{
  "id": 104884,
  "title": "Best single model",
  "url": "/competitions/champs-scalar-coupling/discussion/104884",
  "author_name": "Jiwei Liu",
  "post_date": "2019-08-19T20:12:34.992000",
  "votes": 17,
  "comment_count": 22,
  "views": 0,
  "content": "<p>Our best xgb is about -1.87 and our best GNN is -1.92. How about yours?</p>",
  "messages": [
    {
      "id": 603060,
      "postDate": "2019-08-19T20:12:34.993Z",
      "content": "<p>Our best xgb is about -1.87 and our best GNN is -1.92. How about yours?</p>",
      "rawMarkdown": "Our best xgb is about -1.87 and our best GNN is -1.92. How about yours?",
      "votes": 17
    },
    {
      "id": 603233,
      "postDate": "2019-08-20T03:20:22.860Z",
      "content": "<p>Conventional NN: -2.047</p>",
      "rawMarkdown": "Conventional NN: -2.047",
      "votes": 3,
      "replies": [
        {
          "id": 603748,
          "postDate": "2019-08-20T16:07:28.557Z",
          "content": "<p>Hi <a href=\"/anjum48\">@anjum48</a> , Could you comment on your features and feature selection?</p>",
          "rawMarkdown": "Hi @anjum48 , Could you comment on your features and feature selection?",
          "votes": 2
        }
      ]
    },
    {
      "id": 603080,
      "postDate": "2019-08-19T20:26:05.943Z",
      "content": "<p>Not fully trained LGBM: - 1.946</p>",
      "rawMarkdown": "Not fully trained LGBM: - 1.946",
      "votes": 3,
      "replies": [
        {
          "id": 603106,
          "postDate": "2019-08-19T21:09:48.777Z",
          "content": "<p>Thank you. It seems that both lgb and gnn have a lot of potentials. 😀 </p>",
          "rawMarkdown": "Thank you. It seems that both lgb and gnn have a lot of potentials. 😀 ",
          "votes": 1
        }
      ]
    },
    {
      "id": 603314,
      "postDate": "2019-08-20T06:17:55.017Z",
      "content": "<p>You have some best models reports in this earlier thread: <a href=\"https://www.kaggle.com/c/champs-scalar-coupling/discussion/98444#latest-602104\">https://www.kaggle.com/c/champs-scalar-coupling/discussion/98444#latest-602104</a></p>",
      "rawMarkdown": "You have some best models reports in this earlier thread: https://www.kaggle.com/c/champs-scalar-coupling/discussion/98444#latest-602104\n",
      "votes": 2,
      "replies": [
        {
          "id": 603585,
          "postDate": "2019-08-20T13:18:55.953Z",
          "content": "<p>Thank you so much. I didn't expect so much info in that thread. 😄 </p>",
          "rawMarkdown": "Thank you so much. I didn't expect so much info in that thread. 😄 ",
          "votes": 1
        }
      ]
    },
    {
      "id": 603205,
      "postDate": "2019-08-20T02:38:05.990Z",
      "content": "<p>GNN: -1.981 (1fold of 3fold CV)\nI am interested in single model score of gold medal area players  : )</p>",
      "rawMarkdown": "GNN: -1.981 (1fold of 3fold CV)\nI am interested in single model score of gold medal area players  : )",
      "votes": 2
    },
    {
      "id": 603120,
      "postDate": "2019-08-19T21:49:48.077Z",
      "content": "<p>LGB: -1.97, GNN: -1.44.</p>",
      "rawMarkdown": "LGB: -1.97, GNN: -1.44.",
      "votes": 2,
      "replies": [
        {
          "id": 603136,
          "postDate": "2019-08-19T22:49:59.310Z",
          "content": "<p>What is your keras NN cv Shuhao? :-)</p>",
          "rawMarkdown": "What is your keras NN cv Shuhao? :-)"
        },
        {
          "id": 603140,
          "postDate": "2019-08-19T23:09:51.783Z",
          "content": "<p>Only -0.97 but I mainly use it to extract features.</p>",
          "rawMarkdown": "Only -0.97 but I mainly use it to extract features."
        }
      ]
    },
    {
      "id": 604738,
      "postDate": "2019-08-21T18:08:23.853Z",
      "content": "<p>Our team best single Lightgbm = -1.745 and Gnn = -1.621. </p>",
      "rawMarkdown": "Our team best single Lightgbm = -1.745 and Gnn = -1.621. "
    },
    {
      "id": 604345,
      "postDate": "2019-08-21T09:34:43.940Z",
      "content": "<p>GNN -1.9, hardly even tuned yet, reckon it would top out -2 to -2.5.</p>",
      "rawMarkdown": "GNN -1.9, hardly even tuned yet, reckon it would top out -2 to -2.5."
    },
    {
      "id": 603736,
      "postDate": "2019-08-20T15:52:28.017Z",
      "content": "<p><a href=\"/jiweiliu\">@jiweiliu</a> : do you have a secret sauce to jump from -1.87 and -1.92 to -2.3 ?\nmy lgbm and gnn are in the same -1.8+ area but I can't figure out how to jump so far.</p>",
      "rawMarkdown": "@jiweiliu : do you have a secret sauce to jump from -1.87 and -1.92 to -2.3 ?\nmy lgbm and gnn are in the same -1.8+ area but I can't figure out how to jump so far."
    },
    {
      "id": 603322,
      "postDate": "2019-08-20T06:30:26.710Z",
      "content": "<p>Let me guess：\n-2.0 and below : tree boost\n-2.5 to -2.0 : GNN\n-2.5 and above : GNN and chemistry domain knowledge</p>",
      "rawMarkdown": "Let me guess：\n-2.0 and below : tree boost\n-2.5 to -2.0 : GNN\n-2.5 and above : GNN and chemistry domain knowledge",
      "replies": [
        {
          "id": 603356,
          "postDate": "2019-08-20T07:16:06.323Z",
          "content": "<p>I don't think Domain Knowledge is needed in this competition if you mean chemistry. It seems using deep learning for quantum chemistry is very common, the reason why domain experts are high on the leaderboard is probably because they are also good at deep learning on unstructred data.</p>",
          "rawMarkdown": "I don't think Domain Knowledge is needed in this competition if you mean chemistry. It seems using deep learning for quantum chemistry is very common, the reason why domain experts are high on the leaderboard is probably because they are also good at deep learning on unstructred data.",
          "votes": 8
        },
        {
          "id": 603357,
          "postDate": "2019-08-20T07:18:15.863Z",
          "content": "<blockquote>\n  <p>the reason why domain experts are high on the leaderboard is probably because they are also good at deep learning on unstructred data.</p>\n</blockquote>\n\n<p>And also because they have a warm start with ready to use NNs.  </p>\n\n<p>The domain knowledge that seems relevant here is not chemistry, it is graph NNs.</p>",
          "rawMarkdown": "&gt; the reason why domain experts are high on the leaderboard is probably because they are also good at deep learning on unstructred data.\n\nAnd also because they have a warm start with ready to use NNs.  \n\nThe domain knowledge that seems relevant here is not chemistry, it is graph NNs.",
          "votes": 7
        },
        {
          "id": 603364,
          "postDate": "2019-08-20T07:23:06.810Z",
          "content": "<blockquote>\n  <p>2.0 and below : tree boost\n  -2.5 to -2.0 : GNN</p>\n</blockquote>\n\n<p>It is possible to get way better than -2 with tree boosting.</p>",
          "rawMarkdown": "&gt;2.0 and below : tree boost\n&gt; -2.5 to -2.0 : GNN\n\nIt is possible to get way better than -2 with tree boosting.\n",
          "votes": 4
        },
        {
          "id": 603593,
          "postDate": "2019-08-20T13:27:59.163Z",
          "content": "<blockquote>\n  <p>way better than -2 </p>\n</blockquote>\n\n<p>I could use more decimal digits 😊 </p>",
          "rawMarkdown": "&gt;way better than -2 \n\nI could use more decimal digits 😊 ",
          "votes": 2
        },
        {
          "id": 604088,
          "postDate": "2019-08-21T02:43:07.950Z",
          "content": "<p>May I ask what's your XGB LB score if your single XGB CV is -1.87? Thanks!</p>",
          "rawMarkdown": "May I ask what's your XGB LB score if your single XGB CV is -1.87? Thanks!"
        },
        {
          "id": 609175,
          "postDate": "2019-08-27T13:15:54.843Z",
          "content": "<p><a href=\"/cpmpml\">@cpmpml</a> Would you share this approach after competition?</p>",
          "rawMarkdown": "@cpmpml Would you share this approach after competition?"
        }
      ]
    },
    {
      "id": 605049,
      "postDate": "2019-08-22T02:31:10.817Z",
      "rawMarkdown": "",
      "isDeleted": true,
      "replies": [
        {
          "id": 609177,
          "postDate": "2019-08-27T13:17:02.830Z",
          "content": "<p>This forum ;)</p>",
          "rawMarkdown": "This forum ;)"
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 603233,
      "author_name": "datasaurus",
      "author_url": "",
      "post_date": "2019-08-20T03:20:22.860000",
      "content": "<p>Conventional NN: -2.047</p>",
      "votes": 3,
      "replies": [
        {
          "id": 603748,
          "author_name": "Kranthi Kumar",
          "author_url": "",
          "post_date": "2019-08-20T16:07:28.557000",
          "content": "<p>Hi <a href=\"/anjum48\">@anjum48</a> , Could you comment on your features and feature selection?</p>",
          "votes": 2,
          "replies": []
        }
      ]
    },
    {
      "id": 603080,
      "author_name": "somnambWl",
      "author_url": "",
      "post_date": "2019-08-19T20:26:05.943000",
      "content": "<p>Not fully trained LGBM: - 1.946</p>",
      "votes": 3,
      "replies": [
        {
          "id": 603106,
          "author_name": "Jiwei Liu",
          "author_url": "",
          "post_date": "2019-08-19T21:09:48.777000",
          "content": "<p>Thank you. It seems that both lgb and gnn have a lot of potentials. 😀 </p>",
          "votes": 1,
          "replies": []
        }
      ]
    },
    {
      "id": 603314,
      "author_name": "CPMP",
      "author_url": "",
      "post_date": "2019-08-20T06:17:55.017000",
      "content": "<p>You have some best models reports in this earlier thread: <a href=\"https://www.kaggle.com/c/champs-scalar-coupling/discussion/98444#latest-602104\">https://www.kaggle.com/c/champs-scalar-coupling/discussion/98444#latest-602104</a></p>",
      "votes": 2,
      "replies": [
        {
          "id": 603585,
          "author_name": "Jiwei Liu",
          "author_url": "",
          "post_date": "2019-08-20T13:18:55.953000",
          "content": "<p>Thank you so much. I didn't expect so much info in that thread. 😄 </p>",
          "votes": 1,
          "replies": []
        }
      ]
    },
    {
      "id": 603205,
      "author_name": "pao",
      "author_url": "",
      "post_date": "2019-08-20T02:38:05.990000",
      "content": "<p>GNN: -1.981 (1fold of 3fold CV)\nI am interested in single model score of gold medal area players  : )</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 603120,
      "author_name": "Shuhao Cao",
      "author_url": "",
      "post_date": "2019-08-19T21:49:48.077000",
      "content": "<p>LGB: -1.97, GNN: -1.44.</p>",
      "votes": 2,
      "replies": [
        {
          "id": 603136,
          "author_name": "CoreyJamesLevinson",
          "author_url": "",
          "post_date": "2019-08-19T22:49:59.310000",
          "content": "<p>What is your keras NN cv Shuhao? :-)</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 603140,
          "author_name": "Shuhao Cao",
          "author_url": "",
          "post_date": "2019-08-19T23:09:51.783000",
          "content": "<p>Only -0.97 but I mainly use it to extract features.</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 604738,
      "author_name": "HarshitMehta",
      "author_url": "",
      "post_date": "2019-08-21T18:08:23.853000",
      "content": "<p>Our team best single Lightgbm = -1.745 and Gnn = -1.621. </p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 604345,
      "author_name": "JM",
      "author_url": "",
      "post_date": "2019-08-21T09:34:43.940000",
      "content": "<p>GNN -1.9, hardly even tuned yet, reckon it would top out -2 to -2.5.</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 603736,
      "author_name": "eagle4",
      "author_url": "",
      "post_date": "2019-08-20T15:52:28.017000",
      "content": "<p><a href=\"/jiweiliu\">@jiweiliu</a> : do you have a secret sauce to jump from -1.87 and -1.92 to -2.3 ?\nmy lgbm and gnn are in the same -1.8+ area but I can't figure out how to jump so far.</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 603322,
      "author_name": "fwj7",
      "author_url": "",
      "post_date": "2019-08-20T06:30:26.710000",
      "content": "<p>Let me guess：\n-2.0 and below : tree boost\n-2.5 to -2.0 : GNN\n-2.5 and above : GNN and chemistry domain knowledge</p>",
      "votes": 0,
      "replies": [
        {
          "id": 603356,
          "author_name": "Ahmet Erdem",
          "author_url": "",
          "post_date": "2019-08-20T07:16:06.323000",
          "content": "<p>I don't think Domain Knowledge is needed in this competition if you mean chemistry. It seems using deep learning for quantum chemistry is very common, the reason why domain experts are high on the leaderboard is probably because they are also good at deep learning on unstructred data.</p>",
          "votes": 8,
          "replies": []
        },
        {
          "id": 603357,
          "author_name": "CPMP",
          "author_url": "",
          "post_date": "2019-08-20T07:18:15.863000",
          "content": "<blockquote>\n  <p>the reason why domain experts are high on the leaderboard is probably because they are also good at deep learning on unstructred data.</p>\n</blockquote>\n\n<p>And also because they have a warm start with ready to use NNs.  </p>\n\n<p>The domain knowledge that seems relevant here is not chemistry, it is graph NNs.</p>",
          "votes": 7,
          "replies": []
        },
        {
          "id": 603364,
          "author_name": "CPMP",
          "author_url": "",
          "post_date": "2019-08-20T07:23:06.810000",
          "content": "<blockquote>\n  <p>2.0 and below : tree boost\n  -2.5 to -2.0 : GNN</p>\n</blockquote>\n\n<p>It is possible to get way better than -2 with tree boosting.</p>",
          "votes": 4,
          "replies": []
        },
        {
          "id": 603593,
          "author_name": "Jiwei Liu",
          "author_url": "",
          "post_date": "2019-08-20T13:27:59.163000",
          "content": "<blockquote>\n  <p>way better than -2 </p>\n</blockquote>\n\n<p>I could use more decimal digits 😊 </p>",
          "votes": 2,
          "replies": []
        },
        {
          "id": 604088,
          "author_name": "fwj7",
          "author_url": "",
          "post_date": "2019-08-21T02:43:07.950000",
          "content": "<p>May I ask what's your XGB LB score if your single XGB CV is -1.87? Thanks!</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 609175,
          "author_name": "vithng",
          "author_url": "",
          "post_date": "2019-08-27T13:15:54.843000",
          "content": "<p><a href=\"/cpmpml\">@cpmpml</a> Would you share this approach after competition?</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 605049,
      "author_name": "",
      "author_url": "",
      "post_date": "2019-08-22T02:31:10.817000",
      "content": "",
      "votes": 0,
      "replies": [
        {
          "id": 609177,
          "author_name": "CPMP",
          "author_url": "",
          "post_date": "2019-08-27T13:17:02.830000",
          "content": "<p>This forum ;)</p>",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "603060": "Our best xgb is about -1.87 and our best GNN is -1.92. How about yours?",
    "603233": "Conventional NN: -2.047",
    "603080": "Not fully trained LGBM: - 1.946",
    "603314": "You have some best models reports in this earlier thread: https://www.kaggle.com/c/champs-scalar-coupling/discussion/98444#latest-602104\n",
    "603205": "GNN: -1.981 (1fold of 3fold CV)\nI am interested in single model score of gold medal area players  : )",
    "603120": "LGB: -1.97, GNN: -1.44.",
    "604738": "Our team best single Lightgbm = -1.745 and Gnn = -1.621. ",
    "604345": "GNN -1.9, hardly even tuned yet, reckon it would top out -2 to -2.5.",
    "603736": "@jiweiliu : do you have a secret sauce to jump from -1.87 and -1.92 to -2.3 ?\nmy lgbm and gnn are in the same -1.8+ area but I can't figure out how to jump so far.",
    "603322": "Let me guess：\n-2.0 and below : tree boost\n-2.5 to -2.0 : GNN\n-2.5 and above : GNN and chemistry domain knowledge",
    "605049": ""
  }
}