{
  "id": 49191,
  "title": "Shake UP?   Correct classification of ~12 images separate the Top 10 teams",
  "url": "/competitions/sp-society-camera-model-identification/discussion/49191",
  "author_name": "",
  "post_date": "2018-02-07T20:28:11.307960Z",
  "votes": 6,
  "comment_count": 1,
  "views": 0,
  "content": "<p>My math might be wrong but from my calculations of the ~936 images in the leader-board evaluation set at 36%.  Correct classification of ~12 images separate the Top 10 teams, I didn't use the weighted accuracy metric they have for this competition but I think with the small test set I see some shakeup potential here :) </p>",
  "messages": [
    {
      "id": "279329",
      "postDate": "02/07/2018 20:28:11",
      "content": "<p>My math might be wrong but from my calculations of the ~936 images in the leader-board evaluation set at 36%.  Correct classification of ~12 images separate the Top 10 teams, I didn't use the weighted accuracy metric they have for this competition but I think with the small test set I see some shakeup potential here :) </p>",
      "rawMarkdown": "My math might be wrong but from my calculations of the ~936 images in the leader-board evaluation set at 36%.  Correct classification of ~12 images separate the Top 10 teams, I didn't use the weighted accuracy metric they have for this competition but I think with the small test set I see some shakeup potential here :)",
      "votes": null
    },
    {
      "id": "279340",
      "postDate": "02/07/2018 21:08:01",
      "content": "<p>36% of 2640 images should be around 950 images. And then it depends if you misclassify altered or original images. Then I noticed that my scores are pretty much correlated to the class predictions balance. I suspect the test set to be perfectly balanced; would be then easy to get an estimate of your final error. </p>",
      "rawMarkdown": "36% of 2640 images should be around 950 images. And then it depends if you misclassify altered or original images. Then I noticed that my scores are pretty much correlated to the class predictions balance. I suspect the test set to be perfectly balanced; would be then easy to get an estimate of your final error.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 279340,
      "author_name": "jeandebleau",
      "author_url": "",
      "post_date": "02/07/2018 21:08:01",
      "content": "<p>36% of 2640 images should be around 950 images. And then it depends if you misclassify altered or original images. Then I noticed that my scores are pretty much correlated to the class predictions balance. I suspect the test set to be perfectly balanced; would be then easy to get an estimate of your final error. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "279329": "My math might be wrong but from my calculations of the ~936 images in the leader-board evaluation set at 36%.  Correct classification of ~12 images separate the Top 10 teams, I didn't use the weighted accuracy metric they have for this competition but I think with the small test set I see some shakeup potential here :)",
    "279340": "36% of 2640 images should be around 950 images. And then it depends if you misclassify altered or original images. Then I noticed that my scores are pretty much correlated to the class predictions balance. I suspect the test set to be perfectly balanced; would be then easy to get an estimate of your final error."
  },
  "source": "meta"
}