{
  "id": 56440,
  "title": "A bit funky",
  "url": "/competitions/avito-demand-prediction/discussion/56440",
  "author_name": "",
  "post_date": "2018-05-09T18:55:18.656257500Z",
  "votes": 5,
  "comment_count": 7,
  "views": 0,
  "content": "<p>Although it is a bit weird trying to duplicate an existing model - there seems to be a lot of structure in the predictions</p>\n\n<p>Almost looks like some isotonic regression</p>",
  "messages": [
    {
      "id": "326457",
      "postDate": "05/09/2018 18:55:18",
      "content": "<p>Although it is a bit weird trying to duplicate an existing model - there seems to be a lot of structure in the predictions</p>\n\n<p>Almost looks like some isotonic regression</p>",
      "rawMarkdown": "Although it is a bit weird trying to duplicate an existing model - there seems to be a lot of structure in the predictions\n\nAlmost looks like some isotonic regression",
      "votes": null
    },
    {
      "id": "327021",
      "postDate": "05/10/2018 17:29:31",
      "content": "<p>Are we... trying to predict the behavior of an existing model?  By the way though, I do agree that it appears to be isotonic regression (I realized this before but didn't share): the predictions can be fit over a relatively short domain.</p>",
      "rawMarkdown": "Are we... trying to predict the behavior of an existing model?  By the way though, I do agree that it appears to be isotonic regression (I realized this before but didn't share): the predictions can be fit over a relatively short domain.",
      "votes": null
    },
    {
      "id": "327049",
      "postDate": "05/10/2018 18:29:17",
      "content": "<p>When dealing with probabilities one usually does logloss / auc classifications on targets that are zeros or ones - the fact that we are using regression on values between zero and one leads me to believe we are emulating an existing model</p>",
      "rawMarkdown": "When dealing with probabilities one usually does logloss / auc classifications on targets that are zeros or ones - the fact that we are using regression on values between zero and one leads me to believe we are emulating an existing model",
      "votes": null
    },
    {
      "id": "327066",
      "postDate": "05/10/2018 19:12:45",
      "content": "<p>Good insight. Thank you for building an argument.</p>",
      "rawMarkdown": "Good insight. Thank you for building an argument.",
      "votes": null
    },
    {
      "id": "327090",
      "postDate": "05/10/2018 19:50:59",
      "content": "<p>Wouldn't it be helpful for them to tell us a little bit more about their model and parameters if they want us to build a good model?</p>",
      "rawMarkdown": "Wouldn't it be helpful for them to tell us a little bit more about their model and parameters if they want us to build a good model?",
      "votes": null
    },
    {
      "id": "327095",
      "postDate": "05/10/2018 20:05:39",
      "content": "<p>Part of the fun ;)</p>",
      "rawMarkdown": "Part of the fun ;)",
      "votes": null
    },
    {
      "id": "327106",
      "postDate": "05/10/2018 20:27:23",
      "content": "<p>Just to make sure... y axis is the deal_probability, and red triangles being each item. But what about the blue line ?</p>\n\n<p>Which software are you using ? Seems pretty nice !</p>",
      "rawMarkdown": "Just to make sure... y axis is the deal_probability, and red triangles being each item. But what about the blue line ?\n\nWhich software are you using ? Seems pretty nice !",
      "votes": null
    },
    {
      "id": "327123",
      "postDate": "05/10/2018 21:13:59",
      "content": "<p>Red Triangles are the actual  probabilites and the blue line is the predictions as my genetic program evolves - you can see the output it spits out in the kernels section.</p>",
      "rawMarkdown": "Red Triangles are the actual  probabilites and the blue line is the predictions as my genetic program evolves - you can see the output it spits out in the kernels section.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 327021,
      "author_name": "matthewa313",
      "author_url": "",
      "post_date": "05/10/2018 17:29:31",
      "content": "<p>Are we... trying to predict the behavior of an existing model?  By the way though, I do agree that it appears to be isotonic regression (I realized this before but didn't share): the predictions can be fit over a relatively short domain.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 327049,
      "author_name": "scirpus",
      "author_url": "",
      "post_date": "05/10/2018 18:29:17",
      "content": "<p>When dealing with probabilities one usually does logloss / auc classifications on targets that are zeros or ones - the fact that we are using regression on values between zero and one leads me to believe we are emulating an existing model</p>",
      "votes": null,
      "replies": [
        {
          "id": 327066,
          "author_name": "matthewa313",
          "author_url": "",
          "post_date": "05/10/2018 19:12:45",
          "content": "<p>Good insight. Thank you for building an argument.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 327090,
          "author_name": "matthewa313",
          "author_url": "",
          "post_date": "05/10/2018 19:50:59",
          "content": "<p>Wouldn't it be helpful for them to tell us a little bit more about their model and parameters if they want us to build a good model?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 327095,
          "author_name": "scirpus",
          "author_url": "",
          "post_date": "05/10/2018 20:05:39",
          "content": "<p>Part of the fun ;)</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 327106,
      "author_name": "areveillon",
      "author_url": "",
      "post_date": "05/10/2018 20:27:23",
      "content": "<p>Just to make sure... y axis is the deal_probability, and red triangles being each item. But what about the blue line ?</p>\n\n<p>Which software are you using ? Seems pretty nice !</p>",
      "votes": null,
      "replies": [
        {
          "id": 327123,
          "author_name": "scirpus",
          "author_url": "",
          "post_date": "05/10/2018 21:13:59",
          "content": "<p>Red Triangles are the actual  probabilites and the blue line is the predictions as my genetic program evolves - you can see the output it spits out in the kernels section.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "326457": "Although it is a bit weird trying to duplicate an existing model - there seems to be a lot of structure in the predictions\n\nAlmost looks like some isotonic regression",
    "327021": "Are we... trying to predict the behavior of an existing model?  By the way though, I do agree that it appears to be isotonic regression (I realized this before but didn't share): the predictions can be fit over a relatively short domain.",
    "327049": "When dealing with probabilities one usually does logloss / auc classifications on targets that are zeros or ones - the fact that we are using regression on values between zero and one leads me to believe we are emulating an existing model",
    "327066": "Good insight. Thank you for building an argument.",
    "327090": "Wouldn't it be helpful for them to tell us a little bit more about their model and parameters if they want us to build a good model?",
    "327095": "Part of the fun ;)",
    "327106": "Just to make sure... y axis is the deal_probability, and red triangles being each item. But what about the blue line ?\n\nWhich software are you using ? Seems pretty nice !",
    "327123": "Red Triangles are the actual  probabilites and the blue line is the predictions as my genetic program evolves - you can see the output it spits out in the kernels section."
  },
  "source": "meta"
}