{
  "id": 584372,
  "title": "Test your submission pipeline for silly mistakes and submission time",
  "url": "/competitions/cmi-detect-behavior-with-sensor-data/discussion/584372",
  "author_name": "",
  "post_date": "2025-06-12T21:17:54.488348700Z",
  "votes": 22,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hello all,</p>\n<p>I curated a API tester dataset with the training data, using 3500 sequences as base. We are informed that we have 3500 sequences in the test set, so using this number replicates the approximate test set. The dataset is <a href=\"https://www.kaggle.com/code/ravi20076/cmi2025-api-public-test-v1\" target=\"_blank\">here</a> for reference. I have also demonstrated how one may use this in the API to make a test <strong>before the LB submission</strong>.</p>\n<p>I also use 1750 sequences with IMU features only and make the other sensors as np.nan. This is mainly to test your submissions for silly mistakes. I think this is crucial to save your submissions from silly mistakes and wasted submission attempts. I urge beginners to use this dataset to test their code for issues beforehand prior to making a LB submission to avoid wasted efforts!</p>\n<p>Also, this dataset could be used to test the <strong>submission time</strong> on the leaderboard. We could find a small difference between the actual time to submit and this dataset, as the sequence lengths could be slightly different across the datasets. </p>\n<p>Wishing you the best! Regards!</p>",
  "messages": [
    {
      "id": "3223127",
      "postDate": "06/12/2025 21:17:54",
      "content": "<p>Hello all,</p>\n<p>I curated a API tester dataset with the training data, using 3500 sequences as base. We are informed that we have 3500 sequences in the test set, so using this number replicates the approximate test set. The dataset is <a href=\"https://www.kaggle.com/code/ravi20076/cmi2025-api-public-test-v1\" target=\"_blank\">here</a> for reference. I have also demonstrated how one may use this in the API to make a test <strong>before the LB submission</strong>.</p>\n<p>I also use 1750 sequences with IMU features only and make the other sensors as np.nan. This is mainly to test your submissions for silly mistakes. I think this is crucial to save your submissions from silly mistakes and wasted submission attempts. I urge beginners to use this dataset to test their code for issues beforehand prior to making a LB submission to avoid wasted efforts!</p>\n<p>Also, this dataset could be used to test the <strong>submission time</strong> on the leaderboard. We could find a small difference between the actual time to submit and this dataset, as the sequence lengths could be slightly different across the datasets. </p>\n<p>Wishing you the best! Regards!</p>",
      "rawMarkdown": "Hello all,\n\nI curated a API tester dataset with the training data, using 3500 sequences as base. We are informed that we have 3500 sequences in the test set, so using this number replicates the approximate test set. The dataset is [here](https://www.kaggle.com/code/ravi20076/cmi2025-api-public-test-v1) for reference. I have also demonstrated how one may use this in the API to make a test **before the LB submission**.\n\nI also use 1750 sequences with IMU features only and make the other sensors as np.nan. This is mainly to test your submissions for silly mistakes. I think this is crucial to save your submissions from silly mistakes and wasted submission attempts. I urge beginners to use this dataset to test their code for issues beforehand prior to making a LB submission to avoid wasted efforts!\n\nAlso, this dataset could be used to test the **submission time** on the leaderboard. We could find a small difference between the actual time to submit and this dataset, as the sequence lengths could be slightly different across the datasets. \n\nWishing you the best! Regards!",
      "votes": null
    },
    {
      "id": "3226155",
      "postDate": "06/17/2025 10:18:34",
      "content": "<p><a href=\"https://www.kaggle.com/ravi20076\" target=\"_blank\">@ravi20076</a> one of the best thing about this competition is the submission time for cpu and gpu is 9hrs each unlike the birdclef competition which had 90 mins submission time hopefully wont suffer as many submission errors as I did suffer in birdclef competition</p>",
      "rawMarkdown": "ravi20076 one of the best thing about this competition is the submission time for cpu and gpu is 9hrs each unlike the birdclef competition which had 90 mins submission time hopefully wont suffer as many submission errors as I did suffer in birdclef competition",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3226155,
      "author_name": "sayedathar11",
      "author_url": "",
      "post_date": "06/17/2025 10:18:34",
      "content": "<p><a href=\"https://www.kaggle.com/ravi20076\" target=\"_blank\">@ravi20076</a> one of the best thing about this competition is the submission time for cpu and gpu is 9hrs each unlike the birdclef competition which had 90 mins submission time hopefully wont suffer as many submission errors as I did suffer in birdclef competition</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3223127": "Hello all,\n\nI curated a API tester dataset with the training data, using 3500 sequences as base. We are informed that we have 3500 sequences in the test set, so using this number replicates the approximate test set. The dataset is [here](https://www.kaggle.com/code/ravi20076/cmi2025-api-public-test-v1) for reference. I have also demonstrated how one may use this in the API to make a test **before the LB submission**.\n\nI also use 1750 sequences with IMU features only and make the other sensors as np.nan. This is mainly to test your submissions for silly mistakes. I think this is crucial to save your submissions from silly mistakes and wasted submission attempts. I urge beginners to use this dataset to test their code for issues beforehand prior to making a LB submission to avoid wasted efforts!\n\nAlso, this dataset could be used to test the **submission time** on the leaderboard. We could find a small difference between the actual time to submit and this dataset, as the sequence lengths could be slightly different across the datasets. \n\nWishing you the best! Regards!",
    "3226155": "ravi20076 one of the best thing about this competition is the submission time for cpu and gpu is 9hrs each unlike the birdclef competition which had 90 mins submission time hopefully wont suffer as many submission errors as I did suffer in birdclef competition"
  },
  "source": "meta"
}