{
  "id": 469241,
  "title": "CNN-1D starter notebbok LB-0.69 (Raw EEG) ",
  "url": "/competitions/hms-harmful-brain-activity-classification/discussion/469241",
  "author_name": "",
  "post_date": "2024-01-19T16:46:46.532087600Z",
  "votes": 18,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Created a PyTorch based CNN-1D model on raw EEG data .<br>\nnotebook link -- <a href=\"url\" target=\"_blank\">https://www.kaggle.com/code/anmolgarg1998/hms-cnn1d-inference</a></p>\n<p>Some of the things done are :</p>\n<ul>\n<li>Extracted 10000 points according to offset for each eeg_id as given in train.csv</li>\n<li>Passed through low and high pass filter of 1hz and 25hz respect.</li>\n<li>Downsampled to 50Hz and took mean of consecutive 4 points</li>\n<li>Took CNN-1D as raw eeg can be assumed as image in single dimension (Somebody pls confirm this)</li>\n<li>Took WeightedSampling to tackle the minor effect of data imbalance according to Expert_consensus</li>\n</ul>\n<p>Things Remaining :</p>\n<ul>\n<li>Not saved extracted and filtered EEG data which is consuming time during training .</li>\n<li>Have done training for 2 epochs only, will train more once after saving extracted EEG as that will save GPU training time</li>\n</ul>\n<p>Reference<br>\ntook reference from <a href=\"https://www.kaggle.com/cdeotte\" target=\"_blank\">@cdeotte</a> <br>\n<a href=\"https://www.kaggle.com/code/cdeotte/wavenet-starter-lb-0-66\" target=\"_blank\">https://www.kaggle.com/code/cdeotte/wavenet-starter-lb-0-66</a></p>",
  "messages": [
    {
      "id": "2609714",
      "postDate": "01/19/2024 16:46:46",
      "content": "<p>Created a PyTorch based CNN-1D model on raw EEG data .<br>\nnotebook link -- <a href=\"url\" target=\"_blank\">https://www.kaggle.com/code/anmolgarg1998/hms-cnn1d-inference</a></p>\n<p>Some of the things done are :</p>\n<ul>\n<li>Extracted 10000 points according to offset for each eeg_id as given in train.csv</li>\n<li>Passed through low and high pass filter of 1hz and 25hz respect.</li>\n<li>Downsampled to 50Hz and took mean of consecutive 4 points</li>\n<li>Took CNN-1D as raw eeg can be assumed as image in single dimension (Somebody pls confirm this)</li>\n<li>Took WeightedSampling to tackle the minor effect of data imbalance according to Expert_consensus</li>\n</ul>\n<p>Things Remaining :</p>\n<ul>\n<li>Not saved extracted and filtered EEG data which is consuming time during training .</li>\n<li>Have done training for 2 epochs only, will train more once after saving extracted EEG as that will save GPU training time</li>\n</ul>\n<p>Reference<br>\ntook reference from <a href=\"https://www.kaggle.com/cdeotte\" target=\"_blank\">@cdeotte</a> <br>\n<a href=\"https://www.kaggle.com/code/cdeotte/wavenet-starter-lb-0-66\" target=\"_blank\">https://www.kaggle.com/code/cdeotte/wavenet-starter-lb-0-66</a></p>",
      "rawMarkdown": "Created a PyTorch based CNN-1D model on raw EEG data .\nnotebook link -- [https://www.kaggle.com/code/anmolgarg1998/hms-cnn1d-inference](url)\n\nSome of the things done are :\n- Extracted 10000 points according to offset for each eeg_id as given in train.csv\n- Passed through low and high pass filter of 1hz and 25hz respect.\n- Downsampled to 50Hz and took mean of consecutive 4 points\n- Took CNN-1D as raw eeg can be assumed as image in single dimension (Somebody pls confirm this)\n- Took WeightedSampling to tackle the minor effect of data imbalance according to Expert_consensus\n\nThings Remaining :\n- Not saved extracted and filtered EEG data which is consuming time during training .\n- Have done training for 2 epochs only, will train more once after saving extracted EEG as that will save GPU training time\n\nReference\ntook reference from @cdeotte \nhttps://www.kaggle.com/code/cdeotte/wavenet-starter-lb-0-66",
      "votes": null
    },
    {
      "id": "2609943",
      "postDate": "01/19/2024 19:30:55",
      "content": "<p><a href=\"https://www.kaggle.com/anmolgarg1998\" target=\"_blank\">@anmolgarg1998</a> Thanks for sharing, Nice score with CNN-1D as i got LB:0.79</p>",
      "rawMarkdown": "anmolgarg1998 Thanks for sharing, Nice score with CNN-1D as i got LB:0.79",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2609943,
      "author_name": "seshurajup",
      "author_url": "",
      "post_date": "01/19/2024 19:30:55",
      "content": "<p><a href=\"https://www.kaggle.com/anmolgarg1998\" target=\"_blank\">@anmolgarg1998</a> Thanks for sharing, Nice score with CNN-1D as i got LB:0.79</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2609714": "Created a PyTorch based CNN-1D model on raw EEG data .\nnotebook link -- [https://www.kaggle.com/code/anmolgarg1998/hms-cnn1d-inference](url)\n\nSome of the things done are :\n- Extracted 10000 points according to offset for each eeg_id as given in train.csv\n- Passed through low and high pass filter of 1hz and 25hz respect.\n- Downsampled to 50Hz and took mean of consecutive 4 points\n- Took CNN-1D as raw eeg can be assumed as image in single dimension (Somebody pls confirm this)\n- Took WeightedSampling to tackle the minor effect of data imbalance according to Expert_consensus\n\nThings Remaining :\n- Not saved extracted and filtered EEG data which is consuming time during training .\n- Have done training for 2 epochs only, will train more once after saving extracted EEG as that will save GPU training time\n\nReference\ntook reference from @cdeotte \nhttps://www.kaggle.com/code/cdeotte/wavenet-starter-lb-0-66",
    "2609943": "anmolgarg1998 Thanks for sharing, Nice score with CNN-1D as i got LB:0.79"
  },
  "source": "meta"
}