{
  "id": 223368,
  "title": "EfficientNet7 Inference Notebook in Tensorflow[LB-0.956]",
  "url": "/competitions/ranzcr-clip-catheter-line-classification/discussion/223368",
  "author_name": "",
  "post_date": "2021-03-03T14:45:48.010501300Z",
  "votes": 5,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I have created a inference notebook which gives me a LB of 95.6 on EfficientNet7 model.<br>\nThe notebook primarily focuses on 2 issues faced as how to make submission on private dataset along with public dataset without using sample_submission file(which only have public dataset information) and how to use EfficientNet model without internet.</p>\n<p>In the Inference  No TTA is used. Image size of 600 used for training on TPU and same size is used for inference(GPU). Light Augmentations are used for training the model. </p>\n<p>Running <a href=\"https://www.kaggle.com/vickygoyal/eff7inference\" target=\"_blank\">this</a> inference notebook without editing gives LB of 95.6</p>",
  "messages": [
    {
      "id": "1225329",
      "postDate": "03/03/2021 14:45:48",
      "content": "<p>I have created a inference notebook which gives me a LB of 95.6 on EfficientNet7 model.<br>\nThe notebook primarily focuses on 2 issues faced as how to make submission on private dataset along with public dataset without using sample_submission file(which only have public dataset information) and how to use EfficientNet model without internet.</p>\n<p>In the Inference  No TTA is used. Image size of 600 used for training on TPU and same size is used for inference(GPU). Light Augmentations are used for training the model. </p>\n<p>Running <a href=\"https://www.kaggle.com/vickygoyal/eff7inference\" target=\"_blank\">this</a> inference notebook without editing gives LB of 95.6</p>",
      "rawMarkdown": "I have created a inference notebook which gives me a LB of 95.6 on EfficientNet7 model.\nThe notebook primarily focuses on 2 issues faced as how to make submission on private dataset along with public dataset without using sample_submission file(which only have public dataset information) and how to use EfficientNet model without internet.\n\nIn the Inference  No TTA is used. Image size of 600 used for training on TPU and same size is used for inference(GPU). Light Augmentations are used for training the model. \n\nRunning [this](https://www.kaggle.com/vickygoyal/eff7inference) inference notebook without editing gives LB of 95.6",
      "votes": null
    },
    {
      "id": "1225477",
      "postDate": "03/03/2021 16:22:35",
      "content": "<p>Great work. </p>\n<p>Note when you <strong>commit</strong> your notebook then <code>sample_submission.csv</code> only lists public test. But when you <strong>submit</strong> your notebook, it will load a different <code>sample_submission.csv</code> which contains both public and private test.</p>",
      "rawMarkdown": "Great work. \n\nNote when you **commit** your notebook then `sample_submission.csv` only lists public test. But when you **submit** your notebook, it will load a different `sample_submission.csv` which contains both public and private test.",
      "votes": null
    },
    {
      "id": "1225499",
      "postDate": "03/03/2021 16:37:55",
      "content": "<p>Thanks a lot Chris</p>",
      "rawMarkdown": "Thanks a lot Chris",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1225477,
      "author_name": "cdeotte",
      "author_url": "",
      "post_date": "03/03/2021 16:22:35",
      "content": "<p>Great work. </p>\n<p>Note when you <strong>commit</strong> your notebook then <code>sample_submission.csv</code> only lists public test. But when you <strong>submit</strong> your notebook, it will load a different <code>sample_submission.csv</code> which contains both public and private test.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1225499,
          "author_name": "vickygoyal",
          "author_url": "",
          "post_date": "03/03/2021 16:37:55",
          "content": "<p>Thanks a lot Chris</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1225329": "I have created a inference notebook which gives me a LB of 95.6 on EfficientNet7 model.\nThe notebook primarily focuses on 2 issues faced as how to make submission on private dataset along with public dataset without using sample_submission file(which only have public dataset information) and how to use EfficientNet model without internet.\n\nIn the Inference  No TTA is used. Image size of 600 used for training on TPU and same size is used for inference(GPU). Light Augmentations are used for training the model. \n\nRunning [this](https://www.kaggle.com/vickygoyal/eff7inference) inference notebook without editing gives LB of 95.6",
    "1225477": "Great work. \n\nNote when you **commit** your notebook then `sample_submission.csv` only lists public test. But when you **submit** your notebook, it will load a different `sample_submission.csv` which contains both public and private test.",
    "1225499": "Thanks a lot Chris"
  },
  "source": "meta"
}