{
  "id": 31609,
  "title": "Always the same few digits in prediction/inference",
  "url": "/competitions/digit-recognizer/discussion/31609",
  "author_name": "Joel McLeod",
  "post_date": "2017-04-14T11:54:30.152000",
  "votes": 0,
  "comment_count": 5,
  "views": null,
  "content": "<p>After training using a tensorflow cnn everything moves as it should, 97% accuracy at the end, validation similar.. I save it using saver and then restore and infer but the resulting predictions are clearly wrong.  Always 3 or 4 numbers repeated, like [2, 9, 0, 0, 2, 9, 5, 0, 2, 2, 9, 0, 0, 0, 2, 9, 0, 3, 9, 8, 0....] sort of thing.  A few of the other numbers in there, but always 95% of predictions are within a group of 3 or 4 numbers. </p>\n\n<p>Anyone seen anything like this or know what this is an indicator of?  </p>",
  "messages": [
    {
      "id": 175589,
      "postDate": "2017-04-16T10:40:33.167Z",
      "content": "<p>I just sent through a message with the code I'm using.  Tried to add it here but haven't used kernals before and just posting the code into a comment wasn't working well.  Let me know if you see anything that could be causing the issue I described</p>",
      "rawMarkdown": "I just sent through a message with the code I'm using.  Tried to add it here but haven't used kernals before and just posting the code into a comment wasn't working well.  Let me know if you see anything that could be causing the issue I described"
    },
    {
      "id": 175587,
      "postDate": "2017-04-16T10:33:38.507Z",
      "rawMarkdown": ""
    },
    {
      "id": 175586,
      "postDate": "2017-04-16T10:31:15.907Z",
      "rawMarkdown": ""
    },
    {
      "id": 175584,
      "postDate": "2017-04-16T10:07:04.847Z",
      "content": "<p>are you able to make a submission today?</p>",
      "rawMarkdown": "are you able to make a submission today?"
    },
    {
      "id": 175546,
      "postDate": "2017-04-16T03:23:30.357Z",
      "content": "<p>I'm not sure that whether the format of your train set and your test set is same. You can check it. Hoping can help you.</p>",
      "rawMarkdown": "I'm not sure that whether the format of your train set and your test set is same. You can check it. Hoping can help you."
    },
    {
      "id": 175220,
      "postDate": "2017-04-14T11:54:30.153Z",
      "content": "<p>After training using a tensorflow cnn everything moves as it should, 97% accuracy at the end, validation similar.. I save it using saver and then restore and infer but the resulting predictions are clearly wrong.  Always 3 or 4 numbers repeated, like [2, 9, 0, 0, 2, 9, 5, 0, 2, 2, 9, 0, 0, 0, 2, 9, 0, 3, 9, 8, 0....] sort of thing.  A few of the other numbers in there, but always 95% of predictions are within a group of 3 or 4 numbers. </p>\n\n<p>Anyone seen anything like this or know what this is an indicator of?  </p>",
      "rawMarkdown": "After training using a tensorflow cnn everything moves as it should, 97% accuracy at the end, validation similar.. I save it using saver and then restore and infer but the resulting predictions are clearly wrong.  Always 3 or 4 numbers repeated, like [2, 9, 0, 0, 2, 9, 5, 0, 2, 2, 9, 0, 0, 0, 2, 9, 0, 3, 9, 8, 0....] sort of thing.  A few of the other numbers in there, but always 95% of predictions are within a group of 3 or 4 numbers. \n\nAnyone seen anything like this or know what this is an indicator of?  "
    }
  ],
  "comments": [
    {
      "id": 175589,
      "author_name": "Joel McLeod",
      "author_url": "",
      "post_date": "2017-04-16T10:40:33.167000",
      "content": "<p>I just sent through a message with the code I'm using.  Tried to add it here but haven't used kernals before and just posting the code into a comment wasn't working well.  Let me know if you see anything that could be causing the issue I described</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 175587,
      "author_name": "Joel McLeod",
      "author_url": "",
      "post_date": "2017-04-16T10:33:38.507000",
      "content": "",
      "votes": 0,
      "replies": []
    },
    {
      "id": 175586,
      "author_name": "Joel McLeod",
      "author_url": "",
      "post_date": "2017-04-16T10:31:15.907000",
      "content": "",
      "votes": 0,
      "replies": []
    },
    {
      "id": 175584,
      "author_name": "Yushuai",
      "author_url": "",
      "post_date": "2017-04-16T10:07:04.847000",
      "content": "<p>are you able to make a submission today?</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 175546,
      "author_name": "zwangs",
      "author_url": "",
      "post_date": "2017-04-16T03:23:30.357000",
      "content": "<p>I'm not sure that whether the format of your train set and your test set is same. You can check it. Hoping can help you.</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "175589": "I just sent through a message with the code I'm using.  Tried to add it here but haven't used kernals before and just posting the code into a comment wasn't working well.  Let me know if you see anything that could be causing the issue I described",
    "175587": "",
    "175586": "",
    "175584": "are you able to make a submission today?",
    "175546": "I'm not sure that whether the format of your train set and your test set is same. You can check it. Hoping can help you.",
    "175220": "After training using a tensorflow cnn everything moves as it should, 97% accuracy at the end, validation similar.. I save it using saver and then restore and infer but the resulting predictions are clearly wrong.  Always 3 or 4 numbers repeated, like [2, 9, 0, 0, 2, 9, 5, 0, 2, 2, 9, 0, 0, 0, 2, 9, 0, 3, 9, 8, 0....] sort of thing.  A few of the other numbers in there, but always 95% of predictions are within a group of 3 or 4 numbers. \n\nAnyone seen anything like this or know what this is an indicator of?  "
  }
}