{
  "id": 183951,
  "title": "Code requirements ",
  "url": "/competitions/rsna-str-pulmonary-embolism-detection/discussion/183951",
  "author_name": "Anshu Trivedi",
  "post_date": "2020-09-18T17:31:17.076000",
  "votes": 0,
  "comment_count": 3,
  "views": 0,
  "content": "<p>This is an inference-only code competition. Your submissions will not have access to the training images, so you must train your models elsewhere and incorporate them into your submission, without reference to the folder containing the train images</p>\n<p>What does it mean ? I didn't get train models elsewhere?<br>\nPlease someone explain to me.</p>",
  "messages": [
    {
      "id": 1016172,
      "postDate": "2020-09-18T18:04:47.163Z",
      "content": "<p>The training and test data can be downloaded from the \"Data\" option in the competition.</p>\n<p>It is a large dataset, so many find it easier to work within Notebooks.</p>\n<p>There are also user produced Datasets with the images in JPEG format which are much smaller.</p>\n<p>As a \"Code Competition\", you must submit a Notebook that makes predictions on a hidden test dataset. You must train your model first (either on your own computer or in a Notebook). Then you save the model and it's weights. In your submitted Notebook you load the model and it is run against the private test data.</p>\n<p>Other \"Code Competitions\" let you train and predict in the same notebook. This dataset is so large, it is probably unrealistic to try that anyway.</p>\n<p>Process:</p>\n<ol>\n<li>Create Training Notebook or work locally</li>\n<li>Train model</li>\n<li>Save model design and weights</li>\n<li>Place model design/weights in a Dataset</li>\n<li>Create Prediction (Inference) Notebook</li>\n<li>Attach your Dataset that contains the model</li>\n<li>Predict the hidden test data using the model.</li>\n</ol>\n<p>You can put Training and Prediction/Inference in the same Notebook for development, but you only have access to about 1/3 of the read test data. So eventually you need to break out the steps.</p>\n<p>Good luck!</p>\n<p>-Rich</p>",
      "rawMarkdown": "The training and test data can be downloaded from the \"Data\" option in the competition.\n\nIt is a large dataset, so many find it easier to work within Notebooks.\n\nThere are also user produced Datasets with the images in JPEG format which are much smaller.\n\nAs a \"Code Competition\", you must submit a Notebook that makes predictions on a hidden test dataset. You must train your model first (either on your own computer or in a Notebook). Then you save the model and it's weights. In your submitted Notebook you load the model and it is run against the private test data.\n\nOther \"Code Competitions\" let you train and predict in the same notebook. This dataset is so large, it is probably unrealistic to try that anyway.\n\nProcess:\n1. Create Training Notebook or work locally\n2. Train model\n3. Save model design and weights\n4. Place model design/weights in a Dataset\n4. Create Prediction (Inference) Notebook\n5. Attach your Dataset that contains the model\n6. Predict the hidden test data using the model.\n\nYou can put Training and Prediction/Inference in the same Notebook for development, but you only have access to about 1/3 of the read test data. So eventually you need to break out the steps.\n\nGood luck!\n\n-Rich",
      "votes": 1,
      "replies": [
        {
          "id": 1016184,
          "postDate": "2020-09-18T18:15:28.463Z",
          "content": "<p>Thank you so much Rich.It is helpful.</p>",
          "rawMarkdown": "Thank you so much Rich.It is helpful."
        }
      ]
    },
    {
      "id": 1020556,
      "postDate": "2020-09-21T09:00:59.633Z",
      "content": "<p>You can save your trained model in a Kaggle dataset and use it to make predictions.</p>",
      "rawMarkdown": "You can save your trained model in a Kaggle dataset and use it to make predictions."
    },
    {
      "id": 1016137,
      "postDate": "2020-09-18T17:31:17.077Z",
      "content": "<p>This is an inference-only code competition. Your submissions will not have access to the training images, so you must train your models elsewhere and incorporate them into your submission, without reference to the folder containing the train images</p>\n<p>What does it mean ? I didn't get train models elsewhere?<br>\nPlease someone explain to me.</p>",
      "rawMarkdown": "This is an inference-only code competition. Your submissions will not have access to the training images, so you must train your models elsewhere and incorporate them into your submission, without reference to the folder containing the train images\n\nWhat does it mean ? I didn't get train models elsewhere?\nPlease someone explain to me."
    }
  ],
  "comments": [
    {
      "id": 1016172,
      "author_name": "quadcore/Richard Epstein",
      "author_url": "",
      "post_date": "2020-09-18T18:04:47.163000",
      "content": "<p>The training and test data can be downloaded from the \"Data\" option in the competition.</p>\n<p>It is a large dataset, so many find it easier to work within Notebooks.</p>\n<p>There are also user produced Datasets with the images in JPEG format which are much smaller.</p>\n<p>As a \"Code Competition\", you must submit a Notebook that makes predictions on a hidden test dataset. You must train your model first (either on your own computer or in a Notebook). Then you save the model and it's weights. In your submitted Notebook you load the model and it is run against the private test data.</p>\n<p>Other \"Code Competitions\" let you train and predict in the same notebook. This dataset is so large, it is probably unrealistic to try that anyway.</p>\n<p>Process:</p>\n<ol>\n<li>Create Training Notebook or work locally</li>\n<li>Train model</li>\n<li>Save model design and weights</li>\n<li>Place model design/weights in a Dataset</li>\n<li>Create Prediction (Inference) Notebook</li>\n<li>Attach your Dataset that contains the model</li>\n<li>Predict the hidden test data using the model.</li>\n</ol>\n<p>You can put Training and Prediction/Inference in the same Notebook for development, but you only have access to about 1/3 of the read test data. So eventually you need to break out the steps.</p>\n<p>Good luck!</p>\n<p>-Rich</p>",
      "votes": 1,
      "replies": [
        {
          "id": 1016184,
          "author_name": "Anshu Trivedi",
          "author_url": "",
          "post_date": "2020-09-18T18:15:28.463000",
          "content": "<p>Thank you so much Rich.It is helpful.</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 1020556,
      "author_name": "Sankalp Sharma",
      "author_url": "",
      "post_date": "2020-09-21T09:00:59.633000",
      "content": "<p>You can save your trained model in a Kaggle dataset and use it to make predictions.</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1016172": "The training and test data can be downloaded from the \"Data\" option in the competition.\n\nIt is a large dataset, so many find it easier to work within Notebooks.\n\nThere are also user produced Datasets with the images in JPEG format which are much smaller.\n\nAs a \"Code Competition\", you must submit a Notebook that makes predictions on a hidden test dataset. You must train your model first (either on your own computer or in a Notebook). Then you save the model and it's weights. In your submitted Notebook you load the model and it is run against the private test data.\n\nOther \"Code Competitions\" let you train and predict in the same notebook. This dataset is so large, it is probably unrealistic to try that anyway.\n\nProcess:\n1. Create Training Notebook or work locally\n2. Train model\n3. Save model design and weights\n4. Place model design/weights in a Dataset\n4. Create Prediction (Inference) Notebook\n5. Attach your Dataset that contains the model\n6. Predict the hidden test data using the model.\n\nYou can put Training and Prediction/Inference in the same Notebook for development, but you only have access to about 1/3 of the read test data. So eventually you need to break out the steps.\n\nGood luck!\n\n-Rich",
    "1020556": "You can save your trained model in a Kaggle dataset and use it to make predictions.",
    "1016137": "This is an inference-only code competition. Your submissions will not have access to the training images, so you must train your models elsewhere and incorporate them into your submission, without reference to the folder containing the train images\n\nWhat does it mean ? I didn't get train models elsewhere?\nPlease someone explain to me."
  }
}