{
  "id": 447467,
  "title": "Notebook Run Successful - submission.csv created - But not scored",
  "url": "/competitions/UBC-OCEAN/discussion/447467",
  "author_name": "Samuel Abramov",
  "post_date": "2023-10-16T02:32:33.943000",
  "votes": 3,
  "comment_count": 2,
  "views": 0,
  "content": "<p>My submission has always status \"Failed\". Notebook finishes without errors and submission.csv is also created.<br>\nNothing indicates why its not score. Any Ideas? Do I need special config?</p>",
  "messages": [
    {
      "id": 2483813,
      "postDate": "2023-10-16T02:32:33.943Z",
      "content": "<p>My submission has always status \"Failed\". Notebook finishes without errors and submission.csv is also created.<br>\nNothing indicates why its not score. Any Ideas? Do I need special config?</p>",
      "rawMarkdown": "My submission has always status \"Failed\". Notebook finishes without errors and submission.csv is also created.\nNothing indicates why its not score. Any Ideas? Do I need special config?",
      "votes": 3
    },
    {
      "id": 2483927,
      "postDate": "2023-10-16T05:22:54.600Z",
      "content": "<p>It's most time due to CPU overuse or GPU during the real inferencing. I have noticed that some packages just fails due to using so much of resources while processing the hidden test data.</p>\n<p>I tried to mimic the testing situation by running the inference on the train data without error (not submitted)  but still got a failed submission during real inference with test data.</p>\n<p>Recommendations.</p>\n<ol>\n<li><p>Avoid using CPU or GPU intensive pre/post processing.. you may reduce batch side if that helps.</p></li>\n<li><p>Don't train during inference. The training data is not available during inferencing of your submitted code.</p></li>\n</ol>",
      "rawMarkdown": "It's most time due to CPU overuse or GPU during the real inferencing. I have noticed that some packages just fails due to using so much of resources while processing the hidden test data.\n\nI tried to mimic the testing situation by running the inference on the train data without error (not submitted)  but still got a failed submission during real inference with test data.\n\nRecommendations.\n\n1. Avoid using CPU or GPU intensive pre/post processing.. you may reduce batch side if that helps.\n\n2. Don't train during inference. The training data is not available during inferencing of your submitted code.",
      "votes": 4
    },
    {
      "id": 2484372,
      "postDate": "2023-10-16T12:45:27.520Z",
      "content": "<p>I added a EfficientNet b0 model via UI. Models -&gt; add model - choose b0. Is this step allowed?</p>",
      "rawMarkdown": "I added a EfficientNet b0 model via UI. Models -> add model - choose b0. Is this step allowed?"
    }
  ],
  "comments": [
    {
      "id": 2483927,
      "author_name": "HungryLearner",
      "author_url": "",
      "post_date": "2023-10-16T05:22:54.600000",
      "content": "<p>It's most time due to CPU overuse or GPU during the real inferencing. I have noticed that some packages just fails due to using so much of resources while processing the hidden test data.</p>\n<p>I tried to mimic the testing situation by running the inference on the train data without error (not submitted)  but still got a failed submission during real inference with test data.</p>\n<p>Recommendations.</p>\n<ol>\n<li><p>Avoid using CPU or GPU intensive pre/post processing.. you may reduce batch side if that helps.</p></li>\n<li><p>Don't train during inference. The training data is not available during inferencing of your submitted code.</p></li>\n</ol>",
      "votes": 4,
      "replies": []
    },
    {
      "id": 2484372,
      "author_name": "Samuel Abramov",
      "author_url": "",
      "post_date": "2023-10-16T12:45:27.520000",
      "content": "<p>I added a EfficientNet b0 model via UI. Models -&gt; add model - choose b0. Is this step allowed?</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2483813": "My submission has always status \"Failed\". Notebook finishes without errors and submission.csv is also created.\nNothing indicates why its not score. Any Ideas? Do I need special config?",
    "2483927": "It's most time due to CPU overuse or GPU during the real inferencing. I have noticed that some packages just fails due to using so much of resources while processing the hidden test data.\n\nI tried to mimic the testing situation by running the inference on the train data without error (not submitted)  but still got a failed submission during real inference with test data.\n\nRecommendations.\n\n1. Avoid using CPU or GPU intensive pre/post processing.. you may reduce batch side if that helps.\n\n2. Don't train during inference. The training data is not available during inferencing of your submitted code.",
    "2484372": "I added a EfficientNet b0 model via UI. Models -> add model - choose b0. Is this step allowed?"
  }
}