{
  "id": 255480,
  "title": "Test data labels are not present",
  "url": "/competitions/digit-recognizer/discussion/255480",
  "author_name": "Chintamani Virulkar",
  "post_date": "2021-07-27T16:26:48.593000",
  "votes": 1,
  "comment_count": 4,
  "views": null,
  "content": "<p>How to calculate the accuracy of test data as labels are not mentioned for test data?</p>",
  "messages": [
    {
      "id": 1403003,
      "postDate": "2021-07-28T17:48:08.757Z",
      "content": "<p>Hi, when I did this competition, I used a different approach which I think is repeatable in all competitions. Naturally, MNIST is a public dataset so you can use the Keras MNIST dataset but in general, datasets are not public and test labels are not available. So this is what I did and it is pretty standard :<br>\n1/ split the training set into a training (e.g. 90% of images) and a validation set (10% of images). <br>\n2/ use the training set to train your models and the validation set to compare the performance of different models. <br>\n3/ select the best model (the one with the best accuracy on the validation set). <br>\n4/ optionnally (but I would recommend) retrain the best model using the complete training set (100% of images). <br>\nYou have your model and now it's time to check how well (or bad) it does on the test set. For this, you submit your model and the model accuracy on the test set will be displayed in the leaderboard.  <br>\nI hope it helps<br>\nGood luck</p>",
      "rawMarkdown": "Hi, when I did this competition, I used a different approach which I think is repeatable in all competitions. Naturally, MNIST is a public dataset so you can use the Keras MNIST dataset but in general, datasets are not public and test labels are not available. So this is what I did and it is pretty standard :\n1/ split the training set into a training (e.g. 90% of images) and a validation set (10% of images). \n2/ use the training set to train your models and the validation set to compare the performance of different models. \n3/ select the best model (the one with the best accuracy on the validation set). \n4/ optionnally (but I would recommend) retrain the best model using the complete training set (100% of images). \nYou have your model and now it's time to check how well (or bad) it does on the test set. For this, you submit your model and the model accuracy on the test set will be displayed in the leaderboard.  \nI hope it helps\nGood luck",
      "votes": 3
    },
    {
      "id": 1402001,
      "postDate": "2021-07-27T17:56:12.190Z",
      "content": "<p>Can use Keras example dataset. </p>\n<p>`import tensorflow as tf<br>\nmnist = tf.keras.datasets.mnist</p>\n<p>(x_train, y_train), (x_test, y_test) = mnist.load_data()`</p>\n<p>Something like this.</p>",
      "rawMarkdown": "Can use Keras example dataset. \n\n`import tensorflow as tf\nmnist = tf.keras.datasets.mnist\n\n(x_train, y_train), (x_test, y_test) = mnist.load_data()`\n\nSomething like this.",
      "votes": 2
    },
    {
      "id": 1401911,
      "postDate": "2021-07-27T16:26:48.593Z",
      "content": "<p>How to calculate the accuracy of test data as labels are not mentioned for test data?</p>",
      "rawMarkdown": "How to calculate the accuracy of test data as labels are not mentioned for test data?",
      "votes": 1
    },
    {
      "id": 1403424,
      "postDate": "2021-07-29T05:36:47.927Z",
      "content": "<p>Hey, <a href=\"https://www.kaggle.com/arnrob\" target=\"_blank\">@arnrob</a> thank you a lot for replying, this is very helpful.</p>",
      "rawMarkdown": "Hey, @arnrob thank you a lot for replying, this is very helpful."
    },
    {
      "id": 1402336,
      "postDate": "2021-07-28T05:25:43.490Z",
      "content": "<p><a href=\"https://www.kaggle.com/drwhoami\" target=\"_blank\">@drwhoami</a> thanks, this is helpfull.</p>",
      "rawMarkdown": "@drwhoami thanks, this is helpfull."
    }
  ],
  "comments": [
    {
      "id": 1403003,
      "author_name": "Arnaud",
      "author_url": "",
      "post_date": "2021-07-28T17:48:08.757000",
      "content": "<p>Hi, when I did this competition, I used a different approach which I think is repeatable in all competitions. Naturally, MNIST is a public dataset so you can use the Keras MNIST dataset but in general, datasets are not public and test labels are not available. So this is what I did and it is pretty standard :<br>\n1/ split the training set into a training (e.g. 90% of images) and a validation set (10% of images). <br>\n2/ use the training set to train your models and the validation set to compare the performance of different models. <br>\n3/ select the best model (the one with the best accuracy on the validation set). <br>\n4/ optionnally (but I would recommend) retrain the best model using the complete training set (100% of images). <br>\nYou have your model and now it's time to check how well (or bad) it does on the test set. For this, you submit your model and the model accuracy on the test set will be displayed in the leaderboard.  <br>\nI hope it helps<br>\nGood luck</p>",
      "votes": 3,
      "replies": []
    },
    {
      "id": 1402001,
      "author_name": "Aman Chauhan",
      "author_url": "",
      "post_date": "2021-07-27T17:56:12.190000",
      "content": "<p>Can use Keras example dataset. </p>\n<p>`import tensorflow as tf<br>\nmnist = tf.keras.datasets.mnist</p>\n<p>(x_train, y_train), (x_test, y_test) = mnist.load_data()`</p>\n<p>Something like this.</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 1403424,
      "author_name": "Chintamani Virulkar",
      "author_url": "",
      "post_date": "2021-07-29T05:36:47.927000",
      "content": "<p>Hey, <a href=\"https://www.kaggle.com/arnrob\" target=\"_blank\">@arnrob</a> thank you a lot for replying, this is very helpful.</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 1402336,
      "author_name": "Chintamani Virulkar",
      "author_url": "",
      "post_date": "2021-07-28T05:25:43.490000",
      "content": "<p><a href=\"https://www.kaggle.com/drwhoami\" target=\"_blank\">@drwhoami</a> thanks, this is helpfull.</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1403003": "Hi, when I did this competition, I used a different approach which I think is repeatable in all competitions. Naturally, MNIST is a public dataset so you can use the Keras MNIST dataset but in general, datasets are not public and test labels are not available. So this is what I did and it is pretty standard :\n1/ split the training set into a training (e.g. 90% of images) and a validation set (10% of images). \n2/ use the training set to train your models and the validation set to compare the performance of different models. \n3/ select the best model (the one with the best accuracy on the validation set). \n4/ optionnally (but I would recommend) retrain the best model using the complete training set (100% of images). \nYou have your model and now it's time to check how well (or bad) it does on the test set. For this, you submit your model and the model accuracy on the test set will be displayed in the leaderboard.  \nI hope it helps\nGood luck",
    "1402001": "Can use Keras example dataset. \n\n`import tensorflow as tf\nmnist = tf.keras.datasets.mnist\n\n(x_train, y_train), (x_test, y_test) = mnist.load_data()`\n\nSomething like this.",
    "1401911": "How to calculate the accuracy of test data as labels are not mentioned for test data?",
    "1403424": "Hey, @arnrob thank you a lot for replying, this is very helpful.",
    "1402336": "@drwhoami thanks, this is helpfull."
  }
}