
  "views": 0,
  "content": "<p>Notebook: <a href=\"https://www.kaggle.com/haqishen/1st-place-soluiton-code-small-ver\" target=\"_blank\">https://www.kaggle.com/haqishen/1st-place-soluiton-code-small-ver</a> <br>\n(Do remember to use bigger input data when you set bigger <code>image_size</code> 🤓)</p>\n<p>Summary: <a href=\"https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/175412\" target=\"_blank\">https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/175412</a></p>",
  "messages": [
    {
      "id": "975620",
      "postDate": "08/18/2020 11:38:36",
      "content": "<p>Notebook: <a href=\"https://www.kaggle.com/haqishen/1st-place-soluiton-code-small-ver\" target=\"_blank\">https://www.kaggle.com/haqishen/1st-place-soluiton-code-small-ver</a> <br>\n(Do remember to use bigger input data when you set bigger <code>image_size</code> 🤓)</p>\n<p>Summary: <a href=\"https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/175412\" target=\"_blank\">https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/175412</a></p>",
      "rawMarkdown": "Notebook: https://www.kaggle.com/haqishen/1st-place-soluiton-code-small-ver \n(Do remember to use bigger input data when you set bigger `image_size` 🤓)\n\nSummary: https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/175412",
      "votes": null
    },
    {
      "id": "975661",
      "postDate": "08/18/2020 12:07:11",
      "content": "<p>Wow so elegant, thanks for sharing and congrads🎉🎉🎉 </p>",
      "rawMarkdown": "Wow so elegant, thanks for sharing and congrads🎉🎉🎉",
      "votes": null
    },
    {
      "id": "975685",
      "postDate": "08/18/2020 12:23:45",
      "content": "<p>Thanks for your sharing! <br>\nMay I know the reason you use Apex instead of pytorch inbuild torch.cuda.amp.autocast?<br>\nIs Apex better, according to your experience?</p>\n<p>thanks, congrats again for the 1st place !</p>",
      "rawMarkdown": "Thanks for your sharing! \nMay I know the reason you use Apex instead of pytorch inbuild torch.cuda.amp.autocast?\nIs Apex better, according to your experience?\n\nthanks, congrats again for the 1st place !",
      "votes": null
    },
    {
      "id": "975690",
      "postDate": "08/18/2020 12:26:22",
      "content": "<p>I bet it is because he uses pytorch 1.5.  amp was only available in apex for that version.  Also, pytorch 1.6 with amp only cam in very recently, switching pytorch version near the end of a competition isn't necessarily a great idea.</p>",
      "rawMarkdown": "I bet it is because he uses pytorch 1.5.  amp was only available in apex for that version.  Also, pytorch 1.6 with amp only cam in very recently, switching pytorch version near the end of a competition isn't necessarily a great idea.",
      "votes": null
    },
    {
      "id": "975707",
      "postDate": "08/18/2020 12:34:24",
      "content": "<p>Yes, I didn't use pytorch 1.6. Actually I personally have worked on pytorch 1.2 with amp for a long time (maybe a year) …</p>",
      "rawMarkdown": "Yes, I didn't use pytorch 1.6. Actually I personally have worked on pytorch 1.2 with amp for a long time (maybe a year) ...",
      "votes": null
    },
    {
      "id": "975712",
      "postDate": "08/18/2020 12:38:07",
      "content": "<p>Thanks 🙏!</p>",
      "rawMarkdown": "Thanks 🙏!",
      "votes": null
    },
    {
      "id": "975801",
      "postDate": "08/18/2020 13:24:55",
      "content": "<p>Thanks a lot for sharing that amazing notebook 😍</p>",
      "rawMarkdown": "Thanks a lot for sharing that amazing notebook 😍",
      "votes": null
    },
    {
      "id": "975899",
      "postDate": "08/18/2020 14:12:31",
      "content": "<p>Learn alot from detail solution (notebook and discussion), thanks for sharing <a href=\"https://www.kaggle.com/haqishen\" target=\"_blank\">@haqishen</a> and team</p>",
      "rawMarkdown": "Learn alot from detail solution (notebook and discussion), thanks for sharing @haqishen and team",
      "votes": null
    },
    {
      "id": "975950",
      "postDate": "08/18/2020 14:47:00",
      "content": "<p>Awesome Work</p>",
      "rawMarkdown": "Awesome Work",
      "votes": null
    },
    {
      "id": "976572",
      "postDate": "08/19/2020 00:36:10",
      "content": "<p>Congratulations Qishen, Bo, Gary. Thanks for sharing your code.</p>",
      "rawMarkdown": "Congratulations Qishen, Bo, Gary. Thanks for sharing your code.",
      "votes": null
    },
    {
      "id": "976621",
      "postDate": "08/19/2020 01:43:40",
      "content": "<p>Thanks for sharing!</p>",
      "rawMarkdown": "Thanks for sharing!",
      "votes": null
    },
    {
      "id": "976724",
      "postDate": "08/19/2020 04:08:33",
      "content": "<p>Thanks for sharing the code it helps!</p>",
      "rawMarkdown": "Thanks for sharing the code it helps!",
      "votes": null
    },
    {
      "id": "976746",
      "postDate": "08/19/2020 04:26:34",
      "content": "<p>Thank you very much! so amazing！</p>",
      "rawMarkdown": "Thank you very much! so amazing！",
      "votes": null
    },
    {
      "id": "977454",
      "postDate": "08/19/2020 13:40:54",
      "content": "<p>Thanks for sharing the code and congrats on the 1st place finish. I'm surprised to see how strong the augmentations look visually:</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F644036%2Fb97f565b5d6414252382712addf315e8%2FScreen%20Shot%202020-08-19%20at%209.36.47%20AM.png?generation=1597844238105034&amp;alt=media\" alt=\"\"></p>\n<p>Were there any specific augmentations you found added the most benefit? Did such augmentations significantly slow down the training time?</p>",
      "rawMarkdown": "Thanks for sharing the code and congrats on the 1st place finish. I'm surprised to see how strong the augmentations look visually:\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F644036%2Fb97f565b5d6414252382712addf315e8%2FScreen%20Shot%202020-08-19%20at%209.36.47%20AM.png?generation=1597844238105034&alt=media)\n\nWere there any specific augmentations you found added the most benefit? Did such augmentations significantly slow down the training time?",
      "votes": null
    },
    {
      "id": "977575",
      "postDate": "08/19/2020 15:09:32",
      "content": "<p>Congratulations!</p>",
      "rawMarkdown": "Congratulations!",
      "votes": null
    },
    {
      "id": "977577",
      "postDate": "08/19/2020 15:10:45",
      "content": "<p>Every augmentation methods slow down the training…<br>\nThe most powerful one in this dataset is distortions.</p>",
      "rawMarkdown": "Every augmentation methods slow down the training...\nThe most powerful one in this dataset is distortions.",
      "votes": null
    },
    {
      "id": "981205",
      "postDate": "08/22/2020 09:08:48",
      "content": "<p>Thanks for sharing the code!</p>",
      "rawMarkdown": "Thanks for sharing the code!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 975661,
      "author_name": "ivanvoid",
      "author_url": "",
      "post_date": "08/18/2020 12:07:11",
      "content": "<p>Wow so elegant, thanks for sharing and congrads🎉🎉🎉 </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 975685,
      "author_name": "fiyeroleung",
      "author_url": "",
      "post_date": "08/18/2020 12:23:45",
      "content": "<p>Thanks for your sharing! <br>\nMay I know the reason you use Apex instead of pytorch inbuild torch.cuda.amp.autocast?<br>\nIs Apex better, according to your experience?</p>\n<p>thanks, congrats again for the 1st place !</p>",
      "votes": null,
      "replies": [
        {
          "id": 975690,
          "author_name": "cpmpml",
          "author_url": "",
          "post_date": "08/18/2020 12:26:22",
          "content": "<p>I bet it is because he uses pytorch 1.5.  amp was only available in apex for that version.  Also, pytorch 1.6 with amp only cam in very recently, switching pytorch version near the end of a competition isn't necessarily a great idea.</p>",
          "votes": null,
          "replies": [
            {
              "id": 975707,
              "author_name": "haqishen",
              "author_url": "",
              "post_date": "08/18/2020 12:34:24",
              "content": "<p>Yes, I didn't use pytorch 1.6. Actually I personally have worked on pytorch 1.2 with amp for a long time (maybe a year) …</p>",
              "votes": null,
              "replies": []
            },
            {
              "id": 975712,
              "author_name": "fiyeroleung",
              "author_url": "",
              "post_date": "08/18/2020 12:38:07",
              "content": "<p>Thanks 🙏!</p>",
              "votes": null,
              "replies": []
            }
          ]
        }
      ]
    },
    {
      "id": 975801,
      "author_name": "rahim3",
      "author_url": "",
      "post_date": "08/18/2020 13:24:55",
      "content": "<p>Thanks a lot for sharing that amazing notebook 😍</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 975899,
      "author_name": "duykhanh99",
      "author_url": "",
      "post_date": "08/18/2020 14:12:31",
      "content": "<p>Learn alot from detail solution (notebook and discussion), thanks for sharing <a href=\"https://www.kaggle.com/haqishen\" target=\"_blank\">@haqishen</a> and team</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 976572,
      "author_name": "cdeotte",
      "author_url": "",
      "post_date": "08/19/2020 00:36:10",
      "content": "<p>Congratulations Qishen, Bo, Gary. Thanks for sharing your code.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 976621,
      "author_name": "y8acemak3r",
      "author_url": "",
      "post_date": "08/19/2020 01:43:40",
      "content": "<p>Thanks for sharing!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 976724,
      "author_name": "aakashveera",
      "author_url": "",
      "post_date": "08/19/2020 04:08:33",
      "content": "<p>Thanks for sharing the code it helps!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 976746,
      "author_name": "mnk812",
      "author_url": "",
      "post_date": "08/19/2020 04:26:34",
      "content": "<p>Thank you very much! so amazing！</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 977454,
      "author_name": "robikscube",
      "author_url": "",
      "post_date": "08/19/2020 13:40:54",
      "content": "<p>Thanks for sharing the code and congrats on the 1st place finish. I'm surprised to see how strong the augmentations look visually:</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F644036%2Fb97f565b5d6414252382712addf315e8%2FScreen%20Shot%202020-08-19%20at%209.36.47%20AM.png?generation=1597844238105034&amp;alt=media\" alt=\"\"></p>\n<p>Were there any specific augmentations you found added the most benefit? Did such augmentations significantly slow down the training time?</p>",
      "votes": null,
      "replies": [
        {
          "id": 977577,
          "author_name": "haqishen",
          "author_url": "",
          "post_date": "08/19/2020 15:10:45",
          "content": "<p>Every augmentation methods slow down the training…<br>\nThe most powerful one in this dataset is distortions.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 981205,
      "author_name": "dengbowen",
      "author_url": "",
      "post_date": "08/22/2020 09:08:48",
      "content": "<p>Thanks for sharing the code!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 975950,
      "author_name": "raoofnaushad",
      "author_url": "",
      "post_date": "08/18/2020 14:47:00",
      "content": "<p>Awesome Work</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 977575,
      "author_name": "raoofnaushad",
      "author_url": "",
      "post_date": "08/19/2020 15:09:32",
      "content": "<p>Congratulations!</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "975620": "Notebook: https://www.kaggle.com/haqishen/1st-place-soluiton-code-small-ver \n(Do remember to use bigger input data when you set bigger `image_size` 🤓)\n\nSummary: https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/175412",
    "975661": "Wow so elegant, thanks for sharing and congrads🎉🎉🎉",
    "975685": "Thanks for your sharing! \nMay I know the reason you use Apex instead of pytorch inbuild torch.cuda.amp.autocast?\nIs Apex better, according to your experience?\n\nthanks, congrats again for the 1st place !",
    "975690": "I bet it is because he uses pytorch 1.5.  amp was only available in apex for that version.  Also, pytorch 1.6 with amp only cam in very recently, switching pytorch version near the end of a competition isn't necessarily a great idea.",
    "975707": "Yes, I didn't use pytorch 1.6. Actually I personally have worked on pytorch 1.2 with amp for a long time (maybe a year) ...",
    "975712": "Thanks 🙏!",
    "975801": "Thanks a lot for sharing that amazing notebook 😍",
    "975899": "Learn alot from detail solution (notebook and discussion), thanks for sharing @haqishen and team",
    "975950": "Awesome Work",
    "976572": "Congratulations Qishen, Bo, Gary. Thanks for sharing your code.",
    "976621": "Thanks for sharing!",
    "976724": "Thanks for sharing the code it helps!",
    "976746": "Thank you very much! so amazing！",
    "977454": "Thanks for sharing the code and congrats on the 1st place finish. I'm surprised to see how strong the augmentations look visually:\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F644036%2Fb97f565b5d6414252382712addf315e8%2FScreen%20Shot%202020-08-19%20at%209.36.47%20AM.png?generation=1597844238105034&alt=media)\n\nWere there any specific augmentations you found added the most benefit? Did such augmentations significantly slow down the training time?",
    "977575": "Congratulations!",
    "977577": "Every augmentation methods slow down the training...\nThe most powerful one in this dataset is distortions.",
    "981205": "Thanks for sharing the code!"
  },
  "source": "meta"
}