{
  "id": 358036,
  "title": "Batch Inference",
  "url": "/competitions/g2net-detecting-continuous-gravitational-waves/discussion/358036",
  "author_name": "Marco Ciavarella",
  "post_date": "2022-10-06T11:51:41.068000",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hey all.<br>\nThanks to the input from <a href=\"https://www.kaggle.com/brandenkmurray\" target=\"_blank\">@brandenkmurray</a> I was able to circumvent the memory limitations and make predictions on the whole test dataset (only using the data from the H1 interferometer).<br>\nThe solution I adopted is predicting using small batches of input data and then using the del statement and gc.collect() to remove the batch data from memory.</p>\n<p>Batch Inference Notebook: <a href=\"https://www.kaggle.com/code/chazzer/neural-net-starter-g2net-infer\" target=\"_blank\">https://www.kaggle.com/code/chazzer/neural-net-starter-g2net-infer</a></p>\n<p>The model's performance is still really bad though lol</p>",
  "messages": [
    {
      "id": 1975003,
      "postDate": "2022-10-06T15:16:01.673Z",
      "content": "<p>That shows… We are not on right path… I have tried the same. I think more work is to be done to get data for deep learning, and H1 and L1 ensemble. <br>\nStill trying ways to prepare data. </p>",
      "rawMarkdown": "That shows... We are not on right path... I have tried the same. I think more work is to be done to get data for deep learning, and H1 and L1 ensemble. \nStill trying ways to prepare data. ",
      "votes": 1,
      "replies": [
        {
          "id": 1975008,
          "postDate": "2022-10-06T15:18:45.530Z",
          "content": "<p>Yup, this is looking like a really tough competition. <br>\nMy model is not able to extract useful features from the input data for now.</p>",
          "rawMarkdown": "Yup, this is looking like a really tough competition. \nMy model is not able to extract useful features from the input data for now.",
          "votes": 1
        }
      ]
    },
    {
      "id": 1974697,
      "postDate": "2022-10-06T11:51:41.070Z",
      "content": "<p>Hey all.<br>\nThanks to the input from <a href=\"https://www.kaggle.com/brandenkmurray\" target=\"_blank\">@brandenkmurray</a> I was able to circumvent the memory limitations and make predictions on the whole test dataset (only using the data from the H1 interferometer).<br>\nThe solution I adopted is predicting using small batches of input data and then using the del statement and gc.collect() to remove the batch data from memory.</p>\n<p>Batch Inference Notebook: <a href=\"https://www.kaggle.com/code/chazzer/neural-net-starter-g2net-infer\" target=\"_blank\">https://www.kaggle.com/code/chazzer/neural-net-starter-g2net-infer</a></p>\n<p>The model's performance is still really bad though lol</p>",
      "rawMarkdown": "Hey all.\nThanks to the input from @brandenkmurray I was able to circumvent the memory limitations and make predictions on the whole test dataset (only using the data from the H1 interferometer).\nThe solution I adopted is predicting using small batches of input data and then using the del statement and gc.collect() to remove the batch data from memory.\n\nBatch Inference Notebook: https://www.kaggle.com/code/chazzer/neural-net-starter-g2net-infer\n\nThe model's performance is still really bad though lol",
      "votes": 1
    }
  ],
  "comments": [
    {
      "id": 1975003,
      "author_name": "Bibhabasu Mohapatra",
      "author_url": "",
      "post_date": "2022-10-06T15:16:01.673000",
      "content": "<p>That shows… We are not on right path… I have tried the same. I think more work is to be done to get data for deep learning, and H1 and L1 ensemble. <br>\nStill trying ways to prepare data. </p>",
      "votes": 1,
      "replies": [
        {
          "id": 1975008,
          "author_name": "Marco Ciavarella",
          "author_url": "",
          "post_date": "2022-10-06T15:18:45.530000",
          "content": "<p>Yup, this is looking like a really tough competition. <br>\nMy model is not able to extract useful features from the input data for now.</p>",
          "votes": 1,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1975003": "That shows... We are not on right path... I have tried the same. I think more work is to be done to get data for deep learning, and H1 and L1 ensemble. \nStill trying ways to prepare data. ",
    "1974697": "Hey all.\nThanks to the input from @brandenkmurray I was able to circumvent the memory limitations and make predictions on the whole test dataset (only using the data from the H1 interferometer).\nThe solution I adopted is predicting using small batches of input data and then using the del statement and gc.collect() to remove the batch data from memory.\n\nBatch Inference Notebook: https://www.kaggle.com/code/chazzer/neural-net-starter-g2net-infer\n\nThe model's performance is still really bad though lol"
  }
}