{
  "id": 41294,
  "title": "sample_submission.csv  do not have label like labels.csv  ",
  "url": "/competitions/dog-breed-identification/discussion/41294",
  "author_name": "",
  "post_date": "2017-10-16T03:08:57.318787300Z",
  "votes": 2,
  "comment_count": 6,
  "views": 0,
  "content": "<p>how to use test data set ,i have look sample_submission.csv  and each element is 0.0083 . what the mean?</p>",
  "messages": [
    {
      "id": "231799",
      "postDate": "10/16/2017 03:08:57",
      "content": "<p>how to use test data set ,i have look sample_submission.csv  and each element is 0.0083 . what the mean?</p>",
      "rawMarkdown": "how to use test data set ,i have look sample_submission.csv  and each element is 0.0083 . what the mean?",
      "votes": null
    },
    {
      "id": "231960",
      "postDate": "10/16/2017 14:12:25",
      "content": "<p>Each cell is the 'probability' that that photo (row) is that dog breed (column) - if you sum the row it should add up to 1.</p>\n\n<p>There are 120 different breeds of dogs (classes). In the sample each one 0.0083 is just 1 / 120 - you can interpret that as the submission is showing no idea which breed a photo is so its giving all the same probability. Its just an example remember.</p>\n\n<p>If you generated this for real you would see that one class (column) on each row has a number close to 1 (pretty sure that the photo is this breed) and other others close to 0.</p>\n\n<p>If the classifier isn't sure what breed a dog is you may see a few columns with number of say 0.3.</p>",
      "rawMarkdown": "Each cell is the 'probability' that that photo (row) is that dog breed (column) - if you sum the row it should add up to 1.\n\nThere are 120 different breeds of dogs (classes). In the sample each one 0.0083 is just 1 / 120 - you can interpret that as the submission is showing no idea which breed a photo is so its giving all the same probability. Its just an example remember.\n\nIf you generated this for real you would see that one class (column) on each row has a number close to 1 (pretty sure that the photo is this breed) and other others close to 0.\n\nIf the classifier isn't sure what breed a dog is you may see a few columns with number of say 0.3.",
      "votes": null
    },
    {
      "id": "231977",
      "postDate": "10/16/2017 14:47:56",
      "content": "<p>i am sure i have understood what you say ,but the test set just only some image with end name jpg, so i want to say ,the test data do not have label ,i can get get each image label in test set,can you help me ?</p>",
      "rawMarkdown": "i am sure i have understood what you say ,but the test set just only some image with end name jpg, so i want to say ,the test data do not have label ,i can get get each image label in test set,can you help me ?",
      "votes": null
    },
    {
      "id": "231978",
      "postDate": "10/16/2017 14:48:45",
      "content": "<p>i am sure i have understood what you say ,but the test set just only some image with end name jpg, so i want to say ,the test data do not have label ,i can not  get get each image label in test set,can you help me ?</p>",
      "rawMarkdown": "i am sure i have understood what you say ,but the test set just only some image with end name jpg, so i want to say ,the test data do not have label ,i can not  get get each image label in test set,can you help me ?",
      "votes": null
    },
    {
      "id": "232012",
      "postDate": "10/16/2017 15:48:48",
      "content": "<p>Test data don't come with labels because you can only have labels for the training set. </p>\n\n<p>I am guessing what you mean by the labels in the test set is the prediction from the test data after the model has been trained.  </p>\n\n<p>For every machine/deep learning practice, you can abstractly simplify the system as y=Xb+e in your thinking process, where y, X, b and e are vectors and matrixes. </p>\n\n<p>At first, you will be given y and X in the training set to train b and e. </p>\n\n<p>Then input X from the test dataset into the trained b and e to predict y. This is why in the test set, you don't get to be given labels. </p>",
      "rawMarkdown": "Test data don't come with labels because you can only have labels for the training set. \n\nI am guessing what you mean by the labels in the test set is the prediction from the test data after the model has been trained.  \n\nFor every machine/deep learning practice, you can abstractly simplify the system as y=Xb+e in your thinking process, where y, X, b and e are vectors and matrixes. \n\nAt first, you will be given y and X in the training set to train b and e. \n\nThen input X from the test dataset into the trained b and e to predict y. This is why in the test set, you don't get to be given labels.",
      "votes": null
    },
    {
      "id": "233749",
      "postDate": "10/21/2017 02:52:24",
      "content": "<p>As Carmen said, no labels are provided for the test data. Your goal is to generate the predictions for the test set. </p>\n\n<p>If you want to test how well your model is performing, you should split the training data into training and validation sets. Then you train with the new (smaller) training set and and check how well you are doing using the validation set. Once you are happy with your results, use you model to predict the results for the test set.</p>",
      "rawMarkdown": "As Carmen said, no labels are provided for the test data. Your goal is to generate the predictions for the test set. \n\nIf you want to test how well your model is performing, you should split the training data into training and validation sets. Then you train with the new (smaller) training set and and check how well you are doing using the validation set. Once you are happy with your results, use you model to predict the results for the test set.",
      "votes": null
    },
    {
      "id": "616443",
      "postDate": "09/03/2019 06:19:52",
      "content": "",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 231960,
      "author_name": "rhinoo",
      "author_url": "",
      "post_date": "10/16/2017 14:12:25",
      "content": "<p>Each cell is the 'probability' that that photo (row) is that dog breed (column) - if you sum the row it should add up to 1.</p>\n\n<p>There are 120 different breeds of dogs (classes). In the sample each one 0.0083 is just 1 / 120 - you can interpret that as the submission is showing no idea which breed a photo is so its giving all the same probability. Its just an example remember.</p>\n\n<p>If you generated this for real you would see that one class (column) on each row has a number close to 1 (pretty sure that the photo is this breed) and other others close to 0.</p>\n\n<p>If the classifier isn't sure what breed a dog is you may see a few columns with number of say 0.3.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 231977,
      "author_name": "yangshihao",
      "author_url": "",
      "post_date": "10/16/2017 14:47:56",
      "content": "<p>i am sure i have understood what you say ,but the test set just only some image with end name jpg, so i want to say ,the test data do not have label ,i can get get each image label in test set,can you help me ?</p>",
      "votes": null,
      "replies": [
        {
          "id": 616443,
          "author_name": "sd4321",
          "author_url": "",
          "post_date": "09/03/2019 06:19:52",
          "content": "",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 231978,
      "author_name": "yangshihao",
      "author_url": "",
      "post_date": "10/16/2017 14:48:45",
      "content": "<p>i am sure i have understood what you say ,but the test set just only some image with end name jpg, so i want to say ,the test data do not have label ,i can not  get get each image label in test set,can you help me ?</p>",
      "votes": null,
      "replies": [
        {
          "id": 232012,
          "author_name": "carmnejsu",
          "author_url": "",
          "post_date": "10/16/2017 15:48:48",
          "content": "<p>Test data don't come with labels because you can only have labels for the training set. </p>\n\n<p>I am guessing what you mean by the labels in the test set is the prediction from the test data after the model has been trained.  </p>\n\n<p>For every machine/deep learning practice, you can abstractly simplify the system as y=Xb+e in your thinking process, where y, X, b and e are vectors and matrixes. </p>\n\n<p>At first, you will be given y and X in the training set to train b and e. </p>\n\n<p>Then input X from the test dataset into the trained b and e to predict y. This is why in the test set, you don't get to be given labels. </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 233749,
          "author_name": "bluevalhalla",
          "author_url": "",
          "post_date": "10/21/2017 02:52:24",
          "content": "<p>As Carmen said, no labels are provided for the test data. Your goal is to generate the predictions for the test set. </p>\n\n<p>If you want to test how well your model is performing, you should split the training data into training and validation sets. Then you train with the new (smaller) training set and and check how well you are doing using the validation set. Once you are happy with your results, use you model to predict the results for the test set.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "231799": "how to use test data set ,i have look sample_submission.csv  and each element is 0.0083 . what the mean?",
    "231960": "Each cell is the 'probability' that that photo (row) is that dog breed (column) - if you sum the row it should add up to 1.\n\nThere are 120 different breeds of dogs (classes). In the sample each one 0.0083 is just 1 / 120 - you can interpret that as the submission is showing no idea which breed a photo is so its giving all the same probability. Its just an example remember.\n\nIf you generated this for real you would see that one class (column) on each row has a number close to 1 (pretty sure that the photo is this breed) and other others close to 0.\n\nIf the classifier isn't sure what breed a dog is you may see a few columns with number of say 0.3.",
    "231977": "i am sure i have understood what you say ,but the test set just only some image with end name jpg, so i want to say ,the test data do not have label ,i can get get each image label in test set,can you help me ?",
    "231978": "i am sure i have understood what you say ,but the test set just only some image with end name jpg, so i want to say ,the test data do not have label ,i can not  get get each image label in test set,can you help me ?",
    "232012": "Test data don't come with labels because you can only have labels for the training set. \n\nI am guessing what you mean by the labels in the test set is the prediction from the test data after the model has been trained.  \n\nFor every machine/deep learning practice, you can abstractly simplify the system as y=Xb+e in your thinking process, where y, X, b and e are vectors and matrixes. \n\nAt first, you will be given y and X in the training set to train b and e. \n\nThen input X from the test dataset into the trained b and e to predict y. This is why in the test set, you don't get to be given labels.",
    "233749": "As Carmen said, no labels are provided for the test data. Your goal is to generate the predictions for the test set. \n\nIf you want to test how well your model is performing, you should split the training data into training and validation sets. Then you train with the new (smaller) training set and and check how well you are doing using the validation set. Once you are happy with your results, use you model to predict the results for the test set.",
    "616443": ""
  },
  "source": "meta"
}