Hello Everyone. I've always wanted to try and predict future time-frequency data and then I came across this dataset and I wanted to actually implement it. I've looked at a few tutorials on how LSTM data modeling works like having a constant window size and a set prediction quantity, but I have a hard understanding this HDF5 data. Open to any clarification!
\nThanks,
\nNaren
", "messages": [ { "id": 2022143, "postDate": "2022-11-08T18:34:47.487Z", "content": "Check out some of the nice notebooks under Code to learn more about the data, e.g:
https://www.kaggle.com/code/ayuraj/g2net-understand-the-data
", "rawMarkdown": "Check out some of the nice notebooks under `Code` to learn more about the data, e.g:\n\nhttps://www.kaggle.com/code/ayuraj/g2net-understand-the-data", "votes": 2, "replies": [ { "id": 2022162, "postDate": "2022-11-08T19:10:26.933Z", "content": "Thank you so much for your reply and I'll look into it. I've actually found an LSTM implementation for this competition too https://www.kaggle.com/code/essammohamed4320/g2net-time-lstm-vs-frequency-vit-cnn-domain#Installing-and-importing-necessary-dependancies.
", "rawMarkdown": "Thank you so much for your reply and I'll look into it. I've actually found an LSTM implementation for this competition too [https://www.kaggle.com/code/essammohamed4320/g2net-time-lstm-vs-frequency-vit-cnn-domain#Installing-and-importing-necessary-dependancies](url). " }, { "id": 2022214, "postDate": "2022-11-08T20:05:57.703Z", "content": "If you look at the results of the run carefully, I don't think lstm learned anything.
\nThe data we're dealing with is not sequential, and I don't think it's possible to be able to apply a sequential model directly.
\nUnless you convert the data back to time domain data, the
\nhttps://www.kaggle.com/competitions/g2net-detecting-continuous-gravitational-waves/discussion/358835
\nOthers have discussed that direction here.
Hmm. The author did actually create a time-domain dataset from the spectrogram datasets: \"The STFT is invertible, that is, the original signal can be recovered from the transform by the inverse STFT…\" The notebook itself doesn't have any real predictions made from the time-domain data however, which could be tested.
", "rawMarkdown": "Hmm. The author did actually create a time-domain dataset from the spectrogram datasets: \"The STFT is invertible, that is, the original signal can be recovered from the transform by the inverse STFT...\" The notebook itself doesn't have any real predictions made from the time-domain data however, which could be tested. \n\n", "votes": -1 }, { "id": 2027652, "postDate": "2022-11-13T04:09:26.560Z", "content": "Nevermind…you're right. The fit method for the LSTM uses the g-wave-presence-value as its label, which doesn't even make sense for an LSTM.
", "rawMarkdown": "Nevermind...you're right. The fit method for the LSTM uses the g-wave-presence-value as its label, which doesn't even make sense for an LSTM. ", "votes": 1 } ] }, { "id": 2019491, "postDate": "2022-11-06T16:31:39.360Z", "content": "Hello Everyone. I've always wanted to try and predict future time-frequency data and then I came across this dataset and I wanted to actually implement it. I've looked at a few tutorials on how LSTM data modeling works like having a constant window size and a set prediction quantity, but I have a hard understanding this HDF5 data. Open to any clarification!
\nThanks,
\nNaren
", "rawMarkdown": "Hello Everyone. I've always wanted to try and predict future time-frequency data and then I came across this dataset and I wanted to actually implement it. I've looked at a few tutorials on how LSTM data modeling works like having a constant window size and a set prediction quantity, but I have a hard understanding this HDF5 data. Open to any clarification!\n\nThanks,\n\nNaren" } ], "comments": [ { "id": 2022143, "author_name": "Ali Abdin", "author_url": "", "post_date": "2022-11-08T18:34:47.487000", "content": "Check out some of the nice notebooks under Code to learn more about the data, e.g:
https://www.kaggle.com/code/ayuraj/g2net-understand-the-data
", "votes": 2, "replies": [ { "id": 2022162, "author_name": "Naren Manikandan", "author_url": "", "post_date": "2022-11-08T19:10:26.933000", "content": "Thank you so much for your reply and I'll look into it. I've actually found an LSTM implementation for this competition too https://www.kaggle.com/code/essammohamed4320/g2net-time-lstm-vs-frequency-vit-cnn-domain#Installing-and-importing-necessary-dependancies.
", "votes": 0, "replies": [] }, { "id": 2022214, "author_name": "Chen Lin", "author_url": "", "post_date": "2022-11-08T20:05:57.703000", "content": "If you look at the results of the run carefully, I don't think lstm learned anything.
\nThe data we're dealing with is not sequential, and I don't think it's possible to be able to apply a sequential model directly.
\nUnless you convert the data back to time domain data, the
\nhttps://www.kaggle.com/competitions/g2net-detecting-continuous-gravitational-waves/discussion/358835
\nOthers have discussed that direction here.
Hmm. The author did actually create a time-domain dataset from the spectrogram datasets: \"The STFT is invertible, that is, the original signal can be recovered from the transform by the inverse STFT…\" The notebook itself doesn't have any real predictions made from the time-domain data however, which could be tested.
", "votes": -1, "replies": [] }, { "id": 2027652, "author_name": "Naren Manikandan", "author_url": "", "post_date": "2022-11-13T04:09:26.560000", "content": "Nevermind…you're right. The fit method for the LSTM uses the g-wave-presence-value as its label, which doesn't even make sense for an LSTM.
", "votes": 1, "replies": [] } ] } ], "raw_markdown_by_id": { "2022143": "Check out some of the nice notebooks under `Code` to learn more about the data, e.g:\n\nhttps://www.kaggle.com/code/ayuraj/g2net-understand-the-data", "2019491": "Hello Everyone. I've always wanted to try and predict future time-frequency data and then I came across this dataset and I wanted to actually implement it. I've looked at a few tutorials on how LSTM data modeling works like having a constant window size and a set prediction quantity, but I have a hard understanding this HDF5 data. Open to any clarification!\n\nThanks,\n\nNaren" } }