{
  "id": 34216,
  "title": "About accuracy and ImageId",
  "url": "/competitions/digit-recognizer/discussion/34216",
  "author_name": "Rafael Novello",
  "post_date": "2017-06-05T23:52:33.833000",
  "votes": 1,
  "comment_count": 2,
  "views": null,
  "content": "<p>Hi! I'm a new kaggle user and I'm get some trouble to understand my accuracy in digit recognizer.</p>\n\n<p>On my notebook [1] I get more than 90% accuracy, but kaggle said it's 10% more or less. Therefore, I believe there is a mistake in ImageId information. How to get the right ImageId?</p>\n\n<p>In MNIST page [2] said \"test set of 10,000 examples\", then how to get 28.000 results in test set?</p>\n\n<p>Sorry if my question is a \"beginner one\", but I can't find any answer.\nThanks for any help!</p>\n\n<p>[1] - <a href=\"https://github.com/rafaelnovello/Deep-Learning-Foundation/blob/master/MNIST/TFLearn_Digit_Recognition.ipynb\">https://github.com/rafaelnovello/Deep-Learning-Foundation/blob/master/MNIST/TFLearn_Digit_Recognition.ipynb</a></p>\n\n<p>[2] - <a href=\"http://yann.lecun.com/exdb/mnist/\">http://yann.lecun.com/exdb/mnist/</a></p>",
  "messages": [
    {
      "id": 191802,
      "postDate": "2017-06-11T20:43:32.963Z",
      "content": "<p>The test set and training set you are supposed to use are <a href=\"https://www.kaggle.com/c/digit-recognizer/data\">here.</a></p>\n\n<p>The training set is smaller than the normal one at around 35k images, as opposed to 60k.\nThe testing set isn't labelled by number, but the line number is the same as the digit id, so you can do something like:\n<code>predictions = [str(i+1) + \",\" + str(model.predict(test_data[i])) for i in range(0,len(test_data)]</code> and then write that to the file you submit.</p>",
      "rawMarkdown": "The test set and training set you are supposed to use are [here.](https://www.kaggle.com/c/digit-recognizer/data)\n\nThe training set is smaller than the normal one at around 35k images, as opposed to 60k.\nThe testing set isn't labelled by number, but the line number is the same as the digit id, so you can do something like:\n`predictions = [str(i+1) + \",\" + str(model.predict(test_data[i])) for i in range(0,len(test_data)]` and then write that to the file you submit.",
      "votes": 3,
      "replies": [
        {
          "id": 192530,
          "postDate": "2017-06-13T22:43:22.313Z",
          "content": "<p>Hmmm I'll try again with right dataset! Thanks a lot!</p>",
          "rawMarkdown": "Hmmm I'll try again with right dataset! Thanks a lot!",
          "votes": 1
        }
      ]
    },
    {
      "id": 189400,
      "postDate": "2017-06-05T23:52:33.833Z",
      "content": "<p>Hi! I'm a new kaggle user and I'm get some trouble to understand my accuracy in digit recognizer.</p>\n\n<p>On my notebook [1] I get more than 90% accuracy, but kaggle said it's 10% more or less. Therefore, I believe there is a mistake in ImageId information. How to get the right ImageId?</p>\n\n<p>In MNIST page [2] said \"test set of 10,000 examples\", then how to get 28.000 results in test set?</p>\n\n<p>Sorry if my question is a \"beginner one\", but I can't find any answer.\nThanks for any help!</p>\n\n<p>[1] - <a href=\"https://github.com/rafaelnovello/Deep-Learning-Foundation/blob/master/MNIST/TFLearn_Digit_Recognition.ipynb\">https://github.com/rafaelnovello/Deep-Learning-Foundation/blob/master/MNIST/TFLearn_Digit_Recognition.ipynb</a></p>\n\n<p>[2] - <a href=\"http://yann.lecun.com/exdb/mnist/\">http://yann.lecun.com/exdb/mnist/</a></p>",
      "rawMarkdown": "Hi! I'm a new kaggle user and I'm get some trouble to understand my accuracy in digit recognizer.\n\nOn my notebook [1] I get more than 90% accuracy, but kaggle said it's 10% more or less. Therefore, I believe there is a mistake in ImageId information. How to get the right ImageId?\n\nIn MNIST page [2] said \"test set of 10,000 examples\", then how to get 28.000 results in test set?\n\nSorry if my question is a \"beginner one\", but I can't find any answer.\nThanks for any help!\n\n[1] - https://github.com/rafaelnovello/Deep-Learning-Foundation/blob/master/MNIST/TFLearn_Digit_Recognition.ipynb\n\n[2] - http://yann.lecun.com/exdb/mnist/",
      "votes": 1
    }
  ],
  "comments": [
    {
      "id": 191802,
      "author_name": "pogrmman",
      "author_url": "",
      "post_date": "2017-06-11T20:43:32.963000",
      "content": "<p>The test set and training set you are supposed to use are <a href=\"https://www.kaggle.com/c/digit-recognizer/data\">here.</a></p>\n\n<p>The training set is smaller than the normal one at around 35k images, as opposed to 60k.\nThe testing set isn't labelled by number, but the line number is the same as the digit id, so you can do something like:\n<code>predictions = [str(i+1) + \",\" + str(model.predict(test_data[i])) for i in range(0,len(test_data)]</code> and then write that to the file you submit.</p>",
      "votes": 3,
      "replies": [
        {
          "id": 192530,
          "author_name": "Rafael Novello",
          "author_url": "",
          "post_date": "2017-06-13T22:43:22.313000",
          "content": "<p>Hmmm I'll try again with right dataset! Thanks a lot!</p>",
          "votes": 1,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "191802": "The test set and training set you are supposed to use are [here.](https://www.kaggle.com/c/digit-recognizer/data)\n\nThe training set is smaller than the normal one at around 35k images, as opposed to 60k.\nThe testing set isn't labelled by number, but the line number is the same as the digit id, so you can do something like:\n`predictions = [str(i+1) + \",\" + str(model.predict(test_data[i])) for i in range(0,len(test_data)]` and then write that to the file you submit.",
    "189400": "Hi! I'm a new kaggle user and I'm get some trouble to understand my accuracy in digit recognizer.\n\nOn my notebook [1] I get more than 90% accuracy, but kaggle said it's 10% more or less. Therefore, I believe there is a mistake in ImageId information. How to get the right ImageId?\n\nIn MNIST page [2] said \"test set of 10,000 examples\", then how to get 28.000 results in test set?\n\nSorry if my question is a \"beginner one\", but I can't find any answer.\nThanks for any help!\n\n[1] - https://github.com/rafaelnovello/Deep-Learning-Foundation/blob/master/MNIST/TFLearn_Digit_Recognition.ipynb\n\n[2] - http://yann.lecun.com/exdb/mnist/"
  }
}