![](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3223907%2F4e408a0761be2a8fe510d77cb5e6bf2c%2FScreenshot%202024-02-22%20at%2012.40.41PM.png?generation=1708634495240640&alt=media\")
I've found some EEG signals with labeled seizures online (hoping to release the processed data soon). I was able to resample the data to 200 Hz, then create a 10-minute spectrogram. Below is the code I used to create the spectrograms. When I select frequencies between 0.5 and 20 Hz, I get nearly perfect agreement with the frequencies in the competition spectrogram column names (LL_0.59, LL_0.78, LL_0.98, etc), giving dimensions 100x300 for each spectrogram. However, a few rounded frequencies do not align with column names, suggesting my parameters might be slightly different than those used to create competition spectrograms.
\nWhile the slight difference in frequency is not a huge concern, I would like the 10-minute spectrograms to resemble the competition spectrograms as best as possible. Does anyone have a suggestion for modifying the code below for better agreement?
\n scipy.signal spectrogram\n\nfrequencies, times, Sxx = spectrogram(x, fs=sr, window=, nperseg=, noverlap=, nfft=, detrend=, scaling=, mode=)\nCompetition spectrogram:
\n
My spectrograms:
\n![](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3223907%2Fe26632f64b94aaf90d37ded40c11cd71%2FScreenshot%202024-02-22%20at%2012.40.32PM.png?generation=1708634519026709&alt=media\")
I've found some EEG signals with labeled seizures online (hoping to release the processed data soon). I was able to resample the data to 200 Hz, then create a 10-minute spectrogram. Below is the code I used to create the spectrograms. When I select frequencies between 0.5 and 20 Hz, I get nearly perfect agreement with the frequencies in the competition spectrogram column names (LL_0.59, LL_0.78, LL_0.98, etc), giving dimensions 100x300 for each spectrogram. However, a few rounded frequencies do not align with column names, suggesting my parameters might be slightly different than those used to create competition spectrograms.
\nWhile the slight difference in frequency is not a huge concern, I would like the 10-minute spectrograms to resemble the competition spectrograms as best as possible. Does anyone have a suggestion for modifying the code below for better agreement?
\n scipy.signal spectrogram\n\nfrequencies, times, Sxx = spectrogram(x, fs=sr, window=, nperseg=, noverlap=, nfft=, detrend=, scaling=, mode=)\nCompetition spectrogram:
\n
My spectrograms:
\n
Your work is impressive!
\nI'm particularly impressed by how similar your generated spectrums are to the competition's. I will take some time (if possible) to test your spectrums and see if they boost performance.
", "rawMarkdown": "Your work is impressive! \n\nI'm particularly impressed by how similar your generated spectrums are to the competition's. I will take some time (if possible) to test your spectrums and see if they boost performance.", "votes": null }, { "id": "2664453", "postDate": "02/23/2024 03:00:54", "content": "Nice job, they look similar. How do you create the LL montage? Do you create 4 spectrograms for the 4 differences then average those 4 spectrograms to make 1 spectrogram like my formula here
", "rawMarkdown": "Nice job, they look similar. How do you create the LL montage? Do you create 4 spectrograms for the 4 differences then average those 4 spectrograms to make 1 spectrogram like my formula [here][1]\n\n[1]: https://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/discussion/469760", "votes": null }, { "id": "2664518", "postDate": "02/23/2024 04:53:57", "content": "Thanks! Yes, I modified your code to create the spectrograms. Thanks again for sharing your insights. I'll publish this dataset soon
", "rawMarkdown": "Thanks! Yes, I modified your code to create the spectrograms. Thanks again for sharing your insights. I'll publish this dataset soon", "votes": null }, { "id": "2664916", "postDate": "02/23/2024 09:42:08", "content": "Hey, a query, what do you mean by non competition EEG signals? Are you telling there is supplementary data available?
", "rawMarkdown": "Hey, a query, what do you mean by non competition EEG signals? Are you telling there is supplementary data available?", "votes": null }, { "id": "2665218", "postDate": "02/23/2024 14:26:38", "content": "You can find various EEG datasets online, but not all of them relate to harmful brain activity
", "rawMarkdown": "You can find various EEG datasets online, but not all of them relate to harmful brain activity", "votes": null } ], "comments": [ { "id": 2664192, "author_name": "minhsienweng", "author_url": "", "post_date": "02/22/2024 21:10:37", "content": "Your work is impressive!
\nI'm particularly impressed by how similar your generated spectrums are to the competition's. I will take some time (if possible) to test your spectrums and see if they boost performance.
", "votes": null, "replies": [] }, { "id": 2664453, "author_name": "cdeotte", "author_url": "", "post_date": "02/23/2024 03:00:54", "content": "Nice job, they look similar. How do you create the LL montage? Do you create 4 spectrograms for the 4 differences then average those 4 spectrograms to make 1 spectrogram like my formula here
", "votes": null, "replies": [ { "id": 2664518, "author_name": "seanbearden", "author_url": "", "post_date": "02/23/2024 04:53:57", "content": "Thanks! Yes, I modified your code to create the spectrograms. Thanks again for sharing your insights. I'll publish this dataset soon
", "votes": null, "replies": [] } ] }, { "id": 2664916, "author_name": "chaudharypriyanshu", "author_url": "", "post_date": "02/23/2024 09:42:08", "content": "Hey, a query, what do you mean by non competition EEG signals? Are you telling there is supplementary data available?
", "votes": null, "replies": [ { "id": 2665218, "author_name": "seanbearden", "author_url": "", "post_date": "02/23/2024 14:26:38", "content": "You can find various EEG datasets online, but not all of them relate to harmful brain activity
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "2664155": "I've found some EEG signals with labeled seizures online (hoping to release the processed data soon). I was able to resample the data to 200 Hz, then create a 10-minute spectrogram. Below is the code I used to create the spectrograms. When I select frequencies between 0.5 and 20 Hz, I get nearly perfect agreement with the frequencies in the competition spectrogram column names (LL_0.59, LL_0.78, LL_0.98, etc), giving dimensions 100x300 for each spectrogram. However, a few rounded frequencies do not align with column names, suggesting my parameters might be slightly different than those used to create competition spectrograms.\n\nWhile the slight difference in frequency is not a huge concern, I would like the 10-minute spectrograms to resemble the competition spectrograms as best as possible. Does anyone have a suggestion for modifying the code below for better agreement?\n\n\n```python\nfrom scipy.signal import spectrogram\n\nfrequencies, times, Sxx = spectrogram(x, fs=sr, window='hann', nperseg=400, noverlap=0, nfft=1024, detrend=False, scaling='density', mode='magnitude')\n```\nCompetition spectrogram:\n\n\n\nMy spectrograms:\n\n", "2664192": "Your work is impressive! \n\nI'm particularly impressed by how similar your generated spectrums are to the competition's. I will take some time (if possible) to test your spectrums and see if they boost performance.", "2664453": "Nice job, they look similar. How do you create the LL montage? Do you create 4 spectrograms for the 4 differences then average those 4 spectrograms to make 1 spectrogram like my formula [here][1]\n\n[1]: https://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/discussion/469760", "2664518": "Thanks! Yes, I modified your code to create the spectrograms. Thanks again for sharing your insights. I'll publish this dataset soon", "2664916": "Hey, a query, what do you mean by non competition EEG signals? Are you telling there is supplementary data available?", "2665218": "You can find various EEG datasets online, but not all of them relate to harmful brain activity" }, "source": "meta" }