{
  "id": 10785,
  "title": "slow sk-learn learning ",
  "url": "/competitions/avazu-ctr-prediction/discussion/10785",
  "author_name": "",
  "post_date": "2014-10-29T14:25:58.433Z",
  "votes": null,
  "comment_count": 7,
  "views": 2069,
  "content": "<p>Not sure if this belongs here.</p>\n<p>Hi All,</p>\n<p>I am new to the machine learning and I am trying to use sk-learn for this competition and my mac doesn't have enough memory to learn from this big file.</p>\n<p>I was thinking of using an AWS machine to learn. I was wondering if you guys have any better ideas to do machine learning faster.</p>\n<p>Thanks,</p>\n<p>Ujwal</p>",
  "messages": [
    {
      "id": "56987",
      "postDate": "10/29/2014 14:25:58",
      "content": "<p>Not sure if this belongs here.</p>\n<p>Hi All,</p>\n<p>I am new to the machine learning and I am trying to use sk-learn for this competition and my mac doesn't have enough memory to learn from this big file.</p>\n<p>I was thinking of using an AWS machine to learn. I was wondering if you guys have any better ideas to do machine learning faster.</p>\n<p>Thanks,</p>\n<p>Ujwal</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "56989",
      "postDate": "10/29/2014 14:47:28",
      "content": "<p>Two words: online learning. Vowpal Wabbit would be my first suggestion.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "56991",
      "postDate": "10/29/2014 15:09:28",
      "content": "<p>Hi Konrad, when you say online learning you mean cloud servers?&nbsp;</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "56992",
      "postDate": "10/29/2014 15:14:00",
      "content": "<p>No, algorithms that do not require loading everything to memory in one go:&nbsp;</p>\n\n<p><a href=\"http://en.wikipedia.org/wiki/Online_machine_learning\">http://en.wikipedia.org/wiki/Online_machine_learning</a></p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "56993",
      "postDate": "10/29/2014 15:47:27",
      "content": "<p>Vowpal Wabbit is good but it won't win this competition. You need some non linear model IMO.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "56995",
      "postDate": "10/29/2014 15:51:45",
      "content": "<p>@Pawel: true, but:</p>\n<p>1. unless someone has massive hardware at their disposal, bagging + feature engineering seem (from the top of my head) like the only way to get this to run e.g. in sklearn&nbsp;</p>\n<p>2. nn + interactions in vowpal can get you quite far, imo</p>\n<p>3. one has to start somewhere...</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "57005",
      "postDate": "10/29/2014 17:58:36",
      "content": "<p>@Pawel what do you recommend for non-linear model that could be fast?</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "57036",
      "postDate": "10/30/2014 08:13:22",
      "content": "<p>xgboost is fast. I recommend starting the model on some sample and measure the performance increase you get when you increase the sample size.</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 56989,
      "author_name": "konradb",
      "author_url": "",
      "post_date": "10/29/2014 14:47:28",
      "content": "<p>Two words: online learning. Vowpal Wabbit would be my first suggestion.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 56991,
      "author_name": "ujwaltrivedi",
      "author_url": "",
      "post_date": "10/29/2014 15:09:28",
      "content": "<p>Hi Konrad, when you say online learning you mean cloud servers?&nbsp;</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 56992,
      "author_name": "konradb",
      "author_url": "",
      "post_date": "10/29/2014 15:14:00",
      "content": "<p>No, algorithms that do not require loading everything to memory in one go:&nbsp;</p>\n\n<p><a href=\"http://en.wikipedia.org/wiki/Online_machine_learning\">http://en.wikipedia.org/wiki/Online_machine_learning</a></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 56993,
      "author_name": "paweljankiewicz",
      "author_url": "",
      "post_date": "10/29/2014 15:47:27",
      "content": "<p>Vowpal Wabbit is good but it won't win this competition. You need some non linear model IMO.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 56995,
      "author_name": "konradb",
      "author_url": "",
      "post_date": "10/29/2014 15:51:45",
      "content": "<p>@Pawel: true, but:</p>\n<p>1. unless someone has massive hardware at their disposal, bagging + feature engineering seem (from the top of my head) like the only way to get this to run e.g. in sklearn&nbsp;</p>\n<p>2. nn + interactions in vowpal can get you quite far, imo</p>\n<p>3. one has to start somewhere...</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 57005,
      "author_name": "ujwaltrivedi",
      "author_url": "",
      "post_date": "10/29/2014 17:58:36",
      "content": "<p>@Pawel what do you recommend for non-linear model that could be fast?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 57036,
      "author_name": "paweljankiewicz",
      "author_url": "",
      "post_date": "10/30/2014 08:13:22",
      "content": "<p>xgboost is fast. I recommend starting the model on some sample and measure the performance increase you get when you increase the sample size.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "56987": "",
    "56989": "",
    "56991": "",
    "56992": "",
    "56993": "",
    "56995": "",
    "57005": "",
    "57036": ""
  },
  "source": "meta"
}