{
  "id": 22684,
  "title": "xgboost warm start",
  "url": "/competitions/grupo-bimbo-inventory-demand/discussion/22684",
  "author_name": "",
  "post_date": "2016-08-04T11:04:22.327Z",
  "votes": null,
  "comment_count": 1,
  "views": 685,
  "content": "<p>Is there a parameter &quot;warm start&quot; in xgboost ?? </p>\n\n<p>warm_start : bool, default: False\nWhen set to True, reuse the solution of the previous call to fit and add more estimators to the ensemble, otherwise, just erase the previous solution.</p>\n\n<p>Thanks</p>",
  "messages": [
    {
      "id": "130145",
      "postDate": "08/04/2016 11:04:22",
      "content": "<p>Is there a parameter &quot;warm start&quot; in xgboost ?? </p>\n\n<p>warm_start : bool, default: False\nWhen set to True, reuse the solution of the previous call to fit and add more estimators to the ensemble, otherwise, just erase the previous solution.</p>\n\n<p>Thanks</p>",
      "rawMarkdown": "Is there a parameter \"warm start\" in xgboost ?? \r\n\r\nwarm_start : bool, default: False\r\nWhen set to True, reuse the solution of the previous call to fit and add more estimators to the ensemble, otherwise, just erase the previous solution.\r\n\r\nThanks",
      "votes": null
    },
    {
      "id": "130214",
      "postDate": "08/04/2016 17:30:12",
      "content": "<p>yes, there is something like that. When you set up the Dmatrix, in R: </p>\n\n<pre><code>xgb.DMatrix(data = your_train_data, label = y,  base_margin= your_previous result)\n</code></pre>\n\n<p>That previous_result should be a raw result that can be obtained from the predict  call with a flag <code>outputmargin=TRUE</code></p>",
      "rawMarkdown": "yes, there is something like that. When you set up the Dmatrix, in R: \r\n\r\n    xgb.DMatrix(data = your_train_data, label = y,  base_margin= your_previous result)\r\n\r\nThat previous_result should be a raw result that can be obtained from the predict  call with a flag `outputmargin=TRUE`",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 130214,
      "author_name": "mrooijer",
      "author_url": "",
      "post_date": "08/04/2016 17:30:12",
      "content": "<p>yes, there is something like that. When you set up the Dmatrix, in R: </p>\n\n<pre><code>xgb.DMatrix(data = your_train_data, label = y,  base_margin= your_previous result)\n</code></pre>\n\n<p>That previous_result should be a raw result that can be obtained from the predict  call with a flag <code>outputmargin=TRUE</code></p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "130145": "Is there a parameter \"warm start\" in xgboost ?? \r\n\r\nwarm_start : bool, default: False\r\nWhen set to True, reuse the solution of the previous call to fit and add more estimators to the ensemble, otherwise, just erase the previous solution.\r\n\r\nThanks",
    "130214": "yes, there is something like that. When you set up the Dmatrix, in R: \r\n\r\n    xgb.DMatrix(data = your_train_data, label = y,  base_margin= your_previous result)\r\n\r\nThat previous_result should be a raw result that can be obtained from the predict  call with a flag `outputmargin=TRUE`"
  },
  "source": "meta"
}