{
  "id": 217714,
  "title": "Questions about local validation and submission",
  "url": "/competitions/hpa-single-cell-image-classification/discussion/217714",
  "author_name": "Martin Chobanyan",
  "post_date": "2021-02-08T03:52:54.856000",
  "votes": 3,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hi, I had a couple of questions:</p>\n<ol>\n<li>Since the labels in this competition are at the whole image level, there's no way to calculate the mAP score on a local validation split? The only way to get the mAP score is to make a submission, right?</li>\n<li>Are we allowed to train a model locally, upload it, and then use it for inference in a notebook? If so, what is the best way to go about this?</li>\n</ol>",
  "messages": [
    {
      "id": 1191366,
      "postDate": "2021-02-08T12:49:53.217Z",
      "content": "<ol>\n<li>Seems like you are right</li>\n<li>Yes. Just upload model weights as an external dataset. Then load the model in you kernel and do inference.</li>\n</ol>",
      "rawMarkdown": "1. Seems like you are right\n2. Yes. Just upload model weights as an external dataset. Then load the model in you kernel and do inference.",
      "votes": 3,
      "replies": [
        {
          "id": 1192130,
          "postDate": "2021-02-09T01:45:36.597Z",
          "content": "<p>Good to know. Thanks for the response!</p>",
          "rawMarkdown": "Good to know. Thanks for the response!"
        }
      ]
    },
    {
      "id": 1190829,
      "postDate": "2021-02-08T03:52:54.857Z",
      "content": "<p>Hi, I had a couple of questions:</p>\n<ol>\n<li>Since the labels in this competition are at the whole image level, there's no way to calculate the mAP score on a local validation split? The only way to get the mAP score is to make a submission, right?</li>\n<li>Are we allowed to train a model locally, upload it, and then use it for inference in a notebook? If so, what is the best way to go about this?</li>\n</ol>",
      "rawMarkdown": "Hi, I had a couple of questions:\n1. Since the labels in this competition are at the whole image level, there's no way to calculate the mAP score on a local validation split? The only way to get the mAP score is to make a submission, right?\n2. Are we allowed to train a model locally, upload it, and then use it for inference in a notebook? If so, what is the best way to go about this?",
      "votes": 2
    }
  ],
  "comments": [
    {
      "id": 1191366,
      "author_name": "Vladislav Ostankovich",
      "author_url": "",
      "post_date": "2021-02-08T12:49:53.217000",
      "content": "<ol>\n<li>Seems like you are right</li>\n<li>Yes. Just upload model weights as an external dataset. Then load the model in you kernel and do inference.</li>\n</ol>",
      "votes": 3,
      "replies": [
        {
          "id": 1192130,
          "author_name": "Martin Chobanyan",
          "author_url": "",
          "post_date": "2021-02-09T01:45:36.597000",
          "content": "<p>Good to know. Thanks for the response!</p>",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1191366": "1. Seems like you are right\n2. Yes. Just upload model weights as an external dataset. Then load the model in you kernel and do inference.",
    "1190829": "Hi, I had a couple of questions:\n1. Since the labels in this competition are at the whole image level, there's no way to calculate the mAP score on a local validation split? The only way to get the mAP score is to make a submission, right?\n2. Are we allowed to train a model locally, upload it, and then use it for inference in a notebook? If so, what is the best way to go about this?"
  }
}