{
  "id": 87894,
  "title": "Besides only predictions, what we can do in kernels now?",
  "url": "/competitions/imet-2019-fgvc6/discussion/87894",
  "author_name": "Dilapsky Lee",
  "post_date": "2019-04-04T09:19:49.242000",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>After we can use our own pretrained models in kaggle kernels (as discussed here <a href=\"https://www.kaggle.com/c/imet-2019-fgvc6/discussion/87544\">https://www.kaggle.com/c/imet-2019-fgvc6/discussion/87544</a> ), we can train locally and only test in kernels. However, we can come up with other tricks in Kernels, especially when dealing with private test data.</p>\n\n<p>I think we can make pseudo-labels for private test set in Kaggle Kernels, re-trained our pre-trained models for several steps on them. But I only come up with this idea (with my poor deep-learning knowledge)</p>\n\n<p>We can also use TTA</p>\n\n<p>Can someone else mentioned other tricks in Kaggle Kernels? Especially for private test set tricks, because this can not accomplished in local.</p>\n\n<p>Thanks!</p>",
  "messages": [
    {
      "id": 507091,
      "postDate": "2019-04-04T09:19:49.243Z",
      "content": "<p>After we can use our own pretrained models in kaggle kernels (as discussed here <a href=\"https://www.kaggle.com/c/imet-2019-fgvc6/discussion/87544\">https://www.kaggle.com/c/imet-2019-fgvc6/discussion/87544</a> ), we can train locally and only test in kernels. However, we can come up with other tricks in Kernels, especially when dealing with private test data.</p>\n\n<p>I think we can make pseudo-labels for private test set in Kaggle Kernels, re-trained our pre-trained models for several steps on them. But I only come up with this idea (with my poor deep-learning knowledge)</p>\n\n<p>We can also use TTA</p>\n\n<p>Can someone else mentioned other tricks in Kaggle Kernels? Especially for private test set tricks, because this can not accomplished in local.</p>\n\n<p>Thanks!</p>",
      "rawMarkdown": "After we can use our own pretrained models in kaggle kernels (as discussed here https://www.kaggle.com/c/imet-2019-fgvc6/discussion/87544 ), we can train locally and only test in kernels. However, we can come up with other tricks in Kernels, especially when dealing with private test data.\n\nI think we can make pseudo-labels for private test set in Kaggle Kernels, re-trained our pre-trained models for several steps on them. But I only come up with this idea (with my poor deep-learning knowledge)\n\nWe can also use TTA\n\nCan someone else mentioned other tricks in Kaggle Kernels? Especially for private test set tricks, because this can not accomplished in local.\n\nThanks!\n  "
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "507091": "After we can use our own pretrained models in kaggle kernels (as discussed here https://www.kaggle.com/c/imet-2019-fgvc6/discussion/87544 ), we can train locally and only test in kernels. However, we can come up with other tricks in Kernels, especially when dealing with private test data.\n\nI think we can make pseudo-labels for private test set in Kaggle Kernels, re-trained our pre-trained models for several steps on them. But I only come up with this idea (with my poor deep-learning knowledge)\n\nWe can also use TTA\n\nCan someone else mentioned other tricks in Kaggle Kernels? Especially for private test set tricks, because this can not accomplished in local.\n\nThanks!\n  "
  }
}