{
  "id": 444773,
  "title": "Train dataset with no NaNs 3 times bigger than previous piblic",
  "url": "/competitions/child-mind-institute-detect-sleep-states/discussion/444773",
  "author_name": "Oleg Khudyakov",
  "post_date": "2023-10-03T14:21:55.740000",
  "votes": 3,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Hello everyone!</p>\n<p>Until today, I used <a href=\"https://www.kaggle.com/competitions/child-mind-institute-detect-sleep-states/discussion/437345\" target=\"_blank\">this dataset</a>, consisting of 35 series without NaN values.</p>\n<p>To increase the amount of training data, I decided to use more series by truncating them to the first NaN value in the event column.</p>\n<p>My notebook is <a href=\"https://www.kaggle.com/code/kaggledummie007/train-series-of-week-and-longer-w-o-nans\" target=\"_blank\">here</a>.</p>\n<p>In addition, I dropped the series consisting of less than 7 nights so that the model could take into account the influence of the weekend.</p>\n<p>As a result, the <strong>dataset has grown by about 3 times</strong>.</p>\n<p>I split the dataset into 3 sets (lists, actually) of series containing approximately an equal number of nights. I use these sets of series A B C to train models with 3 folds cross validation.</p>\n<p>In the OUTPUT section, you can take files containing the desired train series and their corresponding events.</p>\n<p>Also at the end I put a visualization of the series with events marked. The visualization code was taken from <a href=\"https://www.kaggle.com/code/patrick0302/viz-of-sleeping-time-series\" target=\"_blank\">this excellent notebook</a>.</p>",
  "messages": [
    {
      "id": 2466059,
      "postDate": "2023-10-03T14:21:55.740Z",
      "content": "<p>Hello everyone!</p>\n<p>Until today, I used <a href=\"https://www.kaggle.com/competitions/child-mind-institute-detect-sleep-states/discussion/437345\" target=\"_blank\">this dataset</a>, consisting of 35 series without NaN values.</p>\n<p>To increase the amount of training data, I decided to use more series by truncating them to the first NaN value in the event column.</p>\n<p>My notebook is <a href=\"https://www.kaggle.com/code/kaggledummie007/train-series-of-week-and-longer-w-o-nans\" target=\"_blank\">here</a>.</p>\n<p>In addition, I dropped the series consisting of less than 7 nights so that the model could take into account the influence of the weekend.</p>\n<p>As a result, the <strong>dataset has grown by about 3 times</strong>.</p>\n<p>I split the dataset into 3 sets (lists, actually) of series containing approximately an equal number of nights. I use these sets of series A B C to train models with 3 folds cross validation.</p>\n<p>In the OUTPUT section, you can take files containing the desired train series and their corresponding events.</p>\n<p>Also at the end I put a visualization of the series with events marked. The visualization code was taken from <a href=\"https://www.kaggle.com/code/patrick0302/viz-of-sleeping-time-series\" target=\"_blank\">this excellent notebook</a>.</p>",
      "rawMarkdown": "Hello everyone!\n\nUntil today, I used [this dataset](https://www.kaggle.com/competitions/child-mind-institute-detect-sleep-states/discussion/437345), consisting of 35 series without NaN values.\n\nTo increase the amount of training data, I decided to use more series by truncating them to the first NaN value in the event column.\n\nMy notebook is [here](https://www.kaggle.com/code/kaggledummie007/train-series-of-week-and-longer-w-o-nans).\n\nIn addition, I dropped the series consisting of less than 7 nights so that the model could take into account the influence of the weekend.\n\nAs a result, the **dataset has grown by about 3 times**.\n\nI split the dataset into 3 sets (lists, actually) of series containing approximately an equal number of nights. I use these sets of series A B C to train models with 3 folds cross validation.\n\nIn the OUTPUT section, you can take files containing the desired train series and their corresponding events.\n\nAlso at the end I put a visualization of the series with events marked. The visualization code was taken from [this excellent notebook](https://www.kaggle.com/code/patrick0302/viz-of-sleeping-time-series).\n\n",
      "votes": 3
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2466059": "Hello everyone!\n\nUntil today, I used [this dataset](https://www.kaggle.com/competitions/child-mind-institute-detect-sleep-states/discussion/437345), consisting of 35 series without NaN values.\n\nTo increase the amount of training data, I decided to use more series by truncating them to the first NaN value in the event column.\n\nMy notebook is [here](https://www.kaggle.com/code/kaggledummie007/train-series-of-week-and-longer-w-o-nans).\n\nIn addition, I dropped the series consisting of less than 7 nights so that the model could take into account the influence of the weekend.\n\nAs a result, the **dataset has grown by about 3 times**.\n\nI split the dataset into 3 sets (lists, actually) of series containing approximately an equal number of nights. I use these sets of series A B C to train models with 3 folds cross validation.\n\nIn the OUTPUT section, you can take files containing the desired train series and their corresponding events.\n\nAlso at the end I put a visualization of the series with events marked. The visualization code was taken from [this excellent notebook](https://www.kaggle.com/code/patrick0302/viz-of-sleeping-time-series).\n\n"
  }
}