There are some aspects of the published data I don't get. I don't have any domain knowledge, so maybe someone could help me with my concerns.
\n\nWhy is the recorded period so small? We are predicting very short range of values, what's the point of that? I imagine information like \"there will be an earthquake in one hour\", or \"next year\", or \"tomorrow morning\", could be useful for the people, but what's the point of saying \"there will be the earthquake in 0.003 second\", or \"there will be ten earthquakes in the following 16 seconds\".
Why the sampling frequency is so huge? 4MHz can measure waves of length equal to 0.0000005, or something like that, is it even dangerous? How can a wave that is so small propagate for such long distance? My suspicion is that only long waves like 1 mm or longer are dangerous because they make the earth movement visible.
For example in cardiology, an expert cardiologist can label the signal normal or abnormal. How is the data labeled in case of earthquakes? Are there any earthquakes experts, how to become one? ;) Or maybe the confirmation of the occurrence of the earthquake results from the caused damages? When the earthquake starts, and ends? How could this data be possibly tagged?
My guess would be that sampling frequency is, in fact, 40Hz: 629145000 / 40 (Hz) / 3600 (h) / 24 (day) = ~182 days so we have frequencies between 0 and 20 Hz, smallest waves equal to 5 cm, 182 days of recordings when 16 earthquakes happened.
\n\nWithout the understanding of the 3 above points, I have no motivation for doing the competition unless sampling frequency isn't so huge and time_to_failure has the comma in a wrong place.
\n\nWhat do you think?
", "messages": [ { "id": "456467", "postDate": "01/15/2019 21:38:21", "content": "There are some aspects of the published data I don't get. I don't have any domain knowledge, so maybe someone could help me with my concerns.
\n\nWhy is the recorded period so small? We are predicting very short range of values, what's the point of that? I imagine information like \"there will be an earthquake in one hour\", or \"next year\", or \"tomorrow morning\", could be useful for the people, but what's the point of saying \"there will be the earthquake in 0.003 second\", or \"there will be ten earthquakes in the following 16 seconds\".
Why the sampling frequency is so huge? 4MHz can measure waves of length equal to 0.0000005, or something like that, is it even dangerous? How can a wave that is so small propagate for such long distance? My suspicion is that only long waves like 1 mm or longer are dangerous because they make the earth movement visible.
For example in cardiology, an expert cardiologist can label the signal normal or abnormal. How is the data labeled in case of earthquakes? Are there any earthquakes experts, how to become one? ;) Or maybe the confirmation of the occurrence of the earthquake results from the caused damages? When the earthquake starts, and ends? How could this data be possibly tagged?
My guess would be that sampling frequency is, in fact, 40Hz: 629145000 / 40 (Hz) / 3600 (h) / 24 (day) = ~182 days so we have frequencies between 0 and 20 Hz, smallest waves equal to 5 cm, 182 days of recordings when 16 earthquakes happened.
\n\nWithout the understanding of the 3 above points, I have no motivation for doing the competition unless sampling frequency isn't so huge and time_to_failure has the comma in a wrong place.
\n\nWhat do you think?
", "rawMarkdown": "There are some aspects of the published data I don't get. I don't have any domain knowledge, so maybe someone could help me with my concerns.\n\n1. Why is the recorded period so small? We are predicting very short range of values, what's the point of that? I imagine information like \"there will be an earthquake in one hour\", or \"next year\", or \"tomorrow morning\", could be useful for the people, but what's the point of saying \"there will be the earthquake in 0.003 second\", or \"there will be ten earthquakes in the following 16 seconds\".\n\n2. Why the sampling frequency is so huge? 4MHz can measure waves of length equal to 0.0000005, or something like that, is it even dangerous? How can a wave that is so small propagate for such long distance? My suspicion is that only long waves like 1 mm or longer are dangerous because they make the earth movement visible.\n\n3. For example in cardiology, an expert cardiologist can label the signal normal or abnormal. How is the data labeled in case of earthquakes? Are there any earthquakes experts, how to become one? ;) Or maybe the confirmation of the occurrence of the earthquake results from the caused damages? When the earthquake starts, and ends? How could this data be possibly tagged? \n\nMy guess would be that sampling frequency is, in fact, 40Hz: 629145000 / 40 (Hz) / 3600 (h) / 24 (day) = ~182 days so we have frequencies between 0 and 20 Hz, smallest waves equal to 5 cm, 182 days of recordings when 16 earthquakes happened. \n\nWithout the understanding of the 3 above points, I have no motivation for doing the competition unless sampling frequency isn't so huge and time_to_failure has the comma in a wrong place.\n\nWhat do you think?", "votes": null }, { "id": "456749", "postDate": "01/16/2019 13:04:14", "content": "Hi DavidS, I think the answer can be found in the description: \"In this competition, you will address when the earthquake will take place. Specifically, you’ll predict the time remaining before laboratory earthquakes occur from real-time seismic data. \"
\n\nIt's an experiment in a laboratory, not real life earthquakes. You can read about the experiment in the provided papers here: https://www.kaggle.com/c/LANL-Earthquake-Prediction/discussion/77525
", "rawMarkdown": "Hi DavidS, I think the answer can be found in the description: \"In this competition, you will address when the earthquake will take place. Specifically, you’ll predict the time remaining before **laboratory** earthquakes occur from real-time seismic data. \"\n\nIt's an experiment in a laboratory, not real life earthquakes. You can read about the experiment in the provided papers here: https://www.kaggle.com/c/LANL-Earthquake-Prediction/discussion/77525", "votes": null }, { "id": "540617", "postDate": "05/31/2019 18:34:54", "content": "We have about 163 seconds of data in train.
", "rawMarkdown": "We have about 163 seconds of data in train.", "votes": null }, { "id": "540623", "postDate": "05/31/2019 18:47:39", "content": "I cannot move these thoughts forward. I think I was wrong.
", "rawMarkdown": "I cannot move these thoughts forward. I think I was wrong.", "votes": null }, { "id": "540633", "postDate": "05/31/2019 19:21:17", "content": "Well, not really. I don't think I was wrong. I can't prove I'm right, and I don't think it makes any difference. After reading\nhttps://www.nature.com/articles/s41561-018-0274-6.epdf?author_access_token=ufRH8L72R8YTtQHPYd8EBdRgN0jAjWel9jnR3ZoTv0MVCBq1iB8iH5RzSr1QIjwlUDw2153BkSXUbxpE3ZdR99KcH0eOpxD7moW-HGhvNXb6fRCPBFfLusxZ_pSFQ7lrbSIMXOUM5YfrlUBa1I2dlw%3D%3D\nI don't believe there are 163 seconds nor that the sampling rate is ~4MHz.
", "rawMarkdown": "Well, not really. I don't think I was wrong. I can't prove I'm right, and I don't think it makes any difference. After reading\nhttps://www.nature.com/articles/s41561-018-0274-6.epdf?author_access_token=ufRH8L72R8YTtQHPYd8EBdRgN0jAjWel9jnR3ZoTv0MVCBq1iB8iH5RzSr1QIjwlUDw2153BkSXUbxpE3ZdR99KcH0eOpxD7moW-HGhvNXb6fRCPBFfLusxZ_pSFQ7lrbSIMXOUM5YfrlUBa1I2dlw%3D%3D\nI don't believe there are 163 seconds nor that the sampling rate is ~4MHz.", "votes": null }, { "id": "540638", "postDate": "05/31/2019 19:28:49", "content": "you are referencing a real world data paper there.\nIn this competition, it's a labquake. And it is indeed 163 s of training data and sampling rate is ~4 MHz
", "rawMarkdown": "you are referencing a real world data paper there.\nIn this competition, it's a labquake. And it is indeed 163 s of training data and sampling rate is ~4 MHz", "votes": null }, { "id": "540674", "postDate": "05/31/2019 20:39:34", "content": "This competition is a laboratory experiment recorded using a 4MHz sound transducer + A/D converter. There is 163s of data in trainset. Also competition description and some papers confirms that info.
", "rawMarkdown": "This competition is a laboratory experiment recorded using a 4MHz sound transducer + A/D converter. There is 163s of data in trainset. Also competition description and some papers confirms that info.", "votes": null } ], "comments": [ { "id": 456749, "author_name": "maximilianhahn", "author_url": "", "post_date": "01/16/2019 13:04:14", "content": "Hi DavidS, I think the answer can be found in the description: \"In this competition, you will address when the earthquake will take place. Specifically, you’ll predict the time remaining before laboratory earthquakes occur from real-time seismic data. \"
\n\nIt's an experiment in a laboratory, not real life earthquakes. You can read about the experiment in the provided papers here: https://www.kaggle.com/c/LANL-Earthquake-Prediction/discussion/77525
", "votes": null, "replies": [] }, { "id": 540617, "author_name": "cpmpml", "author_url": "", "post_date": "05/31/2019 18:34:54", "content": "We have about 163 seconds of data in train.
", "votes": null, "replies": [ { "id": 540623, "author_name": "davids1992", "author_url": "", "post_date": "05/31/2019 18:47:39", "content": "I cannot move these thoughts forward. I think I was wrong.
", "votes": null, "replies": [] }, { "id": 540633, "author_name": "davids1992", "author_url": "", "post_date": "05/31/2019 19:21:17", "content": "Well, not really. I don't think I was wrong. I can't prove I'm right, and I don't think it makes any difference. After reading\nhttps://www.nature.com/articles/s41561-018-0274-6.epdf?author_access_token=ufRH8L72R8YTtQHPYd8EBdRgN0jAjWel9jnR3ZoTv0MVCBq1iB8iH5RzSr1QIjwlUDw2153BkSXUbxpE3ZdR99KcH0eOpxD7moW-HGhvNXb6fRCPBFfLusxZ_pSFQ7lrbSIMXOUM5YfrlUBa1I2dlw%3D%3D\nI don't believe there are 163 seconds nor that the sampling rate is ~4MHz.
", "votes": null, "replies": [] }, { "id": 540638, "author_name": "ilu000", "author_url": "", "post_date": "05/31/2019 19:28:49", "content": "you are referencing a real world data paper there.\nIn this competition, it's a labquake. And it is indeed 163 s of training data and sampling rate is ~4 MHz
", "votes": null, "replies": [] } ] }, { "id": 540674, "author_name": "titericz", "author_url": "", "post_date": "05/31/2019 20:39:34", "content": "This competition is a laboratory experiment recorded using a 4MHz sound transducer + A/D converter. There is 163s of data in trainset. Also competition description and some papers confirms that info.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "456467": "There are some aspects of the published data I don't get. I don't have any domain knowledge, so maybe someone could help me with my concerns.\n\n1. Why is the recorded period so small? We are predicting very short range of values, what's the point of that? I imagine information like \"there will be an earthquake in one hour\", or \"next year\", or \"tomorrow morning\", could be useful for the people, but what's the point of saying \"there will be the earthquake in 0.003 second\", or \"there will be ten earthquakes in the following 16 seconds\".\n\n2. Why the sampling frequency is so huge? 4MHz can measure waves of length equal to 0.0000005, or something like that, is it even dangerous? How can a wave that is so small propagate for such long distance? My suspicion is that only long waves like 1 mm or longer are dangerous because they make the earth movement visible.\n\n3. For example in cardiology, an expert cardiologist can label the signal normal or abnormal. How is the data labeled in case of earthquakes? Are there any earthquakes experts, how to become one? ;) Or maybe the confirmation of the occurrence of the earthquake results from the caused damages? When the earthquake starts, and ends? How could this data be possibly tagged? \n\nMy guess would be that sampling frequency is, in fact, 40Hz: 629145000 / 40 (Hz) / 3600 (h) / 24 (day) = ~182 days so we have frequencies between 0 and 20 Hz, smallest waves equal to 5 cm, 182 days of recordings when 16 earthquakes happened. \n\nWithout the understanding of the 3 above points, I have no motivation for doing the competition unless sampling frequency isn't so huge and time_to_failure has the comma in a wrong place.\n\nWhat do you think?", "456749": "Hi DavidS, I think the answer can be found in the description: \"In this competition, you will address when the earthquake will take place. Specifically, you’ll predict the time remaining before **laboratory** earthquakes occur from real-time seismic data. \"\n\nIt's an experiment in a laboratory, not real life earthquakes. You can read about the experiment in the provided papers here: https://www.kaggle.com/c/LANL-Earthquake-Prediction/discussion/77525", "540617": "We have about 163 seconds of data in train.", "540623": "I cannot move these thoughts forward. I think I was wrong.", "540633": "Well, not really. I don't think I was wrong. I can't prove I'm right, and I don't think it makes any difference. After reading\nhttps://www.nature.com/articles/s41561-018-0274-6.epdf?author_access_token=ufRH8L72R8YTtQHPYd8EBdRgN0jAjWel9jnR3ZoTv0MVCBq1iB8iH5RzSr1QIjwlUDw2153BkSXUbxpE3ZdR99KcH0eOpxD7moW-HGhvNXb6fRCPBFfLusxZ_pSFQ7lrbSIMXOUM5YfrlUBa1I2dlw%3D%3D\nI don't believe there are 163 seconds nor that the sampling rate is ~4MHz.", "540638": "you are referencing a real world data paper there.\nIn this competition, it's a labquake. And it is indeed 163 s of training data and sampling rate is ~4 MHz", "540674": "This competition is a laboratory experiment recorded using a 4MHz sound transducer + A/D converter. There is 163s of data in trainset. Also competition description and some papers confirms that info." }, "source": "meta" }