{
  "id": 22039,
  "title": "Baseline of single model",
  "url": "/competitions/avito-duplicate-ads-detection/discussion/22039",
  "author_name": "",
  "post_date": "2016-07-04T16:45:40.977Z",
  "votes": null,
  "comment_count": 4,
  "views": 764,
  "content": "<p>Hi\nJust to get an idea of the potential improvment, how High is it possible to go without stacking/ensembling,  using external data and speaking russian which is my crurent situation. </p>\n\n<p>Thxs</p>",
  "messages": [
    {
      "id": "125934",
      "postDate": "07/04/2016 16:45:40",
      "content": "<p>Hi\nJust to get an idea of the potential improvment, how High is it possible to go without stacking/ensembling,  using external data and speaking russian which is my crurent situation. </p>\n\n<p>Thxs</p>",
      "rawMarkdown": "Hi\r\nJust to get an idea of the potential improvment, how High is it possible to go without stacking/ensembling,  using external data and speaking russian which is my crurent situation. \r\n\r\nThxs",
      "votes": null
    },
    {
      "id": "125937",
      "postDate": "07/04/2016 17:21:26",
      "content": "<p>[quote=jayjay;125934]</p>\n\n<p>Hi\nJust to get an idea of the potential improvment, how High is it possible to go without stacking/ensembling,  using external data and speaking russian which is my crurent situation. </p>\n\n<p>Thxs</p>\n\n<p>[/quote]</p>\n\n<p>Single XGBOOST = 0.94 LB. We are not using any external data but we are using libraries to generate features from images and the rest of our features are using standard R and Python libraries. I think you really don't need to know Russian. I remember the team that won previous Avito competition didn't have any team member who knew Russian.</p>",
      "rawMarkdown": "[quote=jayjay;125934]\r\n\r\nHi\r\nJust to get an idea of the potential improvment, how High is it possible to go without stacking/ensembling,  using external data and speaking russian which is my crurent situation. \r\n\r\nThxs\r\n\r\n[/quote]\r\n\r\nSingle XGBOOST = 0.94 LB. We are not using any external data but we are using libraries to generate features from images and the rest of our features are using standard R and Python libraries. I think you really don't need to know Russian. I remember the team that won previous Avito competition didn't have any team member who knew Russian.",
      "votes": null
    },
    {
      "id": "125940",
      "postDate": "07/04/2016 19:05:07",
      "content": "<p>[quote=DataGeek;125937]</p>\n\n<p>[quote=jayjay;125934]</p>\n\n<p>Hi\nJust to get an idea of the potential improvment, how High is it possible to go without stacking/ensembling,  using external data and speaking russian which is my crurent situation. </p>\n\n<p>Thxs</p>\n\n<p>[/quote]</p>\n\n<p>Single XGBOOST = 0.94 LB. We are not using any external data but we are using libraries to generate features from images and the rest of our features are using standard R and Python libraries. I think you really don't need to know Russian. I remember the team that won previous Avito competition didn't have any team member who knew Russian.</p>\n\n<p>[/quote]</p>\n\n<p>What's your baseline score without the more advanced image features (using only the hashes similarity)? Can you share it? Thanks!</p>",
      "rawMarkdown": "[quote=DataGeek;125937]\r\n\r\n[quote=jayjay;125934]\r\n\r\nHi\r\nJust to get an idea of the potential improvment, how High is it possible to go without stacking/ensembling,  using external data and speaking russian which is my crurent situation. \r\n\r\nThxs\r\n\r\n[/quote]\r\n\r\nSingle XGBOOST = 0.94 LB. We are not using any external data but we are using libraries to generate features from images and the rest of our features are using standard R and Python libraries. I think you really don't need to know Russian. I remember the team that won previous Avito competition didn't have any team member who knew Russian.\r\n\r\n[/quote]\r\n\r\nWhat's your baseline score without the more advanced image features (using only the hashes similarity)? Can you share it? Thanks!",
      "votes": null
    },
    {
      "id": "125949",
      "postDate": "07/04/2016 22:23:11",
      "content": "<p>[quote=Snow Dog;125940]</p>\n\n<p>[quote=DataGeek;125937]</p>\n\n<p>[quote=jayjay;125934]</p>\n\n<p>Hi\nJust to get an idea of the potential improvment, how High is it possible to go without stacking/ensembling,  using external data and speaking russian which is my crurent situation. </p>\n\n<p>Thxs</p>\n\n<p>[/quote]</p>\n\n<p>Single XGBOOST = 0.94 LB. We are not using any external data but we are using libraries to generate features from images and the rest of our features are using standard R and Python libraries. I think you really don't need to know Russian. I remember the team that won previous Avito competition didn't have any team member who knew Russian.</p>\n\n<p>[/quote]</p>\n\n<p>What's your baseline score without the more advanced image features (using only the hashes similarity)? Can you share it? Thanks!</p>\n\n<p>[/quote]</p>\n\n<p>Never kept track and it is bit hard to tell exact score as we added a lot of other features after advanced image features also. My guess would be  0.92+ with hashes only.</p>",
      "rawMarkdown": "[quote=Snow Dog;125940]\r\n\r\n[quote=DataGeek;125937]\r\n\r\n[quote=jayjay;125934]\r\n\r\nHi\r\nJust to get an idea of the potential improvment, how High is it possible to go without stacking/ensembling,  using external data and speaking russian which is my crurent situation. \r\n\r\nThxs\r\n\r\n[/quote]\r\n\r\nSingle XGBOOST = 0.94 LB. We are not using any external data but we are using libraries to generate features from images and the rest of our features are using standard R and Python libraries. I think you really don't need to know Russian. I remember the team that won previous Avito competition didn't have any team member who knew Russian.\r\n\r\n[/quote]\r\n\r\nWhat's your baseline score without the more advanced image features (using only the hashes similarity)? Can you share it? Thanks!\r\n\r\n\r\n\r\n[/quote]\r\n\r\nNever kept track and it is bit hard to tell exact score as we added a lot of other features after advanced image features also. My guess would be  0.92+ with hashes only.",
      "votes": null
    },
    {
      "id": "125972",
      "postDate": "07/05/2016 05:57:31",
      "content": "<p>[quote=DataGeek;125949]\n My guess would be  0.92+ with hashes only.</p>\n\n<p>[/quote]\nIt's possible to get more than 0.93 with hashes only </p>",
      "rawMarkdown": "[quote=DataGeek;125949]\r\n My guess would be  0.92+ with hashes only.\r\n\r\n[/quote]\r\nIt's possible to get more than 0.93 with hashes only",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 125937,
      "author_name": "thakurrajanand",
      "author_url": "",
      "post_date": "07/04/2016 17:21:26",
      "content": "<p>[quote=jayjay;125934]</p>\n\n<p>Hi\nJust to get an idea of the potential improvment, how High is it possible to go without stacking/ensembling,  using external data and speaking russian which is my crurent situation. </p>\n\n<p>Thxs</p>\n\n<p>[/quote]</p>\n\n<p>Single XGBOOST = 0.94 LB. We are not using any external data but we are using libraries to generate features from images and the rest of our features are using standard R and Python libraries. I think you really don't need to know Russian. I remember the team that won previous Avito competition didn't have any team member who knew Russian.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 125940,
      "author_name": "snowdog",
      "author_url": "",
      "post_date": "07/04/2016 19:05:07",
      "content": "<p>[quote=DataGeek;125937]</p>\n\n<p>[quote=jayjay;125934]</p>\n\n<p>Hi\nJust to get an idea of the potential improvment, how High is it possible to go without stacking/ensembling,  using external data and speaking russian which is my crurent situation. </p>\n\n<p>Thxs</p>\n\n<p>[/quote]</p>\n\n<p>Single XGBOOST = 0.94 LB. We are not using any external data but we are using libraries to generate features from images and the rest of our features are using standard R and Python libraries. I think you really don't need to know Russian. I remember the team that won previous Avito competition didn't have any team member who knew Russian.</p>\n\n<p>[/quote]</p>\n\n<p>What's your baseline score without the more advanced image features (using only the hashes similarity)? Can you share it? Thanks!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 125949,
      "author_name": "thakurrajanand",
      "author_url": "",
      "post_date": "07/04/2016 22:23:11",
      "content": "<p>[quote=Snow Dog;125940]</p>\n\n<p>[quote=DataGeek;125937]</p>\n\n<p>[quote=jayjay;125934]</p>\n\n<p>Hi\nJust to get an idea of the potential improvment, how High is it possible to go without stacking/ensembling,  using external data and speaking russian which is my crurent situation. </p>\n\n<p>Thxs</p>\n\n<p>[/quote]</p>\n\n<p>Single XGBOOST = 0.94 LB. We are not using any external data but we are using libraries to generate features from images and the rest of our features are using standard R and Python libraries. I think you really don't need to know Russian. I remember the team that won previous Avito competition didn't have any team member who knew Russian.</p>\n\n<p>[/quote]</p>\n\n<p>What's your baseline score without the more advanced image features (using only the hashes similarity)? Can you share it? Thanks!</p>\n\n<p>[/quote]</p>\n\n<p>Never kept track and it is bit hard to tell exact score as we added a lot of other features after advanced image features also. My guess would be  0.92+ with hashes only.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 125972,
      "author_name": "agrigorev",
      "author_url": "",
      "post_date": "07/05/2016 05:57:31",
      "content": "<p>[quote=DataGeek;125949]\n My guess would be  0.92+ with hashes only.</p>\n\n<p>[/quote]\nIt's possible to get more than 0.93 with hashes only </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "125934": "Hi\r\nJust to get an idea of the potential improvment, how High is it possible to go without stacking/ensembling,  using external data and speaking russian which is my crurent situation. \r\n\r\nThxs",
    "125937": "[quote=jayjay;125934]\r\n\r\nHi\r\nJust to get an idea of the potential improvment, how High is it possible to go without stacking/ensembling,  using external data and speaking russian which is my crurent situation. \r\n\r\nThxs\r\n\r\n[/quote]\r\n\r\nSingle XGBOOST = 0.94 LB. We are not using any external data but we are using libraries to generate features from images and the rest of our features are using standard R and Python libraries. I think you really don't need to know Russian. I remember the team that won previous Avito competition didn't have any team member who knew Russian.",
    "125940": "[quote=DataGeek;125937]\r\n\r\n[quote=jayjay;125934]\r\n\r\nHi\r\nJust to get an idea of the potential improvment, how High is it possible to go without stacking/ensembling,  using external data and speaking russian which is my crurent situation. \r\n\r\nThxs\r\n\r\n[/quote]\r\n\r\nSingle XGBOOST = 0.94 LB. We are not using any external data but we are using libraries to generate features from images and the rest of our features are using standard R and Python libraries. I think you really don't need to know Russian. I remember the team that won previous Avito competition didn't have any team member who knew Russian.\r\n\r\n[/quote]\r\n\r\nWhat's your baseline score without the more advanced image features (using only the hashes similarity)? Can you share it? Thanks!",
    "125949": "[quote=Snow Dog;125940]\r\n\r\n[quote=DataGeek;125937]\r\n\r\n[quote=jayjay;125934]\r\n\r\nHi\r\nJust to get an idea of the potential improvment, how High is it possible to go without stacking/ensembling,  using external data and speaking russian which is my crurent situation. \r\n\r\nThxs\r\n\r\n[/quote]\r\n\r\nSingle XGBOOST = 0.94 LB. We are not using any external data but we are using libraries to generate features from images and the rest of our features are using standard R and Python libraries. I think you really don't need to know Russian. I remember the team that won previous Avito competition didn't have any team member who knew Russian.\r\n\r\n[/quote]\r\n\r\nWhat's your baseline score without the more advanced image features (using only the hashes similarity)? Can you share it? Thanks!\r\n\r\n\r\n\r\n[/quote]\r\n\r\nNever kept track and it is bit hard to tell exact score as we added a lot of other features after advanced image features also. My guess would be  0.92+ with hashes only.",
    "125972": "[quote=DataGeek;125949]\r\n My guess would be  0.92+ with hashes only.\r\n\r\n[/quote]\r\nIt's possible to get more than 0.93 with hashes only"
  },
  "source": "meta"
}