{
  "id": 105195,
  "title": "Vanilla(Shallow) model performance",
  "url": "/competitions/aptos2019-blindness-detection/discussion/105195",
  "author_name": "",
  "post_date": "2019-08-21T17:39:51.123394300Z",
  "votes": 3,
  "comment_count": 1,
  "views": 0,
  "content": "<p>If you read the top solutions from last competitions: <a href=\"https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15801#latest-370950\">1st place</a>, <a href=\"https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15617#latest-600495\">2nd place</a>\nThey all used shallow networks and trained from scratch. I was wondering if anyone has good performance with similar architecture(and training from scratch). For me, it did not work well :( \nThe best I get is <code>LB: 0.731</code></p>",
  "messages": [
    {
      "id": "604727",
      "postDate": "08/21/2019 17:39:51",
      "content": "<p>If you read the top solutions from last competitions: <a href=\"https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15801#latest-370950\">1st place</a>, <a href=\"https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15617#latest-600495\">2nd place</a>\nThey all used shallow networks and trained from scratch. I was wondering if anyone has good performance with similar architecture(and training from scratch). For me, it did not work well :( \nThe best I get is <code>LB: 0.731</code></p>",
      "rawMarkdown": "If you read the top solutions from last competitions: [1st place](https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15801#latest-370950), [2nd place](https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15617#latest-600495)\nThey all used shallow networks and trained from scratch. I was wondering if anyone has good performance with similar architecture(and training from scratch). For me, it did not work well :( \nThe best I get is `LB: 0.731`",
      "votes": null
    },
    {
      "id": "607633",
      "postDate": "08/25/2019 16:16:42",
      "content": "<p>thats because many deep learning models weren't available then, especially for consumers. Now we (used to have) nice kaggle kernels with good compute and consumer graphics cards can also run SOTA models</p>",
      "rawMarkdown": "thats because many deep learning models weren't available then, especially for consumers. Now we (used to have) nice kaggle kernels with good compute and consumer graphics cards can also run SOTA models",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 607633,
      "author_name": "sidhanthholalkere",
      "author_url": "",
      "post_date": "08/25/2019 16:16:42",
      "content": "<p>thats because many deep learning models weren't available then, especially for consumers. Now we (used to have) nice kaggle kernels with good compute and consumer graphics cards can also run SOTA models</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "604727": "If you read the top solutions from last competitions: [1st place](https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15801#latest-370950), [2nd place](https://www.kaggle.com/c/diabetic-retinopathy-detection/discussion/15617#latest-600495)\nThey all used shallow networks and trained from scratch. I was wondering if anyone has good performance with similar architecture(and training from scratch). For me, it did not work well :( \nThe best I get is `LB: 0.731`",
    "607633": "thats because many deep learning models weren't available then, especially for consumers. Now we (used to have) nice kaggle kernels with good compute and consumer graphics cards can also run SOTA models"
  },
  "source": "meta"
}