{
  "id": 179548,
  "title": "Zero scores",
  "url": "/competitions/landmark-recognition-2020/discussion/179548",
  "author_name": "",
  "post_date": "2020-09-02T12:58:17.185863500Z",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hi,</p>\n<p>Did anyone try using simple softmax classification technique and scores? I'm constantly getting a zero score on my submissions. Although a lot of them, what I found by verifying the predicted classes and test images manually, seem to be correct.<br>\nWant to know, if anyone else is facing the same issue. It's just odd to see an absolute zero score till 4 decimal places, even though local validation accuracy is good enough.</p>\n<p>Couple of checks, that I have done : </p>\n<ol>\n<li><p>Submission samples are in the same order as the sample submission samples, since I'm predicting by reading test files from directory.</p></li>\n<li><p>Labels are in sync with training set after inverse transforming , using LabelEncoder.</p></li>\n<li><p>Label and score are in string format -&gt; str(label)+' '+str(score).</p></li>\n</ol>",
  "messages": [
    {
      "id": "995445",
      "postDate": "09/02/2020 12:58:17",
      "content": "<p>Hi,</p>\n<p>Did anyone try using simple softmax classification technique and scores? I'm constantly getting a zero score on my submissions. Although a lot of them, what I found by verifying the predicted classes and test images manually, seem to be correct.<br>\nWant to know, if anyone else is facing the same issue. It's just odd to see an absolute zero score till 4 decimal places, even though local validation accuracy is good enough.</p>\n<p>Couple of checks, that I have done : </p>\n<ol>\n<li><p>Submission samples are in the same order as the sample submission samples, since I'm predicting by reading test files from directory.</p></li>\n<li><p>Labels are in sync with training set after inverse transforming , using LabelEncoder.</p></li>\n<li><p>Label and score are in string format -&gt; str(label)+' '+str(score).</p></li>\n</ol>",
      "rawMarkdown": "Hi,\n\nDid anyone try using simple softmax classification technique and scores? I'm constantly getting a zero score on my submissions. Although a lot of them, what I found by verifying the predicted classes and test images manually, seem to be correct.\nWant to know, if anyone else is facing the same issue. It's just odd to see an absolute zero score till 4 decimal places, even though local validation accuracy is good enough.\n\nCouple of checks, that I have done : \n\n1. Submission samples are in the same order as the sample submission samples, since I'm predicting by reading test files from directory.\n\n2. Labels are in sync with training set after inverse transforming , using LabelEncoder.\n3. Label and score are in string format -> str(label)+' '+str(score).",
      "votes": null
    },
    {
      "id": "1003552",
      "postDate": "09/09/2020 04:55:22",
      "content": "<p>Same here. Please let me know if you find anything</p>",
      "rawMarkdown": "Same here. Please let me know if you find anything",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1003552,
      "author_name": "shanmugam212",
      "author_url": "",
      "post_date": "09/09/2020 04:55:22",
      "content": "<p>Same here. Please let me know if you find anything</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "995445": "Hi,\n\nDid anyone try using simple softmax classification technique and scores? I'm constantly getting a zero score on my submissions. Although a lot of them, what I found by verifying the predicted classes and test images manually, seem to be correct.\nWant to know, if anyone else is facing the same issue. It's just odd to see an absolute zero score till 4 decimal places, even though local validation accuracy is good enough.\n\nCouple of checks, that I have done : \n\n1. Submission samples are in the same order as the sample submission samples, since I'm predicting by reading test files from directory.\n\n2. Labels are in sync with training set after inverse transforming , using LabelEncoder.\n3. Label and score are in string format -> str(label)+' '+str(score).",
    "1003552": "Same here. Please let me know if you find anything"
  },
  "source": "meta"
}