{
  "id": 54430,
  "title": "Batch Size for training of dataset",
  "url": "/competitions/talkingdata-adtracking-fraud-detection/discussion/54430",
  "author_name": "NikhilVerma",
  "post_date": "2018-04-13T06:02:59.043000",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Can somebody help me with choosing an appropriate batch size for my epochs? As this is large dataset It takes a lot of time in training, When choosing 2048*64 as batch size for training, each epoch take 30m on 16-core-cpu/64GB ram (no GPU) machine, I think this is a high batch size for initial epochs as I read somewhere somewhere that I should use low batch size for initial epochs and high for later epochs, so that my model converges better</p>",
  "messages": [
    {
      "id": 313341,
      "postDate": "2018-04-13T06:02:59.043Z",
      "content": "<p>Can somebody help me with choosing an appropriate batch size for my epochs? As this is large dataset It takes a lot of time in training, When choosing 2048*64 as batch size for training, each epoch take 30m on 16-core-cpu/64GB ram (no GPU) machine, I think this is a high batch size for initial epochs as I read somewhere somewhere that I should use low batch size for initial epochs and high for later epochs, so that my model converges better</p>",
      "rawMarkdown": "Can somebody help me with choosing an appropriate batch size for my epochs? As this is large dataset It takes a lot of time in training, When choosing 2048*64 as batch size for training, each epoch take 30m on 16-core-cpu/64GB ram (no GPU) machine, I think this is a high batch size for initial epochs as I read somewhere somewhere that I should use low batch size for initial epochs and high for later epochs, so that my model converges better"
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "313341": "Can somebody help me with choosing an appropriate batch size for my epochs? As this is large dataset It takes a lot of time in training, When choosing 2048*64 as batch size for training, each epoch take 30m on 16-core-cpu/64GB ram (no GPU) machine, I think this is a high batch size for initial epochs as I read somewhere somewhere that I should use low batch size for initial epochs and high for later epochs, so that my model converges better"
  }
}