{
  "id": 12540,
  "title": "Vowpal Wabbit + oaa + logistic loss",
  "url": "/competitions/malware-classification/discussion/12540",
  "author_name": "",
  "post_date": "2015-02-17T17:18:26.780Z",
  "votes": null,
  "comment_count": 3,
  "views": 2082,
  "content": "<p>Decided to try out vw for the first time. How do you guys use multiclass classification with logistic loss? From old blog posts it seems that something like that would work:</p>\n<p>vw -d train.vw -b 30 --oaa 9 --loss_function logistic --passes 20 -f predictor.vw --link logistic</p>\n<p>vw -d test.vw -t -i predictor.vw -p predictions.txt -r raw_predictions.txt</p>\n<p>But that generates an empty raw_predictions file. If I remove logistic loss then it works fine, but of course giving me much worse score than if I'd done it with python.</p>\n<p>Do you calculate logistic loss manually from a raw_predictions.txt or there is some trick to do it with multiclass classification in vw?</p>",
  "messages": [
    {
      "id": "64405",
      "postDate": "02/17/2015 17:18:26",
      "content": "<p>Decided to try out vw for the first time. How do you guys use multiclass classification with logistic loss? From old blog posts it seems that something like that would work:</p>\n<p>vw -d train.vw -b 30 --oaa 9 --loss_function logistic --passes 20 -f predictor.vw --link logistic</p>\n<p>vw -d test.vw -t -i predictor.vw -p predictions.txt -r raw_predictions.txt</p>\n<p>But that generates an empty raw_predictions file. If I remove logistic loss then it works fine, but of course giving me much worse score than if I'd done it with python.</p>\n<p>Do you calculate logistic loss manually from a raw_predictions.txt or there is some trick to do it with multiclass classification in vw?</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "64428",
      "postDate": "02/17/2015 21:18:37",
      "content": "<p>Try removing the examples from the test command (-d test.vw).</p>\n<p>Does that make it work?</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "64430",
      "postDate": "02/17/2015 21:20:47",
      "content": "<p>What I've figured out that it works if I remove --passes :/</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "64431",
      "postDate": "02/17/2015 21:34:56",
      "content": "<p>Looks like it's a bug&nbsp;https://github.com/JohnLangford/vowpal_wabbit/issues/522</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 64428,
      "author_name": "ericwhyne",
      "author_url": "",
      "post_date": "02/17/2015 21:18:37",
      "content": "<p>Try removing the examples from the test command (-d test.vw).</p>\n<p>Does that make it work?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 64430,
      "author_name": "yankov",
      "author_url": "",
      "post_date": "02/17/2015 21:20:47",
      "content": "<p>What I've figured out that it works if I remove --passes :/</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 64431,
      "author_name": "yankov",
      "author_url": "",
      "post_date": "02/17/2015 21:34:56",
      "content": "<p>Looks like it's a bug&nbsp;https://github.com/JohnLangford/vowpal_wabbit/issues/522</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "64405": "",
    "64428": "",
    "64430": "",
    "64431": ""
  },
  "source": "meta"
}