{
  "id": 24523,
  "title": "@Faron, Feature Evaluation",
  "url": "/competitions/bosch-production-line-performance/discussion/24523",
  "author_name": "neuron",
  "post_date": "2016-10-18T02:21:58.437000",
  "votes": 4,
  "comment_count": 0,
  "views": 587,
  "content": "<p>Faron has mentioned this in various topics</p>\n\n<p>To evaluate a single feature</p>\n\n<ol>\n<li>Train xgb on that feature.</li>\n<li>gini score on binned feature</li>\n</ol>\n\n<p>I have few questions</p>\n\n<ol>\n<li><p>When we train xgb on a single feature, how to choose hyper parameters. is there any chance of overfiting. Theoritically we can't overfit to 100 % if not all values are distinct.</p></li>\n<li><p>how bin a feature ? is it fixed number bins or fixed interval ? Once binned, how to calculate the score ? is it calculate gini on each bin and add all ?</p></li>\n</ol>",
  "messages": [
    {
      "id": 140042,
      "postDate": "2016-10-18T02:21:58.437Z",
      "content": "<p>Faron has mentioned this in various topics</p>\n\n<p>To evaluate a single feature</p>\n\n<ol>\n<li>Train xgb on that feature.</li>\n<li>gini score on binned feature</li>\n</ol>\n\n<p>I have few questions</p>\n\n<ol>\n<li><p>When we train xgb on a single feature, how to choose hyper parameters. is there any chance of overfiting. Theoritically we can't overfit to 100 % if not all values are distinct.</p></li>\n<li><p>how bin a feature ? is it fixed number bins or fixed interval ? Once binned, how to calculate the score ? is it calculate gini on each bin and add all ?</p></li>\n</ol>",
      "rawMarkdown": "Faron has mentioned this in various topics\r\n\r\nTo evaluate a single feature\r\n\r\n1. Train xgb on that feature.\r\n2. gini score on binned feature\r\n\r\n\r\nI have few questions\r\n\r\n1. When we train xgb on a single feature, how to choose hyper parameters. is there any chance of overfiting. Theoritically we can't overfit to 100 % if not all values are distinct.\r\n\r\n2. how bin a feature ? is it fixed number bins or fixed interval ? Once binned, how to calculate the score ? is it calculate gini on each bin and add all ?",
      "votes": 4
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "140042": "Faron has mentioned this in various topics\r\n\r\nTo evaluate a single feature\r\n\r\n1. Train xgb on that feature.\r\n2. gini score on binned feature\r\n\r\n\r\nI have few questions\r\n\r\n1. When we train xgb on a single feature, how to choose hyper parameters. is there any chance of overfiting. Theoritically we can't overfit to 100 % if not all values are distinct.\r\n\r\n2. how bin a feature ? is it fixed number bins or fixed interval ? Once binned, how to calculate the score ? is it calculate gini on each bin and add all ?"
  }
}