{
  "id": 484433,
  "title": "Examples of GPD not like described",
  "url": "/competitions/hms-harmful-brain-activity-classification/discussion/484433",
  "author_name": "Idith Haber",
  "post_date": "2024-03-16T17:55:10.995000",
  "votes": 2,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Hi,</p>\n<p>I was wondering if anyone else noticed that many of the subjects labeled GPD, which should have periodic discharges , are very slow and variable over time?   This was not what I expected from watching the linked video and reading the linked paper.    Below is a 10s example (2000 samples) with the row information from the dataframe, followed by the raw subtracted channels in  blue, and then my filtered versions in green. ( The title headers are for the subplot below).  Perhaps I'm reading the file wrong by I copied and checked the code of a lot of people, including the pinned code. (HMS-Keras).  However, even the idealized example they give doesn't seem to fit.</p>\n<p>From text: <br>\n\"FIG. 20. Periodic Discharges (PDs).</p>\n<ol>\n<li>Repetition of a waveform with<br>\nrelatively uniform morphology and<br>\nduration, 2. with a clearly discernable<br>\ninterdischarge interval between<br>\nconsecutive waveforms, and 3.<br>\nrecurrence of the waveform at nearly<br>\nregular intervals: having a cycle length<br>\n(i.e., period) varying by ,50% from one<br>\ncycle to the next in the majority (.50%)<br>\nof cycle pairs. A pattern can qualify as<br>\nrhythmic or periodic if and only if it<br>\ncontinues for at least 6 cycles (e.g. 1 Hz<br>\nfor 6 seconds, or 3 Hz for 2 seconds)\"</li>\n</ol>\n<p>Also, Fp1-F7 has , for example have only 6 \"waveforms\" in 10 seconds.</p>\n<p>eeg_id                                                                     3725659500<br>\neeg_sub_id                                                                          0<br>\neeg_label_offset_seconds                                                          0.0<br>\nspectrogram_id                                                              433813419<br>\nspectrogram_sub_id                                                                  0<br>\nspectrogram_label_offset_seconds                                                  0.0<br>\nlabel_id                                                                   1567053066<br>\npatient_id                                                                      38549<br>\nexpert_consensus                                                                  GPD<br>\nMax_votes                                                                          15<br>\nTotal_vote                                                                         16<br>\nRatio                                                                          0.9375<br>\neeg_path                            /kaggle/input/hms-harmful-brain-activity-class…<br>\nspec_path                           /kaggle/input/hms-harmful-brain-activity-class…<br>\nspec2_path                          /tmp/dataset/hms-hbac/train_spectrograms/43381…<br>\nclass_name                                                                        GPD<br>\nclass_label                                                                         2<br>\nName: 22106, dtype: object<br>\nk,label 18 0 2 16 0.9375<br>\n/kaggle/input/hms-harmful-brain-activity-classification/train_eegs/3725659500.parquet</p>",
  "messages": [
    {
      "id": 2700857,
      "postDate": "2024-03-16T17:55:10.997Z",
      "content": "<p>Hi,</p>\n<p>I was wondering if anyone else noticed that many of the subjects labeled GPD, which should have periodic discharges , are very slow and variable over time?   This was not what I expected from watching the linked video and reading the linked paper.    Below is a 10s example (2000 samples) with the row information from the dataframe, followed by the raw subtracted channels in  blue, and then my filtered versions in green. ( The title headers are for the subplot below).  Perhaps I'm reading the file wrong by I copied and checked the code of a lot of people, including the pinned code. (HMS-Keras).  However, even the idealized example they give doesn't seem to fit.</p>\n<p>From text: <br>\n\"FIG. 20. Periodic Discharges (PDs).</p>\n<ol>\n<li>Repetition of a waveform with<br>\nrelatively uniform morphology and<br>\nduration, 2. with a clearly discernable<br>\ninterdischarge interval between<br>\nconsecutive waveforms, and 3.<br>\nrecurrence of the waveform at nearly<br>\nregular intervals: having a cycle length<br>\n(i.e., period) varying by ,50% from one<br>\ncycle to the next in the majority (.50%)<br>\nof cycle pairs. A pattern can qualify as<br>\nrhythmic or periodic if and only if it<br>\ncontinues for at least 6 cycles (e.g. 1 Hz<br>\nfor 6 seconds, or 3 Hz for 2 seconds)\"</li>\n</ol>\n<p>Also, Fp1-F7 has , for example have only 6 \"waveforms\" in 10 seconds.</p>\n<p>eeg_id                                                                     3725659500<br>\neeg_sub_id                                                                          0<br>\neeg_label_offset_seconds                                                          0.0<br>\nspectrogram_id                                                              433813419<br>\nspectrogram_sub_id                                                                  0<br>\nspectrogram_label_offset_seconds                                                  0.0<br>\nlabel_id                                                                   1567053066<br>\npatient_id                                                                      38549<br>\nexpert_consensus                                                                  GPD<br>\nMax_votes                                                                          15<br>\nTotal_vote                                                                         16<br>\nRatio                                                                          0.9375<br>\neeg_path                            /kaggle/input/hms-harmful-brain-activity-class…<br>\nspec_path                           /kaggle/input/hms-harmful-brain-activity-class…<br>\nspec2_path                          /tmp/dataset/hms-hbac/train_spectrograms/43381…<br>\nclass_name                                                                        GPD<br>\nclass_label                                                                         2<br>\nName: 22106, dtype: object<br>\nk,label 18 0 2 16 0.9375<br>\n/kaggle/input/hms-harmful-brain-activity-classification/train_eegs/3725659500.parquet</p>",
      "rawMarkdown": "Hi,\n\nI was wondering if anyone else noticed that many of the subjects labeled GPD, which should have periodic discharges , are very slow and variable over time?   This was not what I expected from watching the linked video and reading the linked paper.    Below is a 10s example (2000 samples) with the row information from the dataframe, followed by the raw subtracted channels in  blue, and then my filtered versions in green. ( The title headers are for the subplot below).  Perhaps I'm reading the file wrong by I copied and checked the code of a lot of people, including the pinned code. (HMS-Keras).  However, even the idealized example they give doesn't seem to fit.\n\nFrom text: \n\"FIG. 20. Periodic Discharges (PDs).\n1. Repetition of a waveform with\nrelatively uniform morphology and\nduration, 2. with a clearly discernable\ninterdischarge interval between\nconsecutive waveforms, and 3.\nrecurrence of the waveform at nearly\nregular intervals: having a cycle length\n(i.e., period) varying by ,50% from one\ncycle to the next in the majority (.50%)\nof cycle pairs. A pattern can qualify as\nrhythmic or periodic if and only if it\ncontinues for at least 6 cycles (e.g. 1 Hz\nfor 6 seconds, or 3 Hz for 2 seconds)\"\n\n\nAlso, Fp1-F7 has , for example have only 6 \"waveforms\" in 10 seconds.\n\neeg_id                                                                     3725659500\neeg_sub_id                                                                          0\neeg_label_offset_seconds                                                          0.0\nspectrogram_id                                                              433813419\nspectrogram_sub_id                                                                  0\nspectrogram_label_offset_seconds                                                  0.0\nlabel_id                                                                   1567053066\npatient_id                                                                      38549\nexpert_consensus                                                                  GPD\nMax_votes                                                                          15\nTotal_vote                                                                         16\nRatio                                                                          0.9375\neeg_path                            /kaggle/input/hms-harmful-brain-activity-class...\nspec_path                           /kaggle/input/hms-harmful-brain-activity-class...\nspec2_path                          /tmp/dataset/hms-hbac/train_spectrograms/43381...\nclass_name                                                                        GPD\nclass_label                                                                         2\nName: 22106, dtype: object\nk,label 18 0 2 16 0.9375\n/kaggle/input/hms-harmful-brain-activity-classification/train_eegs/3725659500.parquet\n\n\n\n",
      "votes": 2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2700857": "Hi,\n\nI was wondering if anyone else noticed that many of the subjects labeled GPD, which should have periodic discharges , are very slow and variable over time?   This was not what I expected from watching the linked video and reading the linked paper.    Below is a 10s example (2000 samples) with the row information from the dataframe, followed by the raw subtracted channels in  blue, and then my filtered versions in green. ( The title headers are for the subplot below).  Perhaps I'm reading the file wrong by I copied and checked the code of a lot of people, including the pinned code. (HMS-Keras).  However, even the idealized example they give doesn't seem to fit.\n\nFrom text: \n\"FIG. 20. Periodic Discharges (PDs).\n1. Repetition of a waveform with\nrelatively uniform morphology and\nduration, 2. with a clearly discernable\ninterdischarge interval between\nconsecutive waveforms, and 3.\nrecurrence of the waveform at nearly\nregular intervals: having a cycle length\n(i.e., period) varying by ,50% from one\ncycle to the next in the majority (.50%)\nof cycle pairs. A pattern can qualify as\nrhythmic or periodic if and only if it\ncontinues for at least 6 cycles (e.g. 1 Hz\nfor 6 seconds, or 3 Hz for 2 seconds)\"\n\n\nAlso, Fp1-F7 has , for example have only 6 \"waveforms\" in 10 seconds.\n\neeg_id                                                                     3725659500\neeg_sub_id                                                                          0\neeg_label_offset_seconds                                                          0.0\nspectrogram_id                                                              433813419\nspectrogram_sub_id                                                                  0\nspectrogram_label_offset_seconds                                                  0.0\nlabel_id                                                                   1567053066\npatient_id                                                                      38549\nexpert_consensus                                                                  GPD\nMax_votes                                                                          15\nTotal_vote                                                                         16\nRatio                                                                          0.9375\neeg_path                            /kaggle/input/hms-harmful-brain-activity-class...\nspec_path                           /kaggle/input/hms-harmful-brain-activity-class...\nspec2_path                          /tmp/dataset/hms-hbac/train_spectrograms/43381...\nclass_name                                                                        GPD\nclass_label                                                                         2\nName: 22106, dtype: object\nk,label 18 0 2 16 0.9375\n/kaggle/input/hms-harmful-brain-activity-classification/train_eegs/3725659500.parquet\n\n\n\n"
  }
}