{
  "id": 76097,
  "title": "Best linear model",
  "url": "/competitions/quora-insincere-questions-classification/discussion/76097",
  "author_name": "",
  "post_date": "2018-12-29T10:00:56.035535800Z",
  "votes": null,
  "comment_count": 3,
  "views": 0,
  "content": "<p>How is your best linear model doing?</p>\n\n<p>I think it is convenient to have a solid linear model since it won't correlate a lot with neural nets predictions and will help in the blending stage. My best linear model <a href=\"https://www.kaggle.com/david26694/nb-svm-baseline-trigrams\">(NBSVM with trigrams)</a>  is doing 0.63 in validation, and in leaderboard too, using a naive bayes + logistic regression, trigrams, and only 50000 words. Any ideas on possible improvements?</p>",
  "messages": [
    {
      "id": "447177",
      "postDate": "12/29/2018 10:00:56",
      "content": "<p>How is your best linear model doing?</p>\n\n<p>I think it is convenient to have a solid linear model since it won't correlate a lot with neural nets predictions and will help in the blending stage. My best linear model <a href=\"https://www.kaggle.com/david26694/nb-svm-baseline-trigrams\">(NBSVM with trigrams)</a>  is doing 0.63 in validation, and in leaderboard too, using a naive bayes + logistic regression, trigrams, and only 50000 words. Any ideas on possible improvements?</p>",
      "rawMarkdown": "How is your best linear model doing?\n\nI think it is convenient to have a solid linear model since it won't correlate a lot with neural nets predictions and will help in the blending stage. My best linear model [(NBSVM with trigrams)][1]  is doing 0.63 in validation, and in leaderboard too, using a naive bayes + logistic regression, trigrams, and only 50000 words. Any ideas on possible improvements?\n\n\n  [1]: https://www.kaggle.com/david26694/nb-svm-baseline-trigrams",
      "votes": null
    },
    {
      "id": "447758",
      "postDate": "12/30/2018 13:38:01",
      "content": "<p>The problem is that it doesn't really compare to any well trained neural net, so this is like a dead end for me.</p>",
      "rawMarkdown": "The problem is that it doesn't really compare to any well trained neural net, so this is like a dead end for me.",
      "votes": null
    },
    {
      "id": "447775",
      "postDate": "12/30/2018 14:27:03",
      "content": "<p>0.64something. single model</p>",
      "rawMarkdown": "0.64something. single model",
      "votes": null
    },
    {
      "id": "447780",
      "postDate": "12/30/2018 14:58:25",
      "content": "<p>and do you manage to blend it with something else?</p>",
      "rawMarkdown": "and do you manage to blend it with something else?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 447758,
      "author_name": "david26694",
      "author_url": "",
      "post_date": "12/30/2018 13:38:01",
      "content": "<p>The problem is that it doesn't really compare to any well trained neural net, so this is like a dead end for me.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 447775,
      "author_name": "abhishek",
      "author_url": "",
      "post_date": "12/30/2018 14:27:03",
      "content": "<p>0.64something. single model</p>",
      "votes": null,
      "replies": [
        {
          "id": 447780,
          "author_name": "david26694",
          "author_url": "",
          "post_date": "12/30/2018 14:58:25",
          "content": "<p>and do you manage to blend it with something else?</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "447177": "How is your best linear model doing?\n\nI think it is convenient to have a solid linear model since it won't correlate a lot with neural nets predictions and will help in the blending stage. My best linear model [(NBSVM with trigrams)][1]  is doing 0.63 in validation, and in leaderboard too, using a naive bayes + logistic regression, trigrams, and only 50000 words. Any ideas on possible improvements?\n\n\n  [1]: https://www.kaggle.com/david26694/nb-svm-baseline-trigrams",
    "447758": "The problem is that it doesn't really compare to any well trained neural net, so this is like a dead end for me.",
    "447775": "0.64something. single model",
    "447780": "and do you manage to blend it with something else?"
  },
  "source": "meta"
}