Hello everyone !\nCheck out this new kernel in which I demonstrate two ways to denoise the seismic signals. I also visualize the effect of these denoising methods on the signals and explain the processes and results in terms of seismology. Hope you all find it useful :)
", "messages": [ { "id": "522975", "postDate": "04/25/2019 09:52:05", "content": "Hello everyone !\nCheck out this new kernel in which I demonstrate two ways to denoise the seismic signals. I also visualize the effect of these denoising methods on the signals and explain the processes and results in terms of seismology. Hope you all find it useful :)
", "rawMarkdown": "Hello everyone !\nCheck out this [new kernel](https://www.kaggle.com/tarunpaparaju/lanl-earthquake-prediction-signal-denoising) in which I demonstrate two ways to denoise the seismic signals. I also visualize the effect of these denoising methods on the signals and explain the processes and results in terms of seismology. Hope you all find it useful :)", "votes": null }, { "id": "522980", "postDate": "04/25/2019 10:15:14", "content": "There was something very similar in VSB competition. \nDid you try to implement this in your model?
", "rawMarkdown": "There was something very similar in VSB competition. \nDid you try to implement this in your model?", "votes": null }, { "id": "522992", "postDate": "04/25/2019 10:42:58", "content": "Yes, I referenced that kernel by Jack in the VSB competition. I will try it out in my model soon. It should improve the score.
", "rawMarkdown": "Yes, I referenced [that kernel](https://www.kaggle.com/jackvial/dwt-signal-denoising) by Jack in the VSB competition. I will try it out in my model soon. It should improve the score.", "votes": null }, { "id": "523007", "postDate": "04/25/2019 11:31:52", "content": "@tarunpaparaju, I believe denoising in this case can remove useful signal hence may not improve scores .
", "rawMarkdown": "tarunpaparaju, I believe denoising in this case can remove useful signal hence may not improve scores .", "votes": null }, { "id": "523010", "postDate": "04/25/2019 11:34:35", "content": "By denoising, I mean removing the artificial impulse sent by the seismograph. This artificial impulse mixes with the actual seismic impulse from the Earth to produce our current data from the seismograph. Uncovering the actual seismic activity from the Earth and ignoring the artificial impulse could help accurately determine when the earthquake will happen.
", "rawMarkdown": "By denoising, I mean removing the artificial impulse sent by the seismograph. This artificial impulse mixes with the actual seismic impulse from the Earth to produce our current data from the seismograph. Uncovering the actual seismic activity from the Earth and ignoring the artificial impulse could help accurately determine when the earthquake will happen.", "votes": null }, { "id": "523015", "postDate": "04/25/2019 11:52:08", "content": "Thank you for the nice kernel. The high-pass filter you use as the first step removes the mean of the signal. I agree that the mean of the signal is probably artifactual, but so far it seems to me that it improves the model performance, mainly by distinguishing between different earthquakes.\nI think the optimal way may be by subtracting the mean in earthquake-wise manner, but since we don't have this distinction in the test set, it would probably not work.
", "rawMarkdown": "Thank you for the nice kernel. The high-pass filter you use as the first step removes the mean of the signal. I agree that the mean of the signal is probably artifactual, but so far it seems to me that it improves the model performance, mainly by distinguishing between different earthquakes.\nI think the optimal way may be by subtracting the mean in earthquake-wise manner, but since we don't have this distinction in the test set, it would probably not work.", "votes": null }, { "id": "523179", "postDate": "04/25/2019 17:25:58", "content": "Perhaps you could try mean-normalising each segment in your feature engineering to remove this factor?
", "rawMarkdown": "Perhaps you could try mean-normalising each segment in your feature engineering to remove this factor?", "votes": null }, { "id": "523195", "postDate": "04/25/2019 18:02:41", "content": "One could compute the mean on the original signal, keep it as feature and apply filters after that
", "rawMarkdown": "One could compute the mean on the original signal, keep it as feature and apply filters after that", "votes": null }, { "id": "524831", "postDate": "04/29/2019 16:03:33", "content": "It's great for beginners as well, so easy to understand. Thanks.
", "rawMarkdown": "It's great for beginners as well, so easy to understand. Thanks.", "votes": null } ], "comments": [ { "id": 522980, "author_name": "stanislavblinov", "author_url": "", "post_date": "04/25/2019 10:15:14", "content": "There was something very similar in VSB competition. \nDid you try to implement this in your model?
", "votes": null, "replies": [ { "id": 522992, "author_name": "tarunpaparaju", "author_url": "", "post_date": "04/25/2019 10:42:58", "content": "Yes, I referenced that kernel by Jack in the VSB competition. I will try it out in my model soon. It should improve the score.
", "votes": null, "replies": [] }, { "id": 523007, "author_name": "sheriytm", "author_url": "", "post_date": "04/25/2019 11:31:52", "content": "@tarunpaparaju, I believe denoising in this case can remove useful signal hence may not improve scores .
", "votes": null, "replies": [] }, { "id": 523010, "author_name": "tarunpaparaju", "author_url": "", "post_date": "04/25/2019 11:34:35", "content": "By denoising, I mean removing the artificial impulse sent by the seismograph. This artificial impulse mixes with the actual seismic impulse from the Earth to produce our current data from the seismograph. Uncovering the actual seismic activity from the Earth and ignoring the artificial impulse could help accurately determine when the earthquake will happen.
", "votes": null, "replies": [] } ] }, { "id": 523015, "author_name": "amjad85", "author_url": "", "post_date": "04/25/2019 11:52:08", "content": "Thank you for the nice kernel. The high-pass filter you use as the first step removes the mean of the signal. I agree that the mean of the signal is probably artifactual, but so far it seems to me that it improves the model performance, mainly by distinguishing between different earthquakes.\nI think the optimal way may be by subtracting the mean in earthquake-wise manner, but since we don't have this distinction in the test set, it would probably not work.
", "votes": null, "replies": [ { "id": 523179, "author_name": "bigironsphere", "author_url": "", "post_date": "04/25/2019 17:25:58", "content": "Perhaps you could try mean-normalising each segment in your feature engineering to remove this factor?
", "votes": null, "replies": [] }, { "id": 523195, "author_name": "stecasasso", "author_url": "", "post_date": "04/25/2019 18:02:41", "content": "One could compute the mean on the original signal, keep it as feature and apply filters after that
", "votes": null, "replies": [] } ] }, { "id": 524831, "author_name": "shrutimechlearn", "author_url": "", "post_date": "04/29/2019 16:03:33", "content": "It's great for beginners as well, so easy to understand. Thanks.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "522975": "Hello everyone !\nCheck out this [new kernel](https://www.kaggle.com/tarunpaparaju/lanl-earthquake-prediction-signal-denoising) in which I demonstrate two ways to denoise the seismic signals. I also visualize the effect of these denoising methods on the signals and explain the processes and results in terms of seismology. Hope you all find it useful :)", "522980": "There was something very similar in VSB competition. \nDid you try to implement this in your model?", "522992": "Yes, I referenced [that kernel](https://www.kaggle.com/jackvial/dwt-signal-denoising) by Jack in the VSB competition. I will try it out in my model soon. It should improve the score.", "523007": "tarunpaparaju, I believe denoising in this case can remove useful signal hence may not improve scores .", "523010": "By denoising, I mean removing the artificial impulse sent by the seismograph. This artificial impulse mixes with the actual seismic impulse from the Earth to produce our current data from the seismograph. Uncovering the actual seismic activity from the Earth and ignoring the artificial impulse could help accurately determine when the earthquake will happen.", "523015": "Thank you for the nice kernel. The high-pass filter you use as the first step removes the mean of the signal. I agree that the mean of the signal is probably artifactual, but so far it seems to me that it improves the model performance, mainly by distinguishing between different earthquakes.\nI think the optimal way may be by subtracting the mean in earthquake-wise manner, but since we don't have this distinction in the test set, it would probably not work.", "523179": "Perhaps you could try mean-normalising each segment in your feature engineering to remove this factor?", "523195": "One could compute the mean on the original signal, keep it as feature and apply filters after that", "524831": "It's great for beginners as well, so easy to understand. Thanks." }, "source": "meta" }