{
  "id": 57388,
  "title": "Image_top_1 possible explanation",
  "url": "/competitions/avito-demand-prediction/discussion/57388",
  "author_name": "",
  "post_date": "2018-05-23T07:34:52.332823700Z",
  "votes": 8,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Hi!\nIt seems to me that Image_top_1 is a class that indicates the object in the image or classifies the image by the object that is in the image.</p>\n\n<p>My partner ran an algorithm on the images to classify what object is in the image.\nAfter taking the highest probabilities (&gt;=99%) i looked to the data to see what i can find.\nI filtered img object found = \"accordion\" to see all the ads that their image is an accordion and then looked at the distribution of these ads accross Image_top_1.\nindeed far most ads were about 50-50  between 2 specific image_top_1 categories, meaning that in some way, Avito classified maybe 2 types of accordions\\musical instruments.</p>\n\n<p>The image classification my partner did  was not quite helpful for the score because Image_top_1 already contains that data.</p>",
  "messages": [
    {
      "id": "332464",
      "postDate": "05/23/2018 07:34:52",
      "content": "<p>Hi!\nIt seems to me that Image_top_1 is a class that indicates the object in the image or classifies the image by the object that is in the image.</p>\n\n<p>My partner ran an algorithm on the images to classify what object is in the image.\nAfter taking the highest probabilities (&gt;=99%) i looked to the data to see what i can find.\nI filtered img object found = \"accordion\" to see all the ads that their image is an accordion and then looked at the distribution of these ads accross Image_top_1.\nindeed far most ads were about 50-50  between 2 specific image_top_1 categories, meaning that in some way, Avito classified maybe 2 types of accordions\\musical instruments.</p>\n\n<p>The image classification my partner did  was not quite helpful for the score because Image_top_1 already contains that data.</p>",
      "rawMarkdown": "Hi!\nIt seems to me that Image_top_1 is a class that indicates the object in the image or classifies the image by the object that is in the image.\n\nMy partner ran an algorithm on the images to classify what object is in the image.\nAfter taking the highest probabilities (&gt;=99%) i looked to the data to see what i can find.\nI filtered img object found = \"accordion\" to see all the ads that their image is an accordion and then looked at the distribution of these ads accross Image_top_1.\nindeed far most ads were about 50-50  between 2 specific image_top_1 categories, meaning that in some way, Avito classified maybe 2 types of accordions\\musical instruments.\n\nThe image classification my partner did  was not quite helpful for the score because Image_top_1 already contains that data.",
      "votes": null
    },
    {
      "id": "332851",
      "postDate": "05/23/2018 22:36:55",
      "content": "<p>indeed, when I was doing my eda, I found there are about 3,000 image_top_1 values. Then I plotted image_top_1 == 2219.0, these images are living rooms. </p>",
      "rawMarkdown": "indeed, when I was doing my eda, I found there are about 3,000 image_top_1 values. Then I plotted image_top_1 == 2219.0, these images are living rooms.",
      "votes": null
    },
    {
      "id": "332852",
      "postDate": "05/23/2018 22:40:11",
      "content": "<p>Does anyone know if the numbers represent some sort of similarity? Is image_top_1 #2219 really similar to #2218 but very different from #22?</p>",
      "rawMarkdown": "Does anyone know if the numbers represent some sort of similarity? Is image_top_1 #2219 really similar to #2218 but very different from #22?",
      "votes": null
    },
    {
      "id": "332853",
      "postDate": "05/23/2018 22:50:13",
      "content": "<p>2219 are living rooms, 2218 are buildings. Each number represents a different type of object.</p>",
      "rawMarkdown": "2219 are living rooms, 2218 are buildings. Each number represents a different type of object.",
      "votes": null
    },
    {
      "id": "332854",
      "postDate": "05/23/2018 22:56:50",
      "content": "<p>Living rooms and buildings are pretty similar, though. So that would provide evidence for my hypothesis and also explain why image_top_1 seems useful as both numeric and categorical.</p>",
      "rawMarkdown": "Living rooms and buildings are pretty similar, though. So that would provide evidence for my hypothesis and also explain why image_top_1 seems useful as both numeric and categorical.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 332851,
      "author_name": "konohayui",
      "author_url": "",
      "post_date": "05/23/2018 22:36:55",
      "content": "<p>indeed, when I was doing my eda, I found there are about 3,000 image_top_1 values. Then I plotted image_top_1 == 2219.0, these images are living rooms. </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 332852,
      "author_name": "peterhurford",
      "author_url": "",
      "post_date": "05/23/2018 22:40:11",
      "content": "<p>Does anyone know if the numbers represent some sort of similarity? Is image_top_1 #2219 really similar to #2218 but very different from #22?</p>",
      "votes": null,
      "replies": [
        {
          "id": 332853,
          "author_name": "konohayui",
          "author_url": "",
          "post_date": "05/23/2018 22:50:13",
          "content": "<p>2219 are living rooms, 2218 are buildings. Each number represents a different type of object.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 332854,
          "author_name": "peterhurford",
          "author_url": "",
          "post_date": "05/23/2018 22:56:50",
          "content": "<p>Living rooms and buildings are pretty similar, though. So that would provide evidence for my hypothesis and also explain why image_top_1 seems useful as both numeric and categorical.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "332464": "Hi!\nIt seems to me that Image_top_1 is a class that indicates the object in the image or classifies the image by the object that is in the image.\n\nMy partner ran an algorithm on the images to classify what object is in the image.\nAfter taking the highest probabilities (&gt;=99%) i looked to the data to see what i can find.\nI filtered img object found = \"accordion\" to see all the ads that their image is an accordion and then looked at the distribution of these ads accross Image_top_1.\nindeed far most ads were about 50-50  between 2 specific image_top_1 categories, meaning that in some way, Avito classified maybe 2 types of accordions\\musical instruments.\n\nThe image classification my partner did  was not quite helpful for the score because Image_top_1 already contains that data.",
    "332851": "indeed, when I was doing my eda, I found there are about 3,000 image_top_1 values. Then I plotted image_top_1 == 2219.0, these images are living rooms.",
    "332852": "Does anyone know if the numbers represent some sort of similarity? Is image_top_1 #2219 really similar to #2218 but very different from #22?",
    "332853": "2219 are living rooms, 2218 are buildings. Each number represents a different type of object.",
    "332854": "Living rooms and buildings are pretty similar, though. So that would provide evidence for my hypothesis and also explain why image_top_1 seems useful as both numeric and categorical."
  },
  "source": "meta"
}