{
  "id": 21920,
  "title": "How far you can go without images?",
  "url": "/competitions/avito-duplicate-ads-detection/discussion/21920",
  "author_name": "",
  "post_date": "2016-06-28T07:38:58.130Z",
  "votes": null,
  "comment_count": 6,
  "views": 818,
  "content": "<p>For me - played with data on this weekend, got ~ 0.85. And I think it is doable to receive ~0.9.</p>",
  "messages": [
    {
      "id": "125261",
      "postDate": "06/28/2016 07:38:58",
      "content": "<p>For me - played with data on this weekend, got ~ 0.85. And I think it is doable to receive ~0.9.</p>",
      "rawMarkdown": "For me - played with data on this weekend, got ~ 0.85. And I think it is doable to receive ~0.9.",
      "votes": null
    },
    {
      "id": "125273",
      "postDate": "06/28/2016 10:02:08",
      "content": "<p>Did you submit or is that local CV? There seems to be some differences between CV vs. LB.</p>\n\n<p>Image data is really powerful. Just adding hash (which I uploaded) to the public script raises LB score from 0.78 =&gt; 0.89. I'd imagine the JSON stuff to be quite additive too maybe in the order of 0.02-0.03. Will try them over the weekend once I sort out my locale issue.</p>",
      "rawMarkdown": "Did you submit or is that local CV? There seems to be some differences between CV vs. LB.\r\n\r\nImage data is really powerful. Just adding hash (which I uploaded) to the public script raises LB score from 0.78 => 0.89. I'd imagine the JSON stuff to be quite additive too maybe in the order of 0.02-0.03. Will try them over the weekend once I sort out my locale issue.",
      "votes": null
    },
    {
      "id": "125275",
      "postDate": "06/28/2016 10:46:55",
      "content": "<p>@yilisg, thanks for comments. Just submitted: 0.85 on local CV = 0.817 on public LB.</p>",
      "rawMarkdown": "yilisg, thanks for comments. Just submitted: 0.85 on local CV = 0.817 on public LB.",
      "votes": null
    },
    {
      "id": "125281",
      "postDate": "06/28/2016 11:56:57",
      "content": "<p>No problem. I am curious, for the 0.817 did you use the JSON attributes? Thanks.</p>",
      "rawMarkdown": "No problem. I am curious, for the 0.817 did you use the JSON attributes? Thanks.",
      "votes": null
    },
    {
      "id": "125282",
      "postDate": "06/28/2016 12:09:25",
      "content": "<p>No, I didn't use them. For textual features I used only description and title.</p>",
      "rawMarkdown": "No, I didn't use them. For textual features I used only description and title.",
      "votes": null
    },
    {
      "id": "125284",
      "postDate": "06/28/2016 12:28:59",
      "content": "<p>[quote=diaman;125261]</p>\n\n<p>For me - played with data on this weekend, got ~ 0.85. And I think it is doable to receive ~0.9.</p>\n\n<p>[/quote]</p>\n\n<p>I think one can reach close to 0.89 on LB without image features but need to do lot of FE. I would like to thank you also as I am using text2vec a lot :)</p>",
      "rawMarkdown": "[quote=diaman;125261]\r\n\r\nFor me - played with data on this weekend, got ~ 0.85. And I think it is doable to receive ~0.9.\r\n\r\n[/quote]\r\n\r\nI think one can reach close to 0.89 on LB without image features but need to do lot of FE. I would like to thank you also as I am using text2vec a lot :)",
      "votes": null
    },
    {
      "id": "125417",
      "postDate": "06/29/2016 09:40:07",
      "content": "<p>Our result without images, json and gbc parameters tuning was ~0.84 on Public LB.</p>",
      "rawMarkdown": "Our result without images, json and gbc parameters tuning was ~0.84 on Public LB.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 125273,
      "author_name": "yilihome",
      "author_url": "",
      "post_date": "06/28/2016 10:02:08",
      "content": "<p>Did you submit or is that local CV? There seems to be some differences between CV vs. LB.</p>\n\n<p>Image data is really powerful. Just adding hash (which I uploaded) to the public script raises LB score from 0.78 =&gt; 0.89. I'd imagine the JSON stuff to be quite additive too maybe in the order of 0.02-0.03. Will try them over the weekend once I sort out my locale issue.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 125275,
      "author_name": "dselivanov",
      "author_url": "",
      "post_date": "06/28/2016 10:46:55",
      "content": "<p>@yilisg, thanks for comments. Just submitted: 0.85 on local CV = 0.817 on public LB.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 125281,
      "author_name": "yilihome",
      "author_url": "",
      "post_date": "06/28/2016 11:56:57",
      "content": "<p>No problem. I am curious, for the 0.817 did you use the JSON attributes? Thanks.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 125282,
      "author_name": "dselivanov",
      "author_url": "",
      "post_date": "06/28/2016 12:09:25",
      "content": "<p>No, I didn't use them. For textual features I used only description and title.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 125284,
      "author_name": "thakurrajanand",
      "author_url": "",
      "post_date": "06/28/2016 12:28:59",
      "content": "<p>[quote=diaman;125261]</p>\n\n<p>For me - played with data on this weekend, got ~ 0.85. And I think it is doable to receive ~0.9.</p>\n\n<p>[/quote]</p>\n\n<p>I think one can reach close to 0.89 on LB without image features but need to do lot of FE. I would like to thank you also as I am using text2vec a lot :)</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 125417,
      "author_name": "dansheb",
      "author_url": "",
      "post_date": "06/29/2016 09:40:07",
      "content": "<p>Our result without images, json and gbc parameters tuning was ~0.84 on Public LB.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "125261": "For me - played with data on this weekend, got ~ 0.85. And I think it is doable to receive ~0.9.",
    "125273": "Did you submit or is that local CV? There seems to be some differences between CV vs. LB.\r\n\r\nImage data is really powerful. Just adding hash (which I uploaded) to the public script raises LB score from 0.78 => 0.89. I'd imagine the JSON stuff to be quite additive too maybe in the order of 0.02-0.03. Will try them over the weekend once I sort out my locale issue.",
    "125275": "yilisg, thanks for comments. Just submitted: 0.85 on local CV = 0.817 on public LB.",
    "125281": "No problem. I am curious, for the 0.817 did you use the JSON attributes? Thanks.",
    "125282": "No, I didn't use them. For textual features I used only description and title.",
    "125284": "[quote=diaman;125261]\r\n\r\nFor me - played with data on this weekend, got ~ 0.85. And I think it is doable to receive ~0.9.\r\n\r\n[/quote]\r\n\r\nI think one can reach close to 0.89 on LB without image features but need to do lot of FE. I would like to thank you also as I am using text2vec a lot :)",
    "125417": "Our result without images, json and gbc parameters tuning was ~0.84 on Public LB."
  },
  "source": "meta"
}