{ "id": 91559, "title": "A question about train data", "url": "/competitions/LANL-Earthquake-Prediction/discussion/91559", "author_name": "", "post_date": "2019-05-06T12:58:52.123319700Z", "votes": null, "comment_count": 7, "views": 0, "content": "

As a non-professional, I have some question that need be answered：\n1. Are the training data obtained continuously over a long period of time without any interruption？ Can we assume that there have been many earthquakes in a given train data and been continuously monitored？\n2. Why are the same time_to_failure in a certain segment of acoustic_data

", "messages": [ { "id": "527836", "postDate": "05/06/2019 12:58:52", "content": "

As a non-professional, I have some question that need be answered：\n1. Are the training data obtained continuously over a long period of time without any interruption？ Can we assume that there have been many earthquakes in a given train data and been continuously monitored？\n2. Why are the same time_to_failure in a certain segment of acoustic_data

", "rawMarkdown": "As a non-professional, I have some question that need be answered：\n1. Are the training data obtained continuously over a long period of time without any interruption？ Can we assume that there have been many earthquakes in a given train data and been continuously monitored？\n2. Why are the same time_to_failure in a certain segment of acoustic_data", "votes": null }, { "id": "527852", "postDate": "05/06/2019 13:17:36", "content": "

Can I comprehend the train data sequences like following pictures: each sequence fragment correspond to a time to failure?

", "rawMarkdown": "Can I comprehend the train data sequences like following pictures: each sequence fragment correspond to a time to failure?", "votes": null }, { "id": "527890", "postDate": "05/06/2019 14:38:18", "content": "

Every time a quake occurs, the time_to_failure resets.

", "rawMarkdown": "Every time a quake occurs, the time_to_failure resets.", "votes": null }, { "id": "527906", "postDate": "05/06/2019 15:47:54", "content": "

Read the four publications contained in the Welcome topic - you will see that an experiment appears to consist of about 255 seconds of data, with an earthquake occurring around every 10 seconds on average. There are 16 quakes in train. The train data is continuous in time.

\n\n

The test data is probably around 10 quakes, but the data has been shuffled into 150000 segments.

\n\n

The experiment also measures stress - we don't have that data, but every time the stress gets to zero, the clock resets.

", "rawMarkdown": "Read the four publications contained in the Welcome topic - you will see that an experiment appears to consist of about 255 seconds of data, with an earthquake occurring around every 10 seconds on average. There are 16 quakes in train. The train data is continuous in time.\n\nThe test data is probably around 10 quakes, but the data has been shuffled into 150000 segments.\n\nThe experiment also measures stress - we don't have that data, but every time the stress gets to zero, the clock resets.", "votes": null }, { "id": "528351", "postDate": "05/07/2019 14:57:44", "content": "
    \n
  1. Can I understand the data like train data.JPG ( 51.83KB ) in my answer? Or I misread the seismic signals？
    2. \n
  2. Hoping you can give a detailed explanation! Hope to be able to give a diagram！\nThanks very much!
    3. \n
", "rawMarkdown": "1. Can I understand the data like train data.JPG ( 51.83KB ) in my answer? Or I misread the seismic signals？\n2. Hoping you can give a detailed explanation! Hope to be able to give a diagram！\nThanks very much!", "votes": null }, { "id": "528355", "postDate": "05/07/2019 15:23:04", "content": "

That's a possible way fo doing it, that was first proposed in @inversion kernel. But you may think of other ways.

", "rawMarkdown": "That's a possible way fo doing it, that was first proposed in @inversion kernel. But you may think of other ways.", "votes": null }, { "id": "528360", "postDate": "05/07/2019 15:27:13", "content": "

I have this feeling that your good score with few features is a result of the \"accurate\" CV strategy you are using? and it is different from the 150k that a lot of us here using?

", "rawMarkdown": "I have this feeling that your good score with few features is a result of the \"accurate\" CV strategy you are using? and it is different from the 150k that a lot of us here using?", "votes": null }, { "id": "528376", "postDate": "05/07/2019 16:06:13", "content": "

I am using segments of 150k points for training ;) That's not where I do things differently from others, if any.

", "rawMarkdown": "I am using segments of 150k points for training ;) That's not where I do things differently from others, if any.", "votes": null } ], "comments": [ { "id": 527852, "author_name": "funkygod", "author_url": "", "post_date": "05/06/2019 13:17:36", "content": "

Can I comprehend the train data sequences like following pictures: each sequence fragment correspond to a time to failure?

", "votes": null, "replies": [ { "id": 528355, "author_name": "cpmpml", "author_url": "", "post_date": "05/07/2019 15:23:04", "content": "

That's a possible way fo doing it, that was first proposed in @inversion kernel. But you may think of other ways.

", "votes": null, "replies": [] }, { "id": 528360, "author_name": "anwarsadique", "author_url": "", "post_date": "05/07/2019 15:27:13", "content": "

I have this feeling that your good score with few features is a result of the \"accurate\" CV strategy you are using? and it is different from the 150k that a lot of us here using?

", "votes": null, "replies": [] }, { "id": 528376, "author_name": "cpmpml", "author_url": "", "post_date": "05/07/2019 16:06:13", "content": "

I am using segments of 150k points for training ;) That's not where I do things differently from others, if any.

", "votes": null, "replies": [] } ] }, { "id": 527890, "author_name": "greenwing1985", "author_url": "", "post_date": "05/06/2019 14:38:18", "content": "

Every time a quake occurs, the time_to_failure resets.

", "votes": null, "replies": [] }, { "id": 527906, "author_name": "pcjimmmy", "author_url": "", "post_date": "05/06/2019 15:47:54", "content": "

Read the four publications contained in the Welcome topic - you will see that an experiment appears to consist of about 255 seconds of data, with an earthquake occurring around every 10 seconds on average. There are 16 quakes in train. The train data is continuous in time.

\n\n

The test data is probably around 10 quakes, but the data has been shuffled into 150000 segments.

\n\n

The experiment also measures stress - we don't have that data, but every time the stress gets to zero, the clock resets.

", "votes": null, "replies": [ { "id": 528351, "author_name": "funkygod", "author_url": "", "post_date": "05/07/2019 14:57:44", "content": "
    \n
  1. Can I understand the data like train data.JPG ( 51.83KB ) in my answer? Or I misread the seismic signals？
    2. \n
  2. Hoping you can give a detailed explanation! Hope to be able to give a diagram！\nThanks very much!
    3. \n
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "527836": "As a non-professional, I have some question that need be answered：\n1. Are the training data obtained continuously over a long period of time without any interruption？ Can we assume that there have been many earthquakes in a given train data and been continuously monitored？\n2. Why are the same time_to_failure in a certain segment of acoustic_data", "527852": "Can I comprehend the train data sequences like following pictures: each sequence fragment correspond to a time to failure?", "527890": "Every time a quake occurs, the time_to_failure resets.", "527906": "Read the four publications contained in the Welcome topic - you will see that an experiment appears to consist of about 255 seconds of data, with an earthquake occurring around every 10 seconds on average. There are 16 quakes in train. The train data is continuous in time.\n\nThe test data is probably around 10 quakes, but the data has been shuffled into 150000 segments.\n\nThe experiment also measures stress - we don't have that data, but every time the stress gets to zero, the clock resets.", "528351": "1. Can I understand the data like train data.JPG ( 51.83KB ) in my answer? Or I misread the seismic signals？\n2. Hoping you can give a detailed explanation! Hope to be able to give a diagram！\nThanks very much!", "528355": "That's a possible way fo doing it, that was first proposed in @inversion kernel. But you may think of other ways.", "528360": "I have this feeling that your good score with few features is a result of the \"accurate\" CV strategy you are using? and it is different from the 150k that a lot of us here using?", "528376": "I am using segments of 150k points for training ;) That's not where I do things differently from others, if any." }, "source": "meta" }