{
  "id": 220197,
  "title": "Code Base Sharing",
  "url": "/competitions/cassava-leaf-disease-classification/discussion/220197",
  "author_name": "",
  "post_date": "2021-02-17T17:20:26.505001300Z",
  "votes": 4,
  "comment_count": 1,
  "views": 0,
  "content": "<p>We are happy to share a code base built on Haven-AI that makes it easy to add and compare between different models, optimizers and data augmentations for the dataset. The code base can be found at <a href=\"https://github.com/Xinhong-Deng/Cassava-Classification\" target=\"_blank\">https://github.com/Xinhong-Deng/Cassava-Classification</a> and Haven-AI can be found at <a href=\"https://github.com/haven-ai/haven-ai\" target=\"_blank\">https://github.com/haven-ai/haven-ai</a>.</p>\n<p>So far we have compared between resnet, resnext, efficientnet and ViT models. We found out that resnext with SAM optimizer performed the best, which was 0.87, and resnet with adam performed the worst. The validation score is close to our LB score. </p>\n<p>The code base also makes visualizations that tells us where the model is getting confused. We are happy to integrate the winning algorithms once they get published after the test phase. This code base could serve as a useful reference for other similar competitions.</p>",
  "messages": [
    {
      "id": "1207014",
      "postDate": "02/17/2021 17:20:26",
      "content": "<p>We are happy to share a code base built on Haven-AI that makes it easy to add and compare between different models, optimizers and data augmentations for the dataset. The code base can be found at <a href=\"https://github.com/Xinhong-Deng/Cassava-Classification\" target=\"_blank\">https://github.com/Xinhong-Deng/Cassava-Classification</a> and Haven-AI can be found at <a href=\"https://github.com/haven-ai/haven-ai\" target=\"_blank\">https://github.com/haven-ai/haven-ai</a>.</p>\n<p>So far we have compared between resnet, resnext, efficientnet and ViT models. We found out that resnext with SAM optimizer performed the best, which was 0.87, and resnet with adam performed the worst. The validation score is close to our LB score. </p>\n<p>The code base also makes visualizations that tells us where the model is getting confused. We are happy to integrate the winning algorithms once they get published after the test phase. This code base could serve as a useful reference for other similar competitions.</p>",
      "rawMarkdown": "We are happy to share a code base built on Haven-AI that makes it easy to add and compare between different models, optimizers and data augmentations for the dataset. The code base can be found at https://github.com/Xinhong-Deng/Cassava-Classification and Haven-AI can be found at https://github.com/haven-ai/haven-ai.\n\nSo far we have compared between resnet, resnext, efficientnet and ViT models. We found out that resnext with SAM optimizer performed the best, which was 0.87, and resnet with adam performed the worst. The validation score is close to our LB score. \n\nThe code base also makes visualizations that tells us where the model is getting confused. We are happy to integrate the winning algorithms once they get published after the test phase. This code base could serve as a useful reference for other similar competitions.",
      "votes": null
    },
    {
      "id": "1208215",
      "postDate": "02/18/2021 07:57:41",
      "content": "<p>Hey guys nice work. I will try this out in another competition as this is ending today.</p>",
      "rawMarkdown": "Hey guys nice work. I will try this out in another competition as this is ending today.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1208215,
      "author_name": "mohneesh7",
      "author_url": "",
      "post_date": "02/18/2021 07:57:41",
      "content": "<p>Hey guys nice work. I will try this out in another competition as this is ending today.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1207014": "We are happy to share a code base built on Haven-AI that makes it easy to add and compare between different models, optimizers and data augmentations for the dataset. The code base can be found at https://github.com/Xinhong-Deng/Cassava-Classification and Haven-AI can be found at https://github.com/haven-ai/haven-ai.\n\nSo far we have compared between resnet, resnext, efficientnet and ViT models. We found out that resnext with SAM optimizer performed the best, which was 0.87, and resnet with adam performed the worst. The validation score is close to our LB score. \n\nThe code base also makes visualizations that tells us where the model is getting confused. We are happy to integrate the winning algorithms once they get published after the test phase. This code base could serve as a useful reference for other similar competitions.",
    "1208215": "Hey guys nice work. I will try this out in another competition as this is ending today."
  },
  "source": "meta"
}