{
  "id": 59022,
  "title": "I wonder the limitation score without image features using lgb or nn",
  "url": "/competitions/avito-demand-prediction/discussion/59022",
  "author_name": "",
  "post_date": "2018-06-17T05:05:45.105215300Z",
  "votes": null,
  "comment_count": 8,
  "views": 0,
  "content": "<p>Because the image dataset is so large that I cannot handle it in my limited computer resources, So I want to know the best score you get if you did not use the image features. my best score is 2220 about for nn and lgbm, and I cannot make them to get better score.</p>",
  "messages": [
    {
      "id": "344153",
      "postDate": "06/17/2018 05:05:45",
      "content": "<p>Because the image dataset is so large that I cannot handle it in my limited computer resources, So I want to know the best score you get if you did not use the image features. my best score is 2220 about for nn and lgbm, and I cannot make them to get better score.</p>",
      "rawMarkdown": "Because the image dataset is so large that I cannot handle it in my limited computer resources, So I want to know the best score you get if you did not use the image features. my best score is 2220 about for nn and lgbm, and I cannot make them to get better score.",
      "votes": null
    },
    {
      "id": "344154",
      "postDate": "06/17/2018 05:08:57",
      "content": "<p>Image features have added ~0.002 for me, so I'd guess it should be possible to get to 0.218X or lower without image features.</p>",
      "rawMarkdown": "Image features have added ~0.002 for me, so I'd guess it should be possible to get to 0.218X or lower without image features.",
      "votes": null
    },
    {
      "id": "344156",
      "postDate": "06/17/2018 05:13:14",
      "content": "<p>thanks for your info , how much time did you take for processing image and download image dataset?</p>",
      "rawMarkdown": "thanks for your info , how much time did you take for processing image and download image dataset?",
      "votes": null
    },
    {
      "id": "344157",
      "postDate": "06/17/2018 05:17:42",
      "content": "<pre><code>so I'd guess it should be possible to get to 0.218X or lower without image features.\n</code></pre>\n\n<p>is there somthing wrong ? I guess you said is with image features could get 218X</p>",
      "rawMarkdown": "so I'd guess it should be possible to get to 0.218X or lower without image features.\n\nis there somthing wrong ? I guess you said is with image features could get 218X",
      "votes": null
    },
    {
      "id": "344161",
      "postDate": "06/17/2018 05:33:00",
      "content": "<blockquote>\n  <p>how much time did you take for processing image and download image dataset</p>\n</blockquote>\n\n<p>I got a bunch of features in about 12 hours, using parallel with 16 cores.</p>\n\n<p>One guy did image processing in Kaggle kernels... see <a href=\"https://www.kaggle.com/sukhyun9673/image-processing-600000-to-750000\">https://www.kaggle.com/sukhyun9673/image-processing-600000-to-750000</a> for one example. It's possible to do it without special hardware.</p>",
      "rawMarkdown": "&gt; how much time did you take for processing image and download image dataset\n\nI got a bunch of features in about 12 hours, using parallel with 16 cores.\n\nOne guy did image processing in Kaggle kernels... see https://www.kaggle.com/sukhyun9673/image-processing-600000-to-750000 for one example. It's possible to do it without special hardware.",
      "votes": null
    },
    {
      "id": "344162",
      "postDate": "06/17/2018 05:47:57",
      "content": "<p>thanks very much</p>",
      "rawMarkdown": "thanks very much",
      "votes": null
    },
    {
      "id": "344173",
      "postDate": "06/17/2018 06:22:51",
      "content": "<p>0.002 is amazing,how many from kernels?</p>",
      "rawMarkdown": "0.002 is amazing,how many from kernels?",
      "votes": null
    },
    {
      "id": "344177",
      "postDate": "06/17/2018 06:45:37",
      "content": "<p>Thanks! Our team has done everything the kernels have done and a lot more. :)</p>",
      "rawMarkdown": "Thanks! Our team has done everything the kernels have done and a lot more. :)",
      "votes": null
    },
    {
      "id": "344202",
      "postDate": "06/17/2018 07:47:17",
      "content": "<p>So without image features, I want to know what is your best score, because for now I dont know which direction to go , do more feature engineer or consider image features? thanks</p>",
      "rawMarkdown": "So without image features, I want to know what is your best score, because for now I dont know which direction to go , do more feature engineer or consider image features? thanks",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 344154,
      "author_name": "peterhurford",
      "author_url": "",
      "post_date": "06/17/2018 05:08:57",
      "content": "<p>Image features have added ~0.002 for me, so I'd guess it should be possible to get to 0.218X or lower without image features.</p>",
      "votes": null,
      "replies": [
        {
          "id": 344156,
          "author_name": "qfzgs1994",
          "author_url": "",
          "post_date": "06/17/2018 05:13:14",
          "content": "<p>thanks for your info , how much time did you take for processing image and download image dataset?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 344157,
          "author_name": "qfzgs1994",
          "author_url": "",
          "post_date": "06/17/2018 05:17:42",
          "content": "<pre><code>so I'd guess it should be possible to get to 0.218X or lower without image features.\n</code></pre>\n\n<p>is there somthing wrong ? I guess you said is with image features could get 218X</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 344161,
          "author_name": "peterhurford",
          "author_url": "",
          "post_date": "06/17/2018 05:33:00",
          "content": "<blockquote>\n  <p>how much time did you take for processing image and download image dataset</p>\n</blockquote>\n\n<p>I got a bunch of features in about 12 hours, using parallel with 16 cores.</p>\n\n<p>One guy did image processing in Kaggle kernels... see <a href=\"https://www.kaggle.com/sukhyun9673/image-processing-600000-to-750000\">https://www.kaggle.com/sukhyun9673/image-processing-600000-to-750000</a> for one example. It's possible to do it without special hardware.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 344162,
          "author_name": "qfzgs1994",
          "author_url": "",
          "post_date": "06/17/2018 05:47:57",
          "content": "<p>thanks very much</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 344173,
          "author_name": "senkin13",
          "author_url": "",
          "post_date": "06/17/2018 06:22:51",
          "content": "<p>0.002 is amazing,how many from kernels?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 344177,
          "author_name": "peterhurford",
          "author_url": "",
          "post_date": "06/17/2018 06:45:37",
          "content": "<p>Thanks! Our team has done everything the kernels have done and a lot more. :)</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 344202,
          "author_name": "qfzgs1994",
          "author_url": "",
          "post_date": "06/17/2018 07:47:17",
          "content": "<p>So without image features, I want to know what is your best score, because for now I dont know which direction to go , do more feature engineer or consider image features? thanks</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "344153": "Because the image dataset is so large that I cannot handle it in my limited computer resources, So I want to know the best score you get if you did not use the image features. my best score is 2220 about for nn and lgbm, and I cannot make them to get better score.",
    "344154": "Image features have added ~0.002 for me, so I'd guess it should be possible to get to 0.218X or lower without image features.",
    "344156": "thanks for your info , how much time did you take for processing image and download image dataset?",
    "344157": "so I'd guess it should be possible to get to 0.218X or lower without image features.\n\nis there somthing wrong ? I guess you said is with image features could get 218X",
    "344161": "&gt; how much time did you take for processing image and download image dataset\n\nI got a bunch of features in about 12 hours, using parallel with 16 cores.\n\nOne guy did image processing in Kaggle kernels... see https://www.kaggle.com/sukhyun9673/image-processing-600000-to-750000 for one example. It's possible to do it without special hardware.",
    "344162": "thanks very much",
    "344173": "0.002 is amazing,how many from kernels?",
    "344177": "Thanks! Our team has done everything the kernels have done and a lot more. :)",
    "344202": "So without image features, I want to know what is your best score, because for now I dont know which direction to go , do more feature engineer or consider image features? thanks"
  },
  "source": "meta"
}