{
  "id": 262003,
  "title": "Any success with LSTM?",
  "url": "/competitions/google-smartphone-decimeter-challenge/discussion/262003",
  "author_name": "majoraregalia",
  "post_date": "2021-08-05T14:12:19.775000",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Did anyone have any success with using the LSTM? I have used the LSTM model for SJC area and landed on 20 public/39 private place, wonder if anyone has achieved a better result.</p>\n<p>I was too lazy to deal with multipath in that area, so I have used lat/lng, lags, IMU data and decided to predict not the raw values, but differences between coord_pred and coord_gt. After that I have used a bit modified snap to grid.</p>",
  "messages": [
    {
      "id": 1452077,
      "postDate": "2021-08-05T14:12:19.777Z",
      "content": "<p>Did anyone have any success with using the LSTM? I have used the LSTM model for SJC area and landed on 20 public/39 private place, wonder if anyone has achieved a better result.</p>\n<p>I was too lazy to deal with multipath in that area, so I have used lat/lng, lags, IMU data and decided to predict not the raw values, but differences between coord_pred and coord_gt. After that I have used a bit modified snap to grid.</p>",
      "rawMarkdown": "Did anyone have any success with using the LSTM? I have used the LSTM model for SJC area and landed on 20 public/39 private place, wonder if anyone has achieved a better result.\n\nI was too lazy to deal with multipath in that area, so I have used lat/lng, lags, IMU data and decided to predict not the raw values, but differences between coord_pred and coord_gt. After that I have used a bit modified snap to grid."
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1452077": "Did anyone have any success with using the LSTM? I have used the LSTM model for SJC area and landed on 20 public/39 private place, wonder if anyone has achieved a better result.\n\nI was too lazy to deal with multipath in that area, so I have used lat/lng, lags, IMU data and decided to predict not the raw values, but differences between coord_pred and coord_gt. After that I have used a bit modified snap to grid."
  }
}