{
  "id": 55922,
  "title": "Little confusion on verification of deal probability ",
  "url": "/competitions/avito-demand-prediction/discussion/55922",
  "author_name": "",
  "post_date": "2018-05-03T06:21:50.494114700Z",
  "votes": 5,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Hi Guys</p>\n\n<p>I got a bit confused and hoping someone can help me understand this. </p>\n\n<p>We are predicting the deal probability for an ad but what's our predicted probability comparing to??</p>\n\n<p>How does Avito generate a 'real deal probability' to verify how accurate our prediction is? </p>",
  "messages": [
    {
      "id": "322531",
      "postDate": "05/03/2018 06:21:50",
      "content": "<p>Hi Guys</p>\n\n<p>I got a bit confused and hoping someone can help me understand this. </p>\n\n<p>We are predicting the deal probability for an ad but what's our predicted probability comparing to??</p>\n\n<p>How does Avito generate a 'real deal probability' to verify how accurate our prediction is? </p>",
      "rawMarkdown": "Hi Guys\n\nI got a bit confused and hoping someone can help me understand this. \n\nWe are predicting the deal probability for an ad but what's our predicted probability comparing to??\n\nHow does Avito generate a 'real deal probability' to verify how accurate our prediction is?",
      "votes": null
    },
    {
      "id": "322713",
      "postDate": "05/03/2018 13:44:39",
      "content": "<p>I will give a try to answer:</p>\n\n<p>deal probability for those cases in test have been collected and saved on server so that we can have public score presented when submitted.</p>\n\n<p>There are various possibilities and reason to say train and test response are different.  However, here, we assume</p>\n\n<ul>\n<li>that the calculation methods of deal probabilities are the same for train and test.  </li>\n<li>train and test are well balanced and split so that true features found in train set can be applied to test set for prediction.</li>\n</ul>\n\n<p>Based on this, we then can say by building a model (features) from training, we can predict deal probabilities for test set.</p>\n\n<p>Hope this answers the questions.</p>",
      "rawMarkdown": "I will give a try to answer:\n\ndeal probability for those cases in test have been collected and saved on server so that we can have public score presented when submitted.\n\nThere are various possibilities and reason to say train and test response are different.  However, here, we assume\n\n - that the calculation methods of deal probabilities are the same for train and test.  \n - train and test are well balanced and split so that true features found in train set can be applied to test set for prediction.\n\nBased on this, we then can say by building a model (features) from training, we can predict deal probabilities for test set.\n\nHope this answers the questions.",
      "votes": null
    },
    {
      "id": "322722",
      "postDate": "05/03/2018 14:05:07",
      "content": "<p>This question confused me a lot when the competition launched. I guess it is similar to <em>Click-through rate (CTR)</em>. It is calculated by the number of times a click is made on the advertisement divided by the total impressions. Then the problem would be can you predict CTR for a new ad.</p>",
      "rawMarkdown": "This question confused me a lot when the competition launched. I guess it is similar to *Click-through rate (CTR)*. It is calculated by the number of times a click is made on the advertisement divided by the total impressions. Then the problem would be can you predict CTR for a new ad.",
      "votes": null
    },
    {
      "id": "322784",
      "postDate": "05/03/2018 16:40:08",
      "content": "<p>Could we get an admin to answer? I'm not sure how helpful it is to just be guessing.</p>",
      "rawMarkdown": "Could we get an admin to answer? I'm not sure how helpful it is to just be guessing.",
      "votes": null
    },
    {
      "id": "322919",
      "postDate": "05/03/2018 23:17:39",
      "content": "<p>It also seemed to me that Avito has already built an ad model (it says that in the data section) and I thought maybe we're just trying to predict its projected <code>deal_probability</code>?</p>",
      "rawMarkdown": "It also seemed to me that Avito has already built an ad model (it says that in the data section) and I thought maybe we're just trying to predict its projected `deal_probability`?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 322713,
      "author_name": "sixianghu",
      "author_url": "",
      "post_date": "05/03/2018 13:44:39",
      "content": "<p>I will give a try to answer:</p>\n\n<p>deal probability for those cases in test have been collected and saved on server so that we can have public score presented when submitted.</p>\n\n<p>There are various possibilities and reason to say train and test response are different.  However, here, we assume</p>\n\n<ul>\n<li>that the calculation methods of deal probabilities are the same for train and test.  </li>\n<li>train and test are well balanced and split so that true features found in train set can be applied to test set for prediction.</li>\n</ul>\n\n<p>Based on this, we then can say by building a model (features) from training, we can predict deal probabilities for test set.</p>\n\n<p>Hope this answers the questions.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 322722,
      "author_name": "cczaixian",
      "author_url": "",
      "post_date": "05/03/2018 14:05:07",
      "content": "<p>This question confused me a lot when the competition launched. I guess it is similar to <em>Click-through rate (CTR)</em>. It is calculated by the number of times a click is made on the advertisement divided by the total impressions. Then the problem would be can you predict CTR for a new ad.</p>",
      "votes": null,
      "replies": [
        {
          "id": 322919,
          "author_name": "matthewa313",
          "author_url": "",
          "post_date": "05/03/2018 23:17:39",
          "content": "<p>It also seemed to me that Avito has already built an ad model (it says that in the data section) and I thought maybe we're just trying to predict its projected <code>deal_probability</code>?</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 322784,
      "author_name": "peterhurford",
      "author_url": "",
      "post_date": "05/03/2018 16:40:08",
      "content": "<p>Could we get an admin to answer? I'm not sure how helpful it is to just be guessing.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "322531": "Hi Guys\n\nI got a bit confused and hoping someone can help me understand this. \n\nWe are predicting the deal probability for an ad but what's our predicted probability comparing to??\n\nHow does Avito generate a 'real deal probability' to verify how accurate our prediction is?",
    "322713": "I will give a try to answer:\n\ndeal probability for those cases in test have been collected and saved on server so that we can have public score presented when submitted.\n\nThere are various possibilities and reason to say train and test response are different.  However, here, we assume\n\n - that the calculation methods of deal probabilities are the same for train and test.  \n - train and test are well balanced and split so that true features found in train set can be applied to test set for prediction.\n\nBased on this, we then can say by building a model (features) from training, we can predict deal probabilities for test set.\n\nHope this answers the questions.",
    "322722": "This question confused me a lot when the competition launched. I guess it is similar to *Click-through rate (CTR)*. It is calculated by the number of times a click is made on the advertisement divided by the total impressions. Then the problem would be can you predict CTR for a new ad.",
    "322784": "Could we get an admin to answer? I'm not sure how helpful it is to just be guessing.",
    "322919": "It also seemed to me that Avito has already built an ad model (it says that in the data section) and I thought maybe we're just trying to predict its projected `deal_probability`?"
  },
  "source": "meta"
}