{
  "id": 468388,
  "title": "Learning from the EEG signals",
  "url": "/competitions/hms-harmful-brain-activity-classification/discussion/468388",
  "author_name": "",
  "post_date": "2024-01-16T13:31:20.251336300Z",
  "votes": 5,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I have attempted to learn from the EEG signals themselves rather than the spectrogram using <a href=\"https://github.com/vlawhern/arl-eegmodels\" target=\"_blank\">https://github.com/vlawhern/arl-eegmodels</a>, specifically the EEGNet model <a href=\"https://arxiv.org/abs/1611.08024\" target=\"_blank\">https://arxiv.org/abs/1611.08024</a></p>\n<p><a href=\"https://www.kaggle.com/code/ybbbat/learning-from-eeg-signals-using-arl-eegmodels\" target=\"_blank\">https://www.kaggle.com/code/ybbbat/learning-from-eeg-signals-using-arl-eegmodels</a></p>\n<p>Unfortunately I have had little success getting good results, the model doesn't seem to be learning very well despite the success of the original paper, my intuition is this is because the original paper may be using much smaller windows of signals than the 50 seconds of 200Hz samples we have here.</p>\n<p>Has anyone had any success with similar techniques?</p>\n<p>Either way, the code may be useful for anyone else trying to figure out how to use the eeg signals with keras models.</p>",
  "messages": [
    {
      "id": "2604533",
      "postDate": "01/16/2024 13:31:20",
      "content": "<p>I have attempted to learn from the EEG signals themselves rather than the spectrogram using <a href=\"https://github.com/vlawhern/arl-eegmodels\" target=\"_blank\">https://github.com/vlawhern/arl-eegmodels</a>, specifically the EEGNet model <a href=\"https://arxiv.org/abs/1611.08024\" target=\"_blank\">https://arxiv.org/abs/1611.08024</a></p>\n<p><a href=\"https://www.kaggle.com/code/ybbbat/learning-from-eeg-signals-using-arl-eegmodels\" target=\"_blank\">https://www.kaggle.com/code/ybbbat/learning-from-eeg-signals-using-arl-eegmodels</a></p>\n<p>Unfortunately I have had little success getting good results, the model doesn't seem to be learning very well despite the success of the original paper, my intuition is this is because the original paper may be using much smaller windows of signals than the 50 seconds of 200Hz samples we have here.</p>\n<p>Has anyone had any success with similar techniques?</p>\n<p>Either way, the code may be useful for anyone else trying to figure out how to use the eeg signals with keras models.</p>",
      "rawMarkdown": "I have attempted to learn from the EEG signals themselves rather than the spectrogram using https://github.com/vlawhern/arl-eegmodels, specifically the EEGNet model https://arxiv.org/abs/1611.08024\n\nhttps://www.kaggle.com/code/ybbbat/learning-from-eeg-signals-using-arl-eegmodels\n\nUnfortunately I have had little success getting good results, the model doesn't seem to be learning very well despite the success of the original paper, my intuition is this is because the original paper may be using much smaller windows of signals than the 50 seconds of 200Hz samples we have here.\n\nHas anyone had any success with similar techniques?\n\nEither way, the code may be useful for anyone else trying to figure out how to use the eeg signals with keras models.",
      "votes": null
    },
    {
      "id": "2604552",
      "postDate": "01/16/2024 13:47:21",
      "content": "<p><a href=\"https://www.kaggle.com/ybbbat\" target=\"_blank\">@ybbbat</a> i'm working on wavnet with EEG signals, will share if i made good progress. </p>\n<p>Top public notebooks are using spectrograms as alternative and EEG -&gt; spectorgram also achieving better scores. go through this <a href=\"https://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/discussion/467877\" target=\"_blank\">Discussion</a></p>\n<p>As per public notebooks, spectorgram better in classifying targets</p>\n<p>Thanks for sharing, i will go through the paper and github</p>\n<pre><code> EEGModels  EEGNet, ShallowConvNet, DeepConvNet\n\nmodel  = EEGNet(nb_classes = ..., Chans = ..., Samples = ...)\n\nmodel2 = ShallowConvNet(nb_classes = ..., Chans = ..., Samples = ...)\n\nmodel3 = DeepConvNet(nb_classes = ..., Chans = ..., Samples = ...)\n</code></pre>",
      "rawMarkdown": "ybbbat i'm working on wavnet with EEG signals, will share if i made good progress. \n\nTop public notebooks are using spectrograms as alternative and EEG -> spectorgram also achieving better scores. go through this [Discussion](https://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/discussion/467877)\n\nAs per public notebooks, spectorgram better in classifying targets\n\nThanks for sharing, i will go through the paper and github\n```python\nfrom EEGModels import EEGNet, ShallowConvNet, DeepConvNet\n\nmodel  = EEGNet(nb_classes = ..., Chans = ..., Samples = ...)\n\nmodel2 = ShallowConvNet(nb_classes = ..., Chans = ..., Samples = ...)\n\nmodel3 = DeepConvNet(nb_classes = ..., Chans = ..., Samples = ...)\n```",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2604552,
      "author_name": "seshurajup",
      "author_url": "",
      "post_date": "01/16/2024 13:47:21",
      "content": "<p><a href=\"https://www.kaggle.com/ybbbat\" target=\"_blank\">@ybbbat</a> i'm working on wavnet with EEG signals, will share if i made good progress. </p>\n<p>Top public notebooks are using spectrograms as alternative and EEG -&gt; spectorgram also achieving better scores. go through this <a href=\"https://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/discussion/467877\" target=\"_blank\">Discussion</a></p>\n<p>As per public notebooks, spectorgram better in classifying targets</p>\n<p>Thanks for sharing, i will go through the paper and github</p>\n<pre><code> EEGModels  EEGNet, ShallowConvNet, DeepConvNet\n\nmodel  = EEGNet(nb_classes = ..., Chans = ..., Samples = ...)\n\nmodel2 = ShallowConvNet(nb_classes = ..., Chans = ..., Samples = ...)\n\nmodel3 = DeepConvNet(nb_classes = ..., Chans = ..., Samples = ...)\n</code></pre>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2604533": "I have attempted to learn from the EEG signals themselves rather than the spectrogram using https://github.com/vlawhern/arl-eegmodels, specifically the EEGNet model https://arxiv.org/abs/1611.08024\n\nhttps://www.kaggle.com/code/ybbbat/learning-from-eeg-signals-using-arl-eegmodels\n\nUnfortunately I have had little success getting good results, the model doesn't seem to be learning very well despite the success of the original paper, my intuition is this is because the original paper may be using much smaller windows of signals than the 50 seconds of 200Hz samples we have here.\n\nHas anyone had any success with similar techniques?\n\nEither way, the code may be useful for anyone else trying to figure out how to use the eeg signals with keras models.",
    "2604552": "ybbbat i'm working on wavnet with EEG signals, will share if i made good progress. \n\nTop public notebooks are using spectrograms as alternative and EEG -> spectorgram also achieving better scores. go through this [Discussion](https://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/discussion/467877)\n\nAs per public notebooks, spectorgram better in classifying targets\n\nThanks for sharing, i will go through the paper and github\n```python\nfrom EEGModels import EEGNet, ShallowConvNet, DeepConvNet\n\nmodel  = EEGNet(nb_classes = ..., Chans = ..., Samples = ...)\n\nmodel2 = ShallowConvNet(nb_classes = ..., Chans = ..., Samples = ...)\n\nmodel3 = DeepConvNet(nb_classes = ..., Chans = ..., Samples = ...)\n```"
  },
  "source": "meta"
}