{
  "id": 413798,
  "title": "How to train ? Seeking tips from fellow kagglers",
  "url": "/competitions/asl-fingerspelling/discussion/413798",
  "author_name": "Vinayak Tiwari",
  "post_date": "2023-05-30T06:06:23.531000",
  "votes": 1,
  "comment_count": 1,
  "views": 0,
  "content": "<p>This is my first competition of this particular category and I want to get advice and tips from experienced Kagglers.</p>\n<p>I can see that there are 53 landmark .parquet files with each file having data for certain no. of frames.<br>\nthe data is pretty distributed and I am a bit confused how to approach model training for such kind of a setup.</p>\n<p>Looking for genuine responses from the community </p>",
  "messages": [
    {
      "id": 2280475,
      "postDate": "2023-05-30T06:06:23.533Z",
      "content": "<p>This is my first competition of this particular category and I want to get advice and tips from experienced Kagglers.</p>\n<p>I can see that there are 53 landmark .parquet files with each file having data for certain no. of frames.<br>\nthe data is pretty distributed and I am a bit confused how to approach model training for such kind of a setup.</p>\n<p>Looking for genuine responses from the community </p>",
      "rawMarkdown": "This is my first competition of this particular category and I want to get advice and tips from experienced Kagglers.\n\nI can see that there are 53 landmark .parquet files with each file having data for certain no. of frames.\nthe data is pretty distributed and I am a bit confused how to approach model training for such kind of a setup.\n\nLooking for genuine responses from the community ",
      "votes": 1
    },
    {
      "id": 2283194,
      "postDate": "2023-06-01T05:49:38.650Z",
      "content": "<p>A simple approach will be to: first use the train.csv file to get the sequence id, the path and the phrase, you have to notice that each parquet file has more than one sequence meaning more than a single training example.</p>\n<p>Then you can encode the phrase to be ready for training, and preprocess the landmarks that you got by reading the parquet file, which is provided by \"path\", anyway you like.</p>\n<p>Hope I answered your question, feel free to ask further questions if there is any anything not clear.</p>",
      "rawMarkdown": "A simple approach will be to: first use the train.csv file to get the sequence id, the path and the phrase, you have to notice that each parquet file has more than one sequence meaning more than a single training example.\n\nThen you can encode the phrase to be ready for training, and preprocess the landmarks that you got by reading the parquet file, which is provided by \"path\", anyway you like.\n\nHope I answered your question, feel free to ask further questions if there is any anything not clear."
    }
  ],
  "comments": [
    {
      "id": 2283194,
      "author_name": "Fanatic Lizard",
      "author_url": "",
      "post_date": "2023-06-01T05:49:38.650000",
      "content": "<p>A simple approach will be to: first use the train.csv file to get the sequence id, the path and the phrase, you have to notice that each parquet file has more than one sequence meaning more than a single training example.</p>\n<p>Then you can encode the phrase to be ready for training, and preprocess the landmarks that you got by reading the parquet file, which is provided by \"path\", anyway you like.</p>\n<p>Hope I answered your question, feel free to ask further questions if there is any anything not clear.</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2280475": "This is my first competition of this particular category and I want to get advice and tips from experienced Kagglers.\n\nI can see that there are 53 landmark .parquet files with each file having data for certain no. of frames.\nthe data is pretty distributed and I am a bit confused how to approach model training for such kind of a setup.\n\nLooking for genuine responses from the community ",
    "2283194": "A simple approach will be to: first use the train.csv file to get the sequence id, the path and the phrase, you have to notice that each parquet file has more than one sequence meaning more than a single training example.\n\nThen you can encode the phrase to be ready for training, and preprocess the landmarks that you got by reading the parquet file, which is provided by \"path\", anyway you like.\n\nHope I answered your question, feel free to ask further questions if there is any anything not clear."
  }
}