{
  "id": 21243,
  "title": "Anyone using XGBoost could share their optimal parameters?",
  "url": "/competitions/expedia-hotel-recommendations/discussion/21243",
  "author_name": "",
  "post_date": "2016-05-26T12:59:27.107Z",
  "votes": null,
  "comment_count": 2,
  "views": 608,
  "content": "<p>Hello all</p>\n\n<p>Anyone using XGBoost could share their optimal parameters? </p>\n\n<p>It's taking me hours (or days if i use whole training set) to even train a XGB model on default parameters and nrounds = 5. I don't think my machine is up for hypertuning the parameters.</p>",
  "messages": [
    {
      "id": "121451",
      "postDate": "05/26/2016 12:59:27",
      "content": "<p>Hello all</p>\n\n<p>Anyone using XGBoost could share their optimal parameters? </p>\n\n<p>It's taking me hours (or days if i use whole training set) to even train a XGB model on default parameters and nrounds = 5. I don't think my machine is up for hypertuning the parameters.</p>",
      "rawMarkdown": "Hello all\r\n\r\nAnyone using XGBoost could share their optimal parameters? \r\n\r\nIt's taking me hours (or days if i use whole training set) to even train a XGB model on default parameters and nrounds = 5. I don't think my machine is up for hypertuning the parameters.",
      "votes": null
    },
    {
      "id": "121471",
      "postDate": "05/26/2016 17:11:48",
      "content": "<p>Wish I could help you, I'm using my own XGBoost implementation (daBoost - it's all about daBoost) and settings aren't really comparable.</p>\n\n<p>But I find that it stops learning quite fast even at a low learning rate. </p>\n\n<p>Here's my settings though;</p>\n\n<blockquote>\n<pre><code>         Forest forest = new Forest(6, training.First().FeatureValues.Length, Objective.LogisticClassification);\n            forest.LearningRate = 0.075f;\n            forest.FractionOfFeaturesToUse = 0.65f;\n            forest.FractionOfTrainingDataPerRandomTree = 0.2f;\n</code></pre>\n</blockquote>",
      "rawMarkdown": "Wish I could help you, I'm using my own XGBoost implementation (daBoost - it's all about daBoost) and settings aren't really comparable.\r\n\r\nBut I find that it stops learning quite fast even at a low learning rate. \r\n\r\nHere's my settings though;\r\n\r\n>              Forest forest = new Forest(6, training.First().FeatureValues.Length, Objective.LogisticClassification);\r\n                forest.LearningRate = 0.075f;\r\n                forest.FractionOfFeaturesToUse = 0.65f;\r\n                forest.FractionOfTrainingDataPerRandomTree = 0.2f;",
      "votes": null
    },
    {
      "id": "121472",
      "postDate": "05/26/2016 17:39:49",
      "content": "<p>@Mattias </p>\n\n<p>Thanks for the input! I gave up my machine, I tried training it (on full train set) and predicting the the full test set on nrounds = 2 - killed the kernal repeatedly.</p>\n\n<p>Now I am setting up a super machine on AWS and downloading the data set using Terminal web browser Lynx. I will just do my magic in my R3x2Large server - 8 cores, 61Gb ram!</p>",
      "rawMarkdown": "Mattias \r\n\r\nThanks for the input! I gave up my machine, I tried training it (on full train set) and predicting the the full test set on nrounds = 2 - killed the kernal repeatedly.\r\n\r\nNow I am setting up a super machine on AWS and downloading the data set using Terminal web browser Lynx. I will just do my magic in my R3x2Large server - 8 cores, 61Gb ram!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 121471,
      "author_name": "mfagerlund",
      "author_url": "",
      "post_date": "05/26/2016 17:11:48",
      "content": "<p>Wish I could help you, I'm using my own XGBoost implementation (daBoost - it's all about daBoost) and settings aren't really comparable.</p>\n\n<p>But I find that it stops learning quite fast even at a low learning rate. </p>\n\n<p>Here's my settings though;</p>\n\n<blockquote>\n<pre><code>         Forest forest = new Forest(6, training.First().FeatureValues.Length, Objective.LogisticClassification);\n            forest.LearningRate = 0.075f;\n            forest.FractionOfFeaturesToUse = 0.65f;\n            forest.FractionOfTrainingDataPerRandomTree = 0.2f;\n</code></pre>\n</blockquote>",
      "votes": null,
      "replies": []
    },
    {
      "id": 121472,
      "author_name": "zyazzy",
      "author_url": "",
      "post_date": "05/26/2016 17:39:49",
      "content": "<p>@Mattias </p>\n\n<p>Thanks for the input! I gave up my machine, I tried training it (on full train set) and predicting the the full test set on nrounds = 2 - killed the kernal repeatedly.</p>\n\n<p>Now I am setting up a super machine on AWS and downloading the data set using Terminal web browser Lynx. I will just do my magic in my R3x2Large server - 8 cores, 61Gb ram!</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "121451": "Hello all\r\n\r\nAnyone using XGBoost could share their optimal parameters? \r\n\r\nIt's taking me hours (or days if i use whole training set) to even train a XGB model on default parameters and nrounds = 5. I don't think my machine is up for hypertuning the parameters.",
    "121471": "Wish I could help you, I'm using my own XGBoost implementation (daBoost - it's all about daBoost) and settings aren't really comparable.\r\n\r\nBut I find that it stops learning quite fast even at a low learning rate. \r\n\r\nHere's my settings though;\r\n\r\n>              Forest forest = new Forest(6, training.First().FeatureValues.Length, Objective.LogisticClassification);\r\n                forest.LearningRate = 0.075f;\r\n                forest.FractionOfFeaturesToUse = 0.65f;\r\n                forest.FractionOfTrainingDataPerRandomTree = 0.2f;",
    "121472": "Mattias \r\n\r\nThanks for the input! I gave up my machine, I tried training it (on full train set) and predicting the the full test set on nrounds = 2 - killed the kernal repeatedly.\r\n\r\nNow I am setting up a super machine on AWS and downloading the data set using Terminal web browser Lynx. I will just do my magic in my R3x2Large server - 8 cores, 61Gb ram!"
  },
  "source": "meta"
}