{
  "id": 98717,
  "title": "New Simple but Powerful Feature: LMBTR",
  "url": "/competitions/champs-scalar-coupling/discussion/98717",
  "author_name": "",
  "post_date": "2019-07-05T19:13:57.989125100Z",
  "votes": 4,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Hello,</p>\n\n<p>Here is an <a href=\"https://www.kaggle.com/yi233333/feature-lmbtr\">example</a> to extract \"Local Many-body Tensor Representation\" for local environments. It's quite simple but very useful. You can read more info about it <a href=\"https://singroup.github.io/dscribe/tutorials/lmbtr.html\">here</a>.</p>\n\n<p>Enjoy!</p>",
  "messages": [
    {
      "id": "568966",
      "postDate": "07/05/2019 19:13:57",
      "content": "<p>Hello,</p>\n\n<p>Here is an <a href=\"https://www.kaggle.com/yi233333/feature-lmbtr\">example</a> to extract \"Local Many-body Tensor Representation\" for local environments. It's quite simple but very useful. You can read more info about it <a href=\"https://singroup.github.io/dscribe/tutorials/lmbtr.html\">here</a>.</p>\n\n<p>Enjoy!</p>",
      "rawMarkdown": "Hello,\n\nHere is an [example](https://www.kaggle.com/yi233333/feature-lmbtr) to extract \"Local Many-body Tensor Representation\" for local environments. It's quite simple but very useful. You can read more info about it [here](https://singroup.github.io/dscribe/tutorials/lmbtr.html).\n\nEnjoy!",
      "votes": null
    },
    {
      "id": "568972",
      "postDate": "07/05/2019 19:27:09",
      "content": "<p>So, do you train a convolutional neural network over the array? Or just a fully connected network? Or are you feeding this representation into gradient boosted methods? Thank you</p>",
      "rawMarkdown": "So, do you train a convolutional neural network over the array? Or just a fully connected network? Or are you feeding this representation into gradient boosted methods? Thank you",
      "votes": null
    },
    {
      "id": "568976",
      "postDate": "07/05/2019 19:36:17",
      "content": "<p>And, as an add on to Corey's question, how much of a boost did this give your submission?</p>",
      "rawMarkdown": "And, as an add on to Corey's question, how much of a boost did this give your submission?",
      "votes": null
    },
    {
      "id": "568991",
      "postDate": "07/05/2019 20:09:20",
      "content": "<p>It will take an eternity to calculate LMBTRs for all data... Tryed it a couple days ago for the smallest type. For my models this features didn't provide any cv/lb boost</p>",
      "rawMarkdown": "It will take an eternity to calculate LMBTRs for all data... Tryed it a couple days ago for the smallest type. For my models this features didn't provide any cv/lb boost",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 568972,
      "author_name": "returnofsputnik",
      "author_url": "",
      "post_date": "07/05/2019 19:27:09",
      "content": "<p>So, do you train a convolutional neural network over the array? Or just a fully connected network? Or are you feeding this representation into gradient boosted methods? Thank you</p>",
      "votes": null,
      "replies": [
        {
          "id": 568976,
          "author_name": "petersk20",
          "author_url": "",
          "post_date": "07/05/2019 19:36:17",
          "content": "<p>And, as an add on to Corey's question, how much of a boost did this give your submission?</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 568991,
      "author_name": "stanislavblinov",
      "author_url": "",
      "post_date": "07/05/2019 20:09:20",
      "content": "<p>It will take an eternity to calculate LMBTRs for all data... Tryed it a couple days ago for the smallest type. For my models this features didn't provide any cv/lb boost</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "568966": "Hello,\n\nHere is an [example](https://www.kaggle.com/yi233333/feature-lmbtr) to extract \"Local Many-body Tensor Representation\" for local environments. It's quite simple but very useful. You can read more info about it [here](https://singroup.github.io/dscribe/tutorials/lmbtr.html).\n\nEnjoy!",
    "568972": "So, do you train a convolutional neural network over the array? Or just a fully connected network? Or are you feeding this representation into gradient boosted methods? Thank you",
    "568976": "And, as an add on to Corey's question, how much of a boost did this give your submission?",
    "568991": "It will take an eternity to calculate LMBTRs for all data... Tryed it a couple days ago for the smallest type. For my models this features didn't provide any cv/lb boost"
  },
  "source": "meta"
}