{
  "id": 266945,
  "title": "How to implement pseudo label code on pytorch",
  "url": "/competitions/seti-breakthrough-listen/discussion/266945",
  "author_name": "Jackie Mai",
  "post_date": "2021-08-21T01:32:03.969000",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I have heard someone said the pseudo label method could improve score on this competition. I am curious how to implement the code? maybe take a simple example i can learn! Thanks you!</p>",
  "messages": [
    {
      "id": 1483944,
      "postDate": "2021-08-21T02:04:45.907Z",
      "content": "<p>One simple way to use pseudo labels in PyTorch is with ‘torch.utils.data.ConcatDataset’. So you create two datasets, one with original data and the other with your pseudo labeled test data, then you concatenate them before training.</p>\n<p>Another option is to simply pre-train your model with pseudo labels, and then finetune it on the given training data. Meaning you first train a model on pseudo labels, save the weights, and then use these weights as your starting point when training with the regular data. Hope this helps. </p>",
      "rawMarkdown": "One simple way to use pseudo labels in PyTorch is with ‘torch.utils.data.ConcatDataset’. So you create two datasets, one with original data and the other with your pseudo labeled test data, then you concatenate them before training.\n\nAnother option is to simply pre-train your model with pseudo labels, and then finetune it on the given training data. Meaning you first train a model on pseudo labels, save the weights, and then use these weights as your starting point when training with the regular data. Hope this helps. ",
      "votes": 1,
      "replies": [
        {
          "id": 1483982,
          "postDate": "2021-08-21T02:48:03.717Z",
          "content": "<p>Thank you </p>",
          "rawMarkdown": "Thank you ",
          "votes": 1
        }
      ]
    },
    {
      "id": 1483925,
      "postDate": "2021-08-21T01:32:03.970Z",
      "content": "<p>I have heard someone said the pseudo label method could improve score on this competition. I am curious how to implement the code? maybe take a simple example i can learn! Thanks you!</p>",
      "rawMarkdown": "I have heard someone said the pseudo label method could improve score on this competition. I am curious how to implement the code? maybe take a simple example i can learn! Thanks you!",
      "votes": 1
    }
  ],
  "comments": [
    {
      "id": 1483944,
      "author_name": "Tucker Arrants",
      "author_url": "",
      "post_date": "2021-08-21T02:04:45.907000",
      "content": "<p>One simple way to use pseudo labels in PyTorch is with ‘torch.utils.data.ConcatDataset’. So you create two datasets, one with original data and the other with your pseudo labeled test data, then you concatenate them before training.</p>\n<p>Another option is to simply pre-train your model with pseudo labels, and then finetune it on the given training data. Meaning you first train a model on pseudo labels, save the weights, and then use these weights as your starting point when training with the regular data. Hope this helps. </p>",
      "votes": 1,
      "replies": [
        {
          "id": 1483982,
          "author_name": "Jackie Mai",
          "author_url": "",
          "post_date": "2021-08-21T02:48:03.717000",
          "content": "<p>Thank you </p>",
          "votes": 1,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1483944": "One simple way to use pseudo labels in PyTorch is with ‘torch.utils.data.ConcatDataset’. So you create two datasets, one with original data and the other with your pseudo labeled test data, then you concatenate them before training.\n\nAnother option is to simply pre-train your model with pseudo labels, and then finetune it on the given training data. Meaning you first train a model on pseudo labels, save the weights, and then use these weights as your starting point when training with the regular data. Hope this helps. ",
    "1483925": "I have heard someone said the pseudo label method could improve score on this competition. I am curious how to implement the code? maybe take a simple example i can learn! Thanks you!"
  }
}