{
  "id": 78967,
  "title": "Anyone is thinking post-processing has a significant role?",
  "url": "/competitions/quora-insincere-questions-classification/discussion/78967",
  "author_name": "",
  "post_date": "2019-01-29T14:57:30.223948300Z",
  "votes": 2,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Just to have some nice discussion and ideas :-)</p>",
  "messages": [
    {
      "id": "463165",
      "postDate": "01/29/2019 14:57:30",
      "content": "<p>Just to have some nice discussion and ideas :-)</p>",
      "rawMarkdown": "Just to have some nice discussion and ideas :-)",
      "votes": null
    },
    {
      "id": "463195",
      "postDate": "01/29/2019 15:41:19",
      "content": "<p>I've not thought about it's application on the Quora challenge, but on other datasets I think there's definitely a place for it. I recall looking into it once when predicting star ratings on Yelp reviews, technically the stars are categorical variables, making it a classification problem, but there are regression-like features inherent in them (i.e. 5&gt;4&gt;3&gt;2&gt;1 star, which is unlike most classification problems). I did a decent amount of looking into \"latent class regression\" and other post-processing manipulation of predictions but ended up not implementing it because I couldn't find enough research to support such behavior. </p>",
      "rawMarkdown": "I've not thought about it's application on the Quora challenge, but on other datasets I think there's definitely a place for it. I recall looking into it once when predicting star ratings on Yelp reviews, technically the stars are categorical variables, making it a classification problem, but there are regression-like features inherent in them (i.e. 5&gt;4&gt;3&gt;2&gt;1 star, which is unlike most classification problems). I did a decent amount of looking into \"latent class regression\" and other post-processing manipulation of predictions but ended up not implementing it because I couldn't find enough research to support such behavior.",
      "votes": null
    },
    {
      "id": "465306",
      "postDate": "02/02/2019 19:27:19",
      "content": "<p>you mean post- process to find the best thredshold for diffrent \"group\" of sentences?</p>",
      "rawMarkdown": "you mean post- process to find the best thredshold for diffrent \"group\" of sentences?",
      "votes": null
    },
    {
      "id": "465418",
      "postDate": "02/03/2019 03:45:52",
      "content": "<p>yeah, one thing is that. Another thing would be trying to find an approx distribution of the test data and making some assumptions on that etc etc, people make some sort of processing based on that on other competitions. But, I am not familiar with that much.</p>",
      "rawMarkdown": "yeah, one thing is that. Another thing would be trying to find an approx distribution of the test data and making some assumptions on that etc etc, people make some sort of processing based on that on other competitions. But, I am not familiar with that much.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 463195,
      "author_name": "alecthekulak",
      "author_url": "",
      "post_date": "01/29/2019 15:41:19",
      "content": "<p>I've not thought about it's application on the Quora challenge, but on other datasets I think there's definitely a place for it. I recall looking into it once when predicting star ratings on Yelp reviews, technically the stars are categorical variables, making it a classification problem, but there are regression-like features inherent in them (i.e. 5&gt;4&gt;3&gt;2&gt;1 star, which is unlike most classification problems). I did a decent amount of looking into \"latent class regression\" and other post-processing manipulation of predictions but ended up not implementing it because I couldn't find enough research to support such behavior. </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 465306,
      "author_name": "tiopon",
      "author_url": "",
      "post_date": "02/02/2019 19:27:19",
      "content": "<p>you mean post- process to find the best thredshold for diffrent \"group\" of sentences?</p>",
      "votes": null,
      "replies": [
        {
          "id": 465418,
          "author_name": "s4sarath",
          "author_url": "",
          "post_date": "02/03/2019 03:45:52",
          "content": "<p>yeah, one thing is that. Another thing would be trying to find an approx distribution of the test data and making some assumptions on that etc etc, people make some sort of processing based on that on other competitions. But, I am not familiar with that much.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "463165": "Just to have some nice discussion and ideas :-)",
    "463195": "I've not thought about it's application on the Quora challenge, but on other datasets I think there's definitely a place for it. I recall looking into it once when predicting star ratings on Yelp reviews, technically the stars are categorical variables, making it a classification problem, but there are regression-like features inherent in them (i.e. 5&gt;4&gt;3&gt;2&gt;1 star, which is unlike most classification problems). I did a decent amount of looking into \"latent class regression\" and other post-processing manipulation of predictions but ended up not implementing it because I couldn't find enough research to support such behavior.",
    "465306": "you mean post- process to find the best thredshold for diffrent \"group\" of sentences?",
    "465418": "yeah, one thing is that. Another thing would be trying to find an approx distribution of the test data and making some assumptions on that etc etc, people make some sort of processing based on that on other competitions. But, I am not familiar with that much."
  },
  "source": "meta"
}