{
  "id": 480556,
  "title": "How the dataset constract?&How to cut the data?",
  "url": "/competitions/hms-harmful-brain-activity-classification/discussion/480556",
  "author_name": "franticXu",
  "post_date": "2024-02-29T03:32:02.093000",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>我不知道如何根据'eeg_label_offset_seconds'和'spectrogram_label_offset_seconds' 这两个值去剪切和对齐eeg和spectrogram。<br>\n我不太了解你们的数据是如何组织的。首先在train.csv表中的'eeg_label_offset_seconds'和'spectrogram_label_offset_seconds' 有什么作用？其后面的评价和投票时针对什么的，是整个eeg和时频图文件，还是针对'eeg_label_offset_seconds'和'spectrogram_label_offset_seconds' 标记的数据片段？然后是你们的eeg文件和时频图文件长度不统一，而示例pdf中的图像和eeg是标准时长的。我有这些问题的根本原因是同一份eeg文件和时频文件并不是一一对应的，而且同一个eeg文件有多个评价。<br>\n测试集的spectrogram文件是10分钟：300个数据点，所以说频谱图每一行的数据都是2秒钟的FFT变换结果，等同说频谱图的采样率是0.5？这也正好导致了每个时间偏移量均是偶数？</p>\n<p>I don't know how to cut and align eeg and spectrogram according to the values 'eeg_label_offset_seconds' and 'spectrogram_label_offset_seconds'.</p>\n<p>I don't know much about how your data is organized. First, what do 'eeg_label_offset_seconds' and 'spectrogram_label_offset_seconds' do in the train.csv table? What were the evaluations and votes were conducted on, the entire eeg and time-frequency graph files, or against the data fragments labeled 'eeg_label_offset_seconds' and 'spectrogram_label_offset_seconds'? The second is that your eeg file and the time-frequency graph file are not the same length, while the image and eeg in the sample pdf are the standard length. The root cause of my problems is that the same eeg file and the time-frequency file are not one-to-one correspondences, and there are multiple evaluations of the same eeg file.</p>\n<p>The spectrogram file of the test set is 10 minutes: 300 data points, so the data in each row of the spectrogram is the result of the FFT transformation of 2 seconds, which means that the sample rate of the spectrogram is 0.5. Which also happens to cause every time offset to be an even number?</p>",
  "messages": [
    {
      "id": 2674027,
      "postDate": "2024-02-29T03:32:02.093Z",
      "content": "<p>我不知道如何根据'eeg_label_offset_seconds'和'spectrogram_label_offset_seconds' 这两个值去剪切和对齐eeg和spectrogram。<br>\n我不太了解你们的数据是如何组织的。首先在train.csv表中的'eeg_label_offset_seconds'和'spectrogram_label_offset_seconds' 有什么作用？其后面的评价和投票时针对什么的，是整个eeg和时频图文件，还是针对'eeg_label_offset_seconds'和'spectrogram_label_offset_seconds' 标记的数据片段？然后是你们的eeg文件和时频图文件长度不统一，而示例pdf中的图像和eeg是标准时长的。我有这些问题的根本原因是同一份eeg文件和时频文件并不是一一对应的，而且同一个eeg文件有多个评价。<br>\n测试集的spectrogram文件是10分钟：300个数据点，所以说频谱图每一行的数据都是2秒钟的FFT变换结果，等同说频谱图的采样率是0.5？这也正好导致了每个时间偏移量均是偶数？</p>\n<p>I don't know how to cut and align eeg and spectrogram according to the values 'eeg_label_offset_seconds' and 'spectrogram_label_offset_seconds'.</p>\n<p>I don't know much about how your data is organized. First, what do 'eeg_label_offset_seconds' and 'spectrogram_label_offset_seconds' do in the train.csv table? What were the evaluations and votes were conducted on, the entire eeg and time-frequency graph files, or against the data fragments labeled 'eeg_label_offset_seconds' and 'spectrogram_label_offset_seconds'? The second is that your eeg file and the time-frequency graph file are not the same length, while the image and eeg in the sample pdf are the standard length. The root cause of my problems is that the same eeg file and the time-frequency file are not one-to-one correspondences, and there are multiple evaluations of the same eeg file.</p>\n<p>The spectrogram file of the test set is 10 minutes: 300 data points, so the data in each row of the spectrogram is the result of the FFT transformation of 2 seconds, which means that the sample rate of the spectrogram is 0.5. Which also happens to cause every time offset to be an even number?</p>",
      "rawMarkdown": "我不知道如何根据'eeg_label_offset_seconds'和'spectrogram_label_offset_seconds' 这两个值去剪切和对齐eeg和spectrogram。\n我不太了解你们的数据是如何组织的。首先在train.csv表中的'eeg_label_offset_seconds'和'spectrogram_label_offset_seconds' 有什么作用？其后面的评价和投票时针对什么的，是整个eeg和时频图文件，还是针对'eeg_label_offset_seconds'和'spectrogram_label_offset_seconds' 标记的数据片段？然后是你们的eeg文件和时频图文件长度不统一，而示例pdf中的图像和eeg是标准时长的。我有这些问题的根本原因是同一份eeg文件和时频文件并不是一一对应的，而且同一个eeg文件有多个评价。\n测试集的spectrogram文件是10分钟：300个数据点，所以说频谱图每一行的数据都是2秒钟的FFT变换结果，等同说频谱图的采样率是0.5？这也正好导致了每个时间偏移量均是偶数？\n\nI don't know how to cut and align eeg and spectrogram according to the values 'eeg_label_offset_seconds' and 'spectrogram_label_offset_seconds'.\n\nI don't know much about how your data is organized. First, what do 'eeg_label_offset_seconds' and 'spectrogram_label_offset_seconds' do in the train.csv table? What were the evaluations and votes were conducted on, the entire eeg and time-frequency graph files, or against the data fragments labeled 'eeg_label_offset_seconds' and 'spectrogram_label_offset_seconds'? The second is that your eeg file and the time-frequency graph file are not the same length, while the image and eeg in the sample pdf are the standard length. The root cause of my problems is that the same eeg file and the time-frequency file are not one-to-one correspondences, and there are multiple evaluations of the same eeg file.\n\nThe spectrogram file of the test set is 10 minutes: 300 data points, so the data in each row of the spectrogram is the result of the FFT transformation of 2 seconds, which means that the sample rate of the spectrogram is 0.5. Which also happens to cause every time offset to be an even number?\n"
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2674027": "我不知道如何根据'eeg_label_offset_seconds'和'spectrogram_label_offset_seconds' 这两个值去剪切和对齐eeg和spectrogram。\n我不太了解你们的数据是如何组织的。首先在train.csv表中的'eeg_label_offset_seconds'和'spectrogram_label_offset_seconds' 有什么作用？其后面的评价和投票时针对什么的，是整个eeg和时频图文件，还是针对'eeg_label_offset_seconds'和'spectrogram_label_offset_seconds' 标记的数据片段？然后是你们的eeg文件和时频图文件长度不统一，而示例pdf中的图像和eeg是标准时长的。我有这些问题的根本原因是同一份eeg文件和时频文件并不是一一对应的，而且同一个eeg文件有多个评价。\n测试集的spectrogram文件是10分钟：300个数据点，所以说频谱图每一行的数据都是2秒钟的FFT变换结果，等同说频谱图的采样率是0.5？这也正好导致了每个时间偏移量均是偶数？\n\nI don't know how to cut and align eeg and spectrogram according to the values 'eeg_label_offset_seconds' and 'spectrogram_label_offset_seconds'.\n\nI don't know much about how your data is organized. First, what do 'eeg_label_offset_seconds' and 'spectrogram_label_offset_seconds' do in the train.csv table? What were the evaluations and votes were conducted on, the entire eeg and time-frequency graph files, or against the data fragments labeled 'eeg_label_offset_seconds' and 'spectrogram_label_offset_seconds'? The second is that your eeg file and the time-frequency graph file are not the same length, while the image and eeg in the sample pdf are the standard length. The root cause of my problems is that the same eeg file and the time-frequency file are not one-to-one correspondences, and there are multiple evaluations of the same eeg file.\n\nThe spectrogram file of the test set is 10 minutes: 300 data points, so the data in each row of the spectrogram is the result of the FFT transformation of 2 seconds, which means that the sample rate of the spectrogram is 0.5. Which also happens to cause every time offset to be an even number?\n"
  }
}