{
  "id": 77984,
  "title": "Inconsistency in Training Data? Sampling rate vs. TimeToFailure",
  "url": "/competitions/LANL-Earthquake-Prediction/discussion/77984",
  "author_name": "",
  "post_date": "2019-01-18T11:09:29.590797400Z",
  "votes": 1,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I need to expand on my comment <a href=\"https://www.kaggle.com/c/LANL-Earthquake-Prediction/discussion/77526\">here</a>. Maybe I'm missing something, but it seems to me that in the training data train.csv either</p>\n\n<p>(a) the sampling rate is exactly 4 MHz and therefore the time elapsed between two successive data points is always exactly 250 ns. Then the TimeToFailure (ttf) column is off by up to 0,5 seconds (!).</p>\n\n<p><em>or</em></p>\n\n<p>(b) the ttf is correct, then the time between two successive data points is not constant, which would be a real problem (e.g. for Fourier transforms).</p>\n\n<p>For example, quake #1 in train.csv occurs at line 5,656,574. The next line (start of new countdown to quake #2) has a ttf of 11.540 sec. quake #2 occurs at line 50,085,878. ttf of this line is 0.001 sec. According to the ttf values, the time between quake #1 and #2 should therefore be 11.540 sec. However, there are 44,429,304 data points in this time span, corresponding to a duration of 44,429,304 * 250 nsec = 11.107 sec. Those two values for the duration are 0.433 sec off. Which one is correct?</p>\n\n<p>Bertrand, could you help me here? Thanks in advance!</p>",
  "messages": [
    {
      "id": "457932",
      "postDate": "01/18/2019 11:09:29",
      "content": "<p>I need to expand on my comment <a href=\"https://www.kaggle.com/c/LANL-Earthquake-Prediction/discussion/77526\">here</a>. Maybe I'm missing something, but it seems to me that in the training data train.csv either</p>\n\n<p>(a) the sampling rate is exactly 4 MHz and therefore the time elapsed between two successive data points is always exactly 250 ns. Then the TimeToFailure (ttf) column is off by up to 0,5 seconds (!).</p>\n\n<p><em>or</em></p>\n\n<p>(b) the ttf is correct, then the time between two successive data points is not constant, which would be a real problem (e.g. for Fourier transforms).</p>\n\n<p>For example, quake #1 in train.csv occurs at line 5,656,574. The next line (start of new countdown to quake #2) has a ttf of 11.540 sec. quake #2 occurs at line 50,085,878. ttf of this line is 0.001 sec. According to the ttf values, the time between quake #1 and #2 should therefore be 11.540 sec. However, there are 44,429,304 data points in this time span, corresponding to a duration of 44,429,304 * 250 nsec = 11.107 sec. Those two values for the duration are 0.433 sec off. Which one is correct?</p>\n\n<p>Bertrand, could you help me here? Thanks in advance!</p>",
      "rawMarkdown": "I need to expand on my comment [here][1]. Maybe I'm missing something, but it seems to me that in the training data train.csv either\n\n(a) the sampling rate is exactly 4 MHz and therefore the time elapsed between two successive data points is always exactly 250 ns. Then the TimeToFailure (ttf) column is off by up to 0,5 seconds (!).\n\n*or*\n\n(b) the ttf is correct, then the time between two successive data points is not constant, which would be a real problem (e.g. for Fourier transforms).\n\nFor example, quake #1 in train.csv occurs at line 5,656,574. The next line (start of new countdown to quake #2) has a ttf of 11.540 sec. quake #2 occurs at line 50,085,878. ttf of this line is 0.001 sec. According to the ttf values, the time between quake #1 and #2 should therefore be 11.540 sec. However, there are 44,429,304 data points in this time span, corresponding to a duration of 44,429,304 * 250 nsec = 11.107 sec. Those two values for the duration are 0.433 sec off. Which one is correct?\n\nBertrand, could you help me here? Thanks in advance!\n\n\n\n\n\n\n\n\n\n\n  [1]: https://www.kaggle.com/c/LANL-Earthquake-Prediction/discussion/77526",
      "votes": null
    },
    {
      "id": "466772",
      "postDate": "02/05/2019 22:54:16",
      "content": "<p>Solved, see main explanation comment.</p>",
      "rawMarkdown": "Solved, see main explanation comment.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 466772,
      "author_name": "friedchips",
      "author_url": "",
      "post_date": "02/05/2019 22:54:16",
      "content": "<p>Solved, see main explanation comment.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "457932": "I need to expand on my comment [here][1]. Maybe I'm missing something, but it seems to me that in the training data train.csv either\n\n(a) the sampling rate is exactly 4 MHz and therefore the time elapsed between two successive data points is always exactly 250 ns. Then the TimeToFailure (ttf) column is off by up to 0,5 seconds (!).\n\n*or*\n\n(b) the ttf is correct, then the time between two successive data points is not constant, which would be a real problem (e.g. for Fourier transforms).\n\nFor example, quake #1 in train.csv occurs at line 5,656,574. The next line (start of new countdown to quake #2) has a ttf of 11.540 sec. quake #2 occurs at line 50,085,878. ttf of this line is 0.001 sec. According to the ttf values, the time between quake #1 and #2 should therefore be 11.540 sec. However, there are 44,429,304 data points in this time span, corresponding to a duration of 44,429,304 * 250 nsec = 11.107 sec. Those two values for the duration are 0.433 sec off. Which one is correct?\n\nBertrand, could you help me here? Thanks in advance!\n\n\n\n\n\n\n\n\n\n\n  [1]: https://www.kaggle.com/c/LANL-Earthquake-Prediction/discussion/77526",
    "466772": "Solved, see main explanation comment."
  },
  "source": "meta"
}