{
  "id": 12412,
  "title": "R glmnet",
  "url": "/competitions/avazu-ctr-prediction/discussion/12412",
  "author_name": "",
  "post_date": "2015-02-04T23:48:30.660Z",
  "votes": 2,
  "comment_count": 1,
  "views": 1194,
  "content": "<p>Hi all,</p>\n<p>Has anyone tried R glmnet for this competition?</p>\n<p>It looks good match to me but I don't have much experience with glmnet or R.</p>\n<p>Best,</p>\n<p>C&nbsp;</p>",
  "messages": [
    {
      "id": "63620",
      "postDate": "02/04/2015 23:48:30",
      "content": "<p>Hi all,</p>\n<p>Has anyone tried R glmnet for this competition?</p>\n<p>It looks good match to me but I don't have much experience with glmnet or R.</p>\n<p>Best,</p>\n<p>C&nbsp;</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "63624",
      "postDate": "02/05/2015 01:14:58",
      "content": "<p>I've used it, it works fine. It was actually the first model i've tried and got 0.40 in the LB.</p>\n<p>Here is a link to the package intro:</p>\n<p>http://web.stanford.edu/~hastie/glmnet/glmnet_alpha.html</p>\n<p>I suggest the following getting started&nbsp;&quot;workflow&quot;:&nbsp;</p>\n<p>1) Read a subset of the training dataset</p>\n<p>2) Convert the variables to factor</p>\n<p>3)&nbsp;train =&nbsp;sparse.model.matrix(click ~ ., data=train)</p>\n<p>to create your&nbsp;model matrix</p>\n<p>4) Try the lasso (alpha = 1), perhaps using interactions of a few selected variables</p>\n\n<p>After that I've been using VW and python for logistic regression. As they are not memory bounded, you can use a larger sample of the data.</p>\n<p>hope it helps!</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 63624,
      "author_name": "marbel",
      "author_url": "",
      "post_date": "02/05/2015 01:14:58",
      "content": "<p>I've used it, it works fine. It was actually the first model i've tried and got 0.40 in the LB.</p>\n<p>Here is a link to the package intro:</p>\n<p>http://web.stanford.edu/~hastie/glmnet/glmnet_alpha.html</p>\n<p>I suggest the following getting started&nbsp;&quot;workflow&quot;:&nbsp;</p>\n<p>1) Read a subset of the training dataset</p>\n<p>2) Convert the variables to factor</p>\n<p>3)&nbsp;train =&nbsp;sparse.model.matrix(click ~ ., data=train)</p>\n<p>to create your&nbsp;model matrix</p>\n<p>4) Try the lasso (alpha = 1), perhaps using interactions of a few selected variables</p>\n\n<p>After that I've been using VW and python for logistic regression. As they are not memory bounded, you can use a larger sample of the data.</p>\n<p>hope it helps!</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "63620": "",
    "63624": ""
  },
  "source": "meta"
}