{
  "id": 584237,
  "title": "How to Format Time Series Data for LSTM Input",
  "url": "/competitions/cmi-detect-behavior-with-sensor-data/discussion/584237",
  "author_name": "",
  "post_date": "2025-06-12T09:48:41.701463800Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Every time I start training my model on Kaggle, it runs out of memory. I'm currently using all the available features in my dataset, which might be contributing to the issue.<br>\nI’m looking for simple and effective ways to start training my model on Kaggle without encountering out-of-memory errors. What are the first things I should do to avoid this problem?</p>",
  "messages": [
    {
      "id": "3222610",
      "postDate": "06/12/2025 09:48:41",
      "content": "<p>Every time I start training my model on Kaggle, it runs out of memory. I'm currently using all the available features in my dataset, which might be contributing to the issue.<br>\nI’m looking for simple and effective ways to start training my model on Kaggle without encountering out-of-memory errors. What are the first things I should do to avoid this problem?</p>",
      "rawMarkdown": "Every time I start training my model on Kaggle, it runs out of memory. I'm currently using all the available features in my dataset, which might be contributing to the issue.\nI’m looking for simple and effective ways to start training my model on Kaggle without encountering out-of-memory errors. What are the first things I should do to avoid this problem?",
      "votes": null
    },
    {
      "id": "3222619",
      "postDate": "06/12/2025 09:57:17",
      "content": "<p>First of all, it would be good to know if you speak of memory or VRAM which is the memory on the GPU you use. And just as you already mentioned, best thing is to start small regarding: how many features you use, batch size, model size and so on. Usually you will then find what fits your hardware pretty quickly.</p>\n<p>Best,<br>\nJan</p>",
      "rawMarkdown": "First of all, it would be good to know if you speak of memory or VRAM which is the memory on the GPU you use. And just as you already mentioned, best thing is to start small regarding: how many features you use, batch size, model size and so on. Usually you will then find what fits your hardware pretty quickly.\n\nBest,\nJan",
      "votes": null
    },
    {
      "id": "3222632",
      "postDate": "06/12/2025 10:12:44",
      "content": "<p>Thanks for replying<br>\nI’m using a model where I first take all input features and apply masks to them for the null values. These masking operations cause the feature dimensions to double, and also result in a longer time sequence for each input.</p>\n<p>Since this is my first machine learning competition, I’m still getting comfortable with all the technical details thanks for your patience.</p>",
      "rawMarkdown": "Thanks for replying\nI’m using a model where I first take all input features and apply masks to them for the null values. These masking operations cause the feature dimensions to double, and also result in a longer time sequence for each input.\n\nSince this is my first machine learning competition, I’m still getting comfortable with all the technical details thanks for your patience.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3222619,
      "author_name": "janbrederecke",
      "author_url": "",
      "post_date": "06/12/2025 09:57:17",
      "content": "<p>First of all, it would be good to know if you speak of memory or VRAM which is the memory on the GPU you use. And just as you already mentioned, best thing is to start small regarding: how many features you use, batch size, model size and so on. Usually you will then find what fits your hardware pretty quickly.</p>\n<p>Best,<br>\nJan</p>",
      "votes": null,
      "replies": [
        {
          "id": 3222632,
          "author_name": "adityapushola",
          "author_url": "",
          "post_date": "06/12/2025 10:12:44",
          "content": "<p>Thanks for replying<br>\nI’m using a model where I first take all input features and apply masks to them for the null values. These masking operations cause the feature dimensions to double, and also result in a longer time sequence for each input.</p>\n<p>Since this is my first machine learning competition, I’m still getting comfortable with all the technical details thanks for your patience.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3222610": "Every time I start training my model on Kaggle, it runs out of memory. I'm currently using all the available features in my dataset, which might be contributing to the issue.\nI’m looking for simple and effective ways to start training my model on Kaggle without encountering out-of-memory errors. What are the first things I should do to avoid this problem?",
    "3222619": "First of all, it would be good to know if you speak of memory or VRAM which is the memory on the GPU you use. And just as you already mentioned, best thing is to start small regarding: how many features you use, batch size, model size and so on. Usually you will then find what fits your hardware pretty quickly.\n\nBest,\nJan",
    "3222632": "Thanks for replying\nI’m using a model where I first take all input features and apply masks to them for the null values. These masking operations cause the feature dimensions to double, and also result in a longer time sequence for each input.\n\nSince this is my first machine learning competition, I’m still getting comfortable with all the technical details thanks for your patience."
  },
  "source": "meta"
}