{
  "id": 104978,
  "title": "There is more than one mechanism",
  "url": "/competitions/champs-scalar-coupling/discussion/104978",
  "author_name": "",
  "post_date": "2019-08-20T11:48:58.142670300Z",
  "votes": 6,
  "comment_count": 4,
  "views": 0,
  "content": "<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F241438%2F84a0c2f04e1d7d94f95b9cc8fc87cda4%2Findex.png?generation=1566301671700423&amp;alt=media\" alt=\"\">\nClustering by type and distance (plus distance&lt;1.5) seem interesting as there are different mechanisms at play.  Distance is the x axis, scalar coupling is the y axis and the color is the type (1JHC, 2JHC etc)\nIf you are using some sort of target encoding (mean,min,max,std) on just type you are going to miss the mark! Just saying...</p>",
  "messages": [
    {
      "id": "603516",
      "postDate": "08/20/2019 11:48:58",
      "content": "<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F241438%2F84a0c2f04e1d7d94f95b9cc8fc87cda4%2Findex.png?generation=1566301671700423&amp;alt=media\" alt=\"\">\nClustering by type and distance (plus distance&lt;1.5) seem interesting as there are different mechanisms at play.  Distance is the x axis, scalar coupling is the y axis and the color is the type (1JHC, 2JHC etc)\nIf you are using some sort of target encoding (mean,min,max,std) on just type you are going to miss the mark! Just saying...</p>",
      "rawMarkdown": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F241438%2F84a0c2f04e1d7d94f95b9cc8fc87cda4%2Findex.png?generation=1566301671700423&amp;alt=media)\nClustering by type and distance (plus distance&lt;1.5) seem interesting as there are different mechanisms at play.  Distance is the x axis, scalar coupling is the y axis and the color is the type (1JHC, 2JHC etc)\nIf you are using some sort of target encoding (mean,min,max,std) on just type you are going to miss the mark! Just saying...",
      "votes": null
    },
    {
      "id": "603669",
      "postDate": "08/20/2019 14:37:08",
      "content": "<p>I guess this is the whole point of this. I would not expect that modeling quantum mechanics would fit nicely in a chart. </p>",
      "rawMarkdown": "I guess this is the whole point of this. I would not expect that modeling quantum mechanics would fit nicely in a chart.",
      "votes": null
    },
    {
      "id": "603690",
      "postDate": "08/20/2019 14:58:43",
      "content": "<p>Your not wrong!!  Congrats to be #1 at the moment - I would have liked to predict experimental values rather than model predictions but hey ho.</p>",
      "rawMarkdown": "Your not wrong!!  Congrats to be #1 at the moment - I would have liked to predict experimental values rather than model predictions but hey ho.",
      "votes": null
    },
    {
      "id": "603710",
      "postDate": "08/20/2019 15:26:14",
      "content": "<p>I would find that 1JHC have strong attractive energy on XX, YY , ZZ.   Is it one of the possible reason?   </p>",
      "rawMarkdown": "I would find that 1JHC have strong attractive energy on XX, YY , ZZ.   Is it one of the possible reason?",
      "votes": null
    },
    {
      "id": "603831",
      "postDate": "08/20/2019 17:45:22",
      "content": "<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F241438%2Feea11acdf6953a37c6ebbee497c024a1%2Findex2.png?generation=1566323072552486&amp;alt=media\" alt=\"\">\n<a href=\"/schwinger\">@schwinger</a> A kNN just based on distance between edge atoms and their respective charges to predict fermi contact.\nGets GroupLog MAE of .4 compared to real fc Group Log MAE of -1.2 but I used nearest neighbor of 100 which is probably overkill. </p>",
      "rawMarkdown": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F241438%2Feea11acdf6953a37c6ebbee497c024a1%2Findex2.png?generation=1566323072552486&amp;alt=media)\n@schwinger A kNN just based on distance between edge atoms and their respective charges to predict fermi contact.\nGets GroupLog MAE of .4 compared to real fc Group Log MAE of -1.2 but I used nearest neighbor of 100 which is probably overkill.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 603669,
      "author_name": "antorsae",
      "author_url": "",
      "post_date": "08/20/2019 14:37:08",
      "content": "<p>I guess this is the whole point of this. I would not expect that modeling quantum mechanics would fit nicely in a chart. </p>",
      "votes": null,
      "replies": [
        {
          "id": 603690,
          "author_name": "scirpus",
          "author_url": "",
          "post_date": "08/20/2019 14:58:43",
          "content": "<p>Your not wrong!!  Congrats to be #1 at the moment - I would have liked to predict experimental values rather than model predictions but hey ho.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 603710,
      "author_name": "schwinger",
      "author_url": "",
      "post_date": "08/20/2019 15:26:14",
      "content": "<p>I would find that 1JHC have strong attractive energy on XX, YY , ZZ.   Is it one of the possible reason?   </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 603831,
      "author_name": "scirpus",
      "author_url": "",
      "post_date": "08/20/2019 17:45:22",
      "content": "<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F241438%2Feea11acdf6953a37c6ebbee497c024a1%2Findex2.png?generation=1566323072552486&amp;alt=media\" alt=\"\">\n<a href=\"/schwinger\">@schwinger</a> A kNN just based on distance between edge atoms and their respective charges to predict fermi contact.\nGets GroupLog MAE of .4 compared to real fc Group Log MAE of -1.2 but I used nearest neighbor of 100 which is probably overkill. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "603516": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F241438%2F84a0c2f04e1d7d94f95b9cc8fc87cda4%2Findex.png?generation=1566301671700423&amp;alt=media)\nClustering by type and distance (plus distance&lt;1.5) seem interesting as there are different mechanisms at play.  Distance is the x axis, scalar coupling is the y axis and the color is the type (1JHC, 2JHC etc)\nIf you are using some sort of target encoding (mean,min,max,std) on just type you are going to miss the mark! Just saying...",
    "603669": "I guess this is the whole point of this. I would not expect that modeling quantum mechanics would fit nicely in a chart.",
    "603690": "Your not wrong!!  Congrats to be #1 at the moment - I would have liked to predict experimental values rather than model predictions but hey ho.",
    "603710": "I would find that 1JHC have strong attractive energy on XX, YY , ZZ.   Is it one of the possible reason?",
    "603831": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F241438%2Feea11acdf6953a37c6ebbee497c024a1%2Findex2.png?generation=1566323072552486&amp;alt=media)\n@schwinger A kNN just based on distance between edge atoms and their respective charges to predict fermi contact.\nGets GroupLog MAE of .4 compared to real fc Group Log MAE of -1.2 but I used nearest neighbor of 100 which is probably overkill."
  },
  "source": "meta"
}