{
  "id": 116100,
  "title": "Estimation of the sample number in PB",
  "url": "/competitions/rsna-intracranial-hemorrhage-detection/discussion/116100",
  "author_name": "Yiheng Wang",
  "post_date": "2019-11-07T05:18:27.914000",
  "votes": 7,
  "comment_count": 4,
  "views": 0,
  "content": "<p>according to <a href=\"/yaroshevskiy\">@yaroshevskiy</a> 's probe results: 16.353 and 18.185, (<a href=\"https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/115976#latest-666784\">https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/115976#latest-666784</a>), the estimated PB examples should be 35 or 70 or 105, ...</p>\n\n<p>For example, if there are 35 samples, then the total weights are:\n&gt; 35*(5*1+1*2) = 245</p>\n\n<p>Here we just consider one example has 7 predictions (2 for any) for convenience. Among of them,  if the number of \"0\" samples is 129, and \"1\" for 116, then we can get:</p>\n\n<p>&gt; (-log(1-(1e-15))*129 + (-log(1e-15))*116)/245 = 16.35305331350873\n(-log(1-(1e-15))*116 + (-log(1e-15))*129)/245 = 18.185723081401953</p>\n\n<p>Related code is attached</p>\n\n<p>----update-----\nWe also have to include 13.837, thus the number should be 455, 665, (455+35*7), 910, ...</p>\n\n<blockquote>\n  <p>455 1677 1508 16.353053313508735\n  455 1508 1677 18.185723081401953\n  455 1909 1276 13.837198957584311</p>\n</blockquote>",
  "messages": [
    {
      "id": 667307,
      "postDate": "2019-11-07T05:18:27.913Z",
      "content": "<p>according to <a href=\"/yaroshevskiy\">@yaroshevskiy</a> 's probe results: 16.353 and 18.185, (<a href=\"https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/115976#latest-666784\">https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/115976#latest-666784</a>), the estimated PB examples should be 35 or 70 or 105, ...</p>\n\n<p>For example, if there are 35 samples, then the total weights are:\n&gt; 35*(5*1+1*2) = 245</p>\n\n<p>Here we just consider one example has 7 predictions (2 for any) for convenience. Among of them,  if the number of \"0\" samples is 129, and \"1\" for 116, then we can get:</p>\n\n<p>&gt; (-log(1-(1e-15))*129 + (-log(1e-15))*116)/245 = 16.35305331350873\n(-log(1-(1e-15))*116 + (-log(1e-15))*129)/245 = 18.185723081401953</p>\n\n<p>Related code is attached</p>\n\n<p>----update-----\nWe also have to include 13.837, thus the number should be 455, 665, (455+35*7), 910, ...</p>\n\n<blockquote>\n  <p>455 1677 1508 16.353053313508735\n  455 1508 1677 18.185723081401953\n  455 1909 1276 13.837198957584311</p>\n</blockquote>",
      "rawMarkdown": "according to @yaroshevskiy 's probe results: 16.353 and 18.185, (https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/115976#latest-666784), the estimated PB examples should be 35 or 70 or 105, ...\n\nFor example, if there are 35 samples, then the total weights are:\n&gt; 35*(5*1+1*2) = 245\n\nHere we just consider one example has 7 predictions (2 for any) for convenience. Among of them,  if the number of \"0\" samples is 129, and \"1\" for 116, then we can get:\n\n&gt; (-log(1-(1e-15))*129 + (-log(1e-15))*116)/245 = 16.35305331350873\n(-log(1-(1e-15))*116 + (-log(1e-15))*129)/245 = 18.185723081401953\n\nRelated code is attached\n\n----update-----\nWe also have to include 13.837, thus the number should be 455, 665, (455+35*7), 910, ...\n\n&gt; 455 1677 1508 16.353053313508735\n455 1508 1677 18.185723081401953\n455 1909 1276 13.837198957584311",
      "votes": 7
    },
    {
      "id": 667313,
      "postDate": "2019-11-07T05:35:24.320Z",
      "content": "<p>You may add score 13.837 to consideration, as this is what <a href=\"https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/116025#667069\">two people reported they got</a> after clipping values to 0/1. I think it invalidates your number 245.</p>",
      "rawMarkdown": "You may add score 13.837 to consideration, as this is what [two people reported they got](https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/116025#667069) after clipping values to 0/1. I think it invalidates your number 245.",
      "votes": 1,
      "replies": [
        {
          "id": 667324,
          "postDate": "2019-11-07T05:53:18.477Z",
          "content": "<p>Thanks, I modified the post.</p>",
          "rawMarkdown": "Thanks, I modified the post."
        }
      ]
    },
    {
      "id": 668257,
      "postDate": "2019-11-08T07:21:39.470Z",
      "content": "<p>Suppose we have around of 1%. It is 1212 images.\nIs 1190=455+2 * 35 * 7 ok by yours calculations?</p>",
      "rawMarkdown": "Suppose we have around of 1%. It is 1212 images.\nIs 1190=455+2 * 35 * 7 ok by yours calculations?",
      "replies": [
        {
          "id": 668340,
          "postDate": "2019-11-08T09:42:00.277Z",
          "content": "<p>not sure for the exact number, I wish it is just 455 according to my bad PB score ; )</p>",
          "rawMarkdown": "not sure for the exact number, I wish it is just 455 according to my bad PB score ; )"
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 667313,
      "author_name": "nosound",
      "author_url": "",
      "post_date": "2019-11-07T05:35:24.320000",
      "content": "<p>You may add score 13.837 to consideration, as this is what <a href=\"https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/116025#667069\">two people reported they got</a> after clipping values to 0/1. I think it invalidates your number 245.</p>",
      "votes": 1,
      "replies": [
        {
          "id": 667324,
          "author_name": "Yiheng Wang",
          "author_url": "",
          "post_date": "2019-11-07T05:53:18.477000",
          "content": "<p>Thanks, I modified the post.</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 668257,
      "author_name": "Sergey Zlobin",
      "author_url": "",
      "post_date": "2019-11-08T07:21:39.470000",
      "content": "<p>Suppose we have around of 1%. It is 1212 images.\nIs 1190=455+2 * 35 * 7 ok by yours calculations?</p>",
      "votes": 0,
      "replies": [
        {
          "id": 668340,
          "author_name": "Yiheng Wang",
          "author_url": "",
          "post_date": "2019-11-08T09:42:00.277000",
          "content": "<p>not sure for the exact number, I wish it is just 455 according to my bad PB score ; )</p>",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "667307": "according to @yaroshevskiy 's probe results: 16.353 and 18.185, (https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/115976#latest-666784), the estimated PB examples should be 35 or 70 or 105, ...\n\nFor example, if there are 35 samples, then the total weights are:\n&gt; 35*(5*1+1*2) = 245\n\nHere we just consider one example has 7 predictions (2 for any) for convenience. Among of them,  if the number of \"0\" samples is 129, and \"1\" for 116, then we can get:\n\n&gt; (-log(1-(1e-15))*129 + (-log(1e-15))*116)/245 = 16.35305331350873\n(-log(1-(1e-15))*116 + (-log(1e-15))*129)/245 = 18.185723081401953\n\nRelated code is attached\n\n----update-----\nWe also have to include 13.837, thus the number should be 455, 665, (455+35*7), 910, ...\n\n&gt; 455 1677 1508 16.353053313508735\n455 1508 1677 18.185723081401953\n455 1909 1276 13.837198957584311",
    "667313": "You may add score 13.837 to consideration, as this is what [two people reported they got](https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/116025#667069) after clipping values to 0/1. I think it invalidates your number 245.",
    "668257": "Suppose we have around of 1%. It is 1212 images.\nIs 1190=455+2 * 35 * 7 ok by yours calculations?"
  }
}