{
  "id": 190766,
  "title": "Welcome to the first competition on Volcanic Eruption Prediction",
  "url": "/competitions/predict-volcanic-eruptions-ingv-oe/discussion/190766",
  "author_name": "Flavio Cannavo",
  "post_date": "2020-10-13T09:12:28.196000",
  "votes": 35,
  "comment_count": 8,
  "views": 0,
  "content": "<p>Hello folks!<br>\nWelcome to the first competition on <strong>Volcanic Eruption Prediction</strong>.<br>\nThe eruption prediction problem is crucial for volcanic risk mitigation in sensible areas (such as human settlements). Indeed, volcanic eruptions could lead to important loss of life, property, and disruption of human activities and could compromise flight safety.<br>\nThe science of volcanic eruption prediction has significantly advanced over the past decades. Many studies indicate that signs of unrest preceding an eruption would be detectable in seismic signals for different time scales (from minutes to months). In solving the short-term prediction problem, some types of observed seismic activity (e.g. very long period events, long period events, volcanic tremor, etc.) are, at the moment, considered reliable precursors of volcanic activity since mostly associated with variations in the fluid-conduit pressure field. Nevertheless, it is believed that seismic waves can carry more information than is known. In this aim, a deep learning approach on massive data could help scientists in exploring currently hidden features in seismic data.<br>\nINGV (National Institute of Geophysics and Volcanology) is the subject institutionally responsible for the monitoring and surveillance of Italian volcanoes. Apart from Italian volcanoes, scientists at INGV are also involved in studies of volcanoes on a global scale.<br>\nIn this challenge, you will try to improve the current weak predictive capabilities by extracting the maximum information encoded in the signals that the Earth sends us. <br>\nIn the data set you will find segments of signals recorded at 10 seismometers deployed on an active volcano. Each segment has associated a time-to-eruption information which represents the time that was passed from the end of the segment to the beginning of the next eruption.<br>\nSince the times of the eruptions are publicly available for all the volcanoes worldwide, we didn’t insert the absolute time, and, anyway, we reserve the right to re-train the posted models and verify test results. <br>\nWe hope you will enjoy the challenge and, whether you win or you don’t, you will contribute to the advancement of our knowledge on volcanoes. <br>\nWe wish you the best of luck.</p>",
  "messages": [
    {
      "id": 1048190,
      "postDate": "2020-10-13T09:12:28.197Z",
      "content": "<p>Hello folks!<br>\nWelcome to the first competition on <strong>Volcanic Eruption Prediction</strong>.<br>\nThe eruption prediction problem is crucial for volcanic risk mitigation in sensible areas (such as human settlements). Indeed, volcanic eruptions could lead to important loss of life, property, and disruption of human activities and could compromise flight safety.<br>\nThe science of volcanic eruption prediction has significantly advanced over the past decades. Many studies indicate that signs of unrest preceding an eruption would be detectable in seismic signals for different time scales (from minutes to months). In solving the short-term prediction problem, some types of observed seismic activity (e.g. very long period events, long period events, volcanic tremor, etc.) are, at the moment, considered reliable precursors of volcanic activity since mostly associated with variations in the fluid-conduit pressure field. Nevertheless, it is believed that seismic waves can carry more information than is known. In this aim, a deep learning approach on massive data could help scientists in exploring currently hidden features in seismic data.<br>\nINGV (National Institute of Geophysics and Volcanology) is the subject institutionally responsible for the monitoring and surveillance of Italian volcanoes. Apart from Italian volcanoes, scientists at INGV are also involved in studies of volcanoes on a global scale.<br>\nIn this challenge, you will try to improve the current weak predictive capabilities by extracting the maximum information encoded in the signals that the Earth sends us. <br>\nIn the data set you will find segments of signals recorded at 10 seismometers deployed on an active volcano. Each segment has associated a time-to-eruption information which represents the time that was passed from the end of the segment to the beginning of the next eruption.<br>\nSince the times of the eruptions are publicly available for all the volcanoes worldwide, we didn’t insert the absolute time, and, anyway, we reserve the right to re-train the posted models and verify test results. <br>\nWe hope you will enjoy the challenge and, whether you win or you don’t, you will contribute to the advancement of our knowledge on volcanoes. <br>\nWe wish you the best of luck.</p>",
      "rawMarkdown": "Hello folks!\nWelcome to the first competition on **Volcanic Eruption Prediction**.\nThe eruption prediction problem is crucial for volcanic risk mitigation in sensible areas (such as human settlements). Indeed, volcanic eruptions could lead to important loss of life, property, and disruption of human activities and could compromise flight safety.\nThe science of volcanic eruption prediction has significantly advanced over the past decades. Many studies indicate that signs of unrest preceding an eruption would be detectable in seismic signals for different time scales (from minutes to months). In solving the short-term prediction problem, some types of observed seismic activity (e.g. very long period events, long period events, volcanic tremor, etc.) are, at the moment, considered reliable precursors of volcanic activity since mostly associated with variations in the fluid-conduit pressure field. Nevertheless, it is believed that seismic waves can carry more information than is known. In this aim, a deep learning approach on massive data could help scientists in exploring currently hidden features in seismic data.\nINGV (National Institute of Geophysics and Volcanology) is the subject institutionally responsible for the monitoring and surveillance of Italian volcanoes. Apart from Italian volcanoes, scientists at INGV are also involved in studies of volcanoes on a global scale.\nIn this challenge, you will try to improve the current weak predictive capabilities by extracting the maximum information encoded in the signals that the Earth sends us. \nIn the data set you will find segments of signals recorded at 10 seismometers deployed on an active volcano. Each segment has associated a time-to-eruption information which represents the time that was passed from the end of the segment to the beginning of the next eruption.\nSince the times of the eruptions are publicly available for all the volcanoes worldwide, we didn’t insert the absolute time, and, anyway, we reserve the right to re-train the posted models and verify test results. \nWe hope you will enjoy the challenge and, whether you win or you don’t, you will contribute to the advancement of our knowledge on volcanoes. \nWe wish you the best of luck.",
      "votes": 35
    },
    {
      "id": 1051794,
      "postDate": "2020-10-16T21:55:44.877Z",
      "content": "<p>Talking with Boris Behncke at INGV in 2007 is actually what got me into geophysics. I am more than excited to see this challenge on here, so I made in <a href=\"https://www.kaggle.com/jesperdramsch/introduction-to-volcanology-seismograms-and-lgbm\" target=\"_blank\">intro notebook</a> that explains volcanology and seismology a little bit.</p>",
      "rawMarkdown": "Talking with Boris Behncke at INGV in 2007 is actually what got me into geophysics. I am more than excited to see this challenge on here, so I made in [intro notebook](https://www.kaggle.com/jesperdramsch/introduction-to-volcanology-seismograms-and-lgbm) that explains volcanology and seismology a little bit.\n",
      "votes": 8
    },
    {
      "id": 1055716,
      "postDate": "2020-10-21T03:27:54.137Z",
      "content": "<p><a href=\"https://www.kaggle.com/pldatacybernetics\" target=\"_blank\">@pldatacybernetics</a> For each segment, its median value was subtracted from the data, plus the data was rescaled by a factor related to the sensor.</p>",
      "rawMarkdown": "@pldatacybernetics For each segment, its median value was subtracted from the data, plus the data was rescaled by a factor related to the sensor.",
      "votes": 1
    },
    {
      "id": 1055126,
      "postDate": "2020-10-20T13:57:00.870Z",
      "content": "<p>I don't believe that I am quite at the level of skill needed to partake in this challenge, but I would love to see what kind of data and predictions participants come up with. Particularly, I hope to see some predictive models about volcanos in the northwestern United States. </p>",
      "rawMarkdown": "I don't believe that I am quite at the level of skill needed to partake in this challenge, but I would love to see what kind of data and predictions participants come up with. Particularly, I hope to see some predictive models about volcanos in the northwestern United States. ",
      "votes": 1
    },
    {
      "id": 1054865,
      "postDate": "2020-10-20T08:36:34.543Z",
      "content": "<p>I have absolutely no clue of geo-physics but to me the data does not look like the classic seismometer data I see on the web. In the data description there is already mentioned that there was some kind of normalization and data cleaning applied.<br>\nIs there more information available on what kind of data cleaning was applied to the raw data? <br>\nThx</p>",
      "rawMarkdown": "I have absolutely no clue of geo-physics but to me the data does not look like the classic seismometer data I see on the web. In the data description there is already mentioned that there was some kind of normalization and data cleaning applied.\nIs there more information available on what kind of data cleaning was applied to the raw data? \nThx"
    },
    {
      "id": 1083010,
      "postDate": "2020-11-18T13:41:07.707Z",
      "content": "<p>Hey!</p>\n<p>What is the sampling rate of the raw and what is the time units?<br>\nThanks</p>",
      "rawMarkdown": "Hey!\n\nWhat is the sampling rate of the raw and what is the time units?\nThanks",
      "replies": [
        {
          "id": 1098411,
          "postDate": "2020-12-01T16:13:56.927Z",
          "content": "<p>BTW, this question was answered here: (<a href=\"https://www.kaggle.com/c/predict-volcanic-eruptions-ingv-oe/discussion/190682)\" target=\"_blank\">https://www.kaggle.com/c/predict-volcanic-eruptions-ingv-oe/discussion/190682)</a>.  Both are based on 100th of a second, so the times to eruption in the training data set vary from one minute to 5 1/2 days.</p>",
          "rawMarkdown": "BTW, this question was answered here: (https://www.kaggle.com/c/predict-volcanic-eruptions-ingv-oe/discussion/190682).  Both are based on 100th of a second, so the times to eruption in the training data set vary from one minute to 5 1/2 days."
        }
      ]
    },
    {
      "id": 1178310,
      "postDate": "2021-01-30T19:31:56.827Z",
      "rawMarkdown": "",
      "isDeleted": true
    },
    {
      "id": 1067821,
      "postDate": "2020-11-02T18:32:11.737Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 1051794,
      "author_name": "Jesper Sören Dramsch",
      "author_url": "",
      "post_date": "2020-10-16T21:55:44.877000",
      "content": "<p>Talking with Boris Behncke at INGV in 2007 is actually what got me into geophysics. I am more than excited to see this challenge on here, so I made in <a href=\"https://www.kaggle.com/jesperdramsch/introduction-to-volcanology-seismograms-and-lgbm\" target=\"_blank\">intro notebook</a> that explains volcanology and seismology a little bit.</p>",
      "votes": 8,
      "replies": []
    },
    {
      "id": 1055716,
      "author_name": "Flavio Cannavo",
      "author_url": "",
      "post_date": "2020-10-21T03:27:54.137000",
      "content": "<p><a href=\"https://www.kaggle.com/pldatacybernetics\" target=\"_blank\">@pldatacybernetics</a> For each segment, its median value was subtracted from the data, plus the data was rescaled by a factor related to the sensor.</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 1055126,
      "author_name": "Nick Shattuck",
      "author_url": "",
      "post_date": "2020-10-20T13:57:00.870000",
      "content": "<p>I don't believe that I am quite at the level of skill needed to partake in this challenge, but I would love to see what kind of data and predictions participants come up with. Particularly, I hope to see some predictive models about volcanos in the northwestern United States. </p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 1054865,
      "author_name": "Philipp Leser",
      "author_url": "",
      "post_date": "2020-10-20T08:36:34.543000",
      "content": "<p>I have absolutely no clue of geo-physics but to me the data does not look like the classic seismometer data I see on the web. In the data description there is already mentioned that there was some kind of normalization and data cleaning applied.<br>\nIs there more information available on what kind of data cleaning was applied to the raw data? <br>\nThx</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 1083010,
      "author_name": "Michael Markzon",
      "author_url": "",
      "post_date": "2020-11-18T13:41:07.707000",
      "content": "<p>Hey!</p>\n<p>What is the sampling rate of the raw and what is the time units?<br>\nThanks</p>",
      "votes": 0,
      "replies": [
        {
          "id": 1098411,
          "author_name": "Bruce Harold",
          "author_url": "",
          "post_date": "2020-12-01T16:13:56.927000",
          "content": "<p>BTW, this question was answered here: (<a href=\"https://www.kaggle.com/c/predict-volcanic-eruptions-ingv-oe/discussion/190682)\" target=\"_blank\">https://www.kaggle.com/c/predict-volcanic-eruptions-ingv-oe/discussion/190682)</a>.  Both are based on 100th of a second, so the times to eruption in the training data set vary from one minute to 5 1/2 days.</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 1178310,
      "author_name": "",
      "author_url": "",
      "post_date": "2021-01-30T19:31:56.827000",
      "content": "",
      "votes": 0,
      "replies": []
    },
    {
      "id": 1067821,
      "author_name": "",
      "author_url": "",
      "post_date": "2020-11-02T18:32:11.737000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1048190": "Hello folks!\nWelcome to the first competition on **Volcanic Eruption Prediction**.\nThe eruption prediction problem is crucial for volcanic risk mitigation in sensible areas (such as human settlements). Indeed, volcanic eruptions could lead to important loss of life, property, and disruption of human activities and could compromise flight safety.\nThe science of volcanic eruption prediction has significantly advanced over the past decades. Many studies indicate that signs of unrest preceding an eruption would be detectable in seismic signals for different time scales (from minutes to months). In solving the short-term prediction problem, some types of observed seismic activity (e.g. very long period events, long period events, volcanic tremor, etc.) are, at the moment, considered reliable precursors of volcanic activity since mostly associated with variations in the fluid-conduit pressure field. Nevertheless, it is believed that seismic waves can carry more information than is known. In this aim, a deep learning approach on massive data could help scientists in exploring currently hidden features in seismic data.\nINGV (National Institute of Geophysics and Volcanology) is the subject institutionally responsible for the monitoring and surveillance of Italian volcanoes. Apart from Italian volcanoes, scientists at INGV are also involved in studies of volcanoes on a global scale.\nIn this challenge, you will try to improve the current weak predictive capabilities by extracting the maximum information encoded in the signals that the Earth sends us. \nIn the data set you will find segments of signals recorded at 10 seismometers deployed on an active volcano. Each segment has associated a time-to-eruption information which represents the time that was passed from the end of the segment to the beginning of the next eruption.\nSince the times of the eruptions are publicly available for all the volcanoes worldwide, we didn’t insert the absolute time, and, anyway, we reserve the right to re-train the posted models and verify test results. \nWe hope you will enjoy the challenge and, whether you win or you don’t, you will contribute to the advancement of our knowledge on volcanoes. \nWe wish you the best of luck.",
    "1051794": "Talking with Boris Behncke at INGV in 2007 is actually what got me into geophysics. I am more than excited to see this challenge on here, so I made in [intro notebook](https://www.kaggle.com/jesperdramsch/introduction-to-volcanology-seismograms-and-lgbm) that explains volcanology and seismology a little bit.\n",
    "1055716": "@pldatacybernetics For each segment, its median value was subtracted from the data, plus the data was rescaled by a factor related to the sensor.",
    "1055126": "I don't believe that I am quite at the level of skill needed to partake in this challenge, but I would love to see what kind of data and predictions participants come up with. Particularly, I hope to see some predictive models about volcanos in the northwestern United States. ",
    "1054865": "I have absolutely no clue of geo-physics but to me the data does not look like the classic seismometer data I see on the web. In the data description there is already mentioned that there was some kind of normalization and data cleaning applied.\nIs there more information available on what kind of data cleaning was applied to the raw data? \nThx",
    "1083010": "Hey!\n\nWhat is the sampling rate of the raw and what is the time units?\nThanks",
    "1178310": "",
    "1067821": ""
  }
}