{
  "id": 80253,
  "title": "Other type of models",
  "url": "/competitions/elo-merchant-category-recommendation/discussion/80253",
  "author_name": "Manoj",
  "post_date": "2019-02-12T06:18:50.413000",
  "votes": 1,
  "comment_count": 12,
  "views": 0,
  "content": "<p>Anybody using any other model apart from Lightgbm and Xgb\nMay be for diversification, can share some experience ?</p>",
  "messages": [
    {
      "id": 469965,
      "postDate": "2019-02-12T06:18:50.413Z",
      "content": "<p>Anybody using any other model apart from Lightgbm and Xgb\nMay be for diversification, can share some experience ?</p>",
      "rawMarkdown": "Anybody using any other model apart from Lightgbm and Xgb\nMay be for diversification, can share some experience ?",
      "votes": 1
    },
    {
      "id": 470203,
      "postDate": "2019-02-12T15:26:50.387Z",
      "content": "<p>NN result does not look consistent for me over 3.7</p>",
      "rawMarkdown": "NN result does not look consistent for me over 3.7",
      "votes": 2
    },
    {
      "id": 470133,
      "postDate": "2019-02-12T12:52:49.947Z",
      "content": "<p>I tried NN but didn't take good results from it. My best NN score was 3.707.</p>",
      "rawMarkdown": "I tried NN but didn't take good results from it. My best NN score was 3.707.",
      "replies": [
        {
          "id": 470134,
          "postDate": "2019-02-12T12:54:30.890Z",
          "content": "<p>same kind of pain here also for NN.</p>",
          "rawMarkdown": "same kind of pain here also for NN."
        }
      ]
    },
    {
      "id": 470023,
      "postDate": "2019-02-12T08:32:17.867Z",
      "content": "<p>even i am ready to hear from tops. I am only using LGB and XGB.</p>",
      "rawMarkdown": "even i am ready to hear from tops. I am only using LGB and XGB.",
      "replies": [
        {
          "id": 470024,
          "postDate": "2019-02-12T08:35:08.267Z",
          "content": "<p>Hi Subikash,\ndid you find any difference in the performance of LGB and XGB for the same data set ?</p>",
          "rawMarkdown": "Hi Subikash,\ndid you find any difference in the performance of LGB and XGB for the same data set ?"
        },
        {
          "id": 470025,
          "postDate": "2019-02-12T08:37:51.097Z",
          "content": "<p>yah..theres a gap but it adds some point in stacking. XGB is performing poor. My top XGB is 3.686.</p>",
          "rawMarkdown": "yah..theres a gap but it adds some point in stacking. XGB is performing poor. My top XGB is 3.686.",
          "votes": 1
        },
        {
          "id": 470058,
          "postDate": "2019-02-12T10:20:10Z",
          "content": "<p>Are your saying your LGB on same data is 3.675 compared to XGB @ 3.686 ?</p>",
          "rawMarkdown": "Are your saying your LGB on same data is 3.675 compared to XGB @ 3.686 ?"
        },
        {
          "id": 470060,
          "postDate": "2019-02-12T10:22:00.110Z",
          "content": "<p>In my case XGB is @ 3.686 and LGB is also @3.686\nHow did you optimized your hyper parameters ?</p>",
          "rawMarkdown": "In my case XGB is @ 3.686 and LGB is also @3.686\nHow did you optimized your hyper parameters ?",
          "votes": 1
        },
        {
          "id": 470099,
          "postDate": "2019-02-12T11:21:08.177Z",
          "content": "<p>I am curious to know too if you don't mind sharing, <a href=\"/subikashpal\">@subikashpal</a>. Basically both my LB and CV rmse don't change much after adding new features, although they show high importance in the lgb importance plot. I tried parameter tuning with optuna but the LB rmse gets worse. </p>",
          "rawMarkdown": "I am curious to know too if you don't mind sharing, @subikashpal. Basically both my LB and CV rmse don't change much after adding new features, although they show high importance in the lgb importance plot. I tried parameter tuning with optuna but the LB rmse gets worse. "
        },
        {
          "id": 470111,
          "postDate": "2019-02-12T11:59:50.697Z",
          "content": "<p>I am not spending time on parameter tuning. for XGB , i am taking it from some kernel. nothing changed from there. Sorry, i couldn't recollect which kernel.</p>\n\n<p>LGB - parameters also adopted from some kernel. </p>\n\n<p>Nothing fancy in terms of parameter tuning.</p>",
          "rawMarkdown": "I am not spending time on parameter tuning. for XGB , i am taking it from some kernel. nothing changed from there. Sorry, i couldn't recollect which kernel.\n\nLGB - parameters also adopted from some kernel. \n\nNothing fancy in terms of parameter tuning.\n",
          "votes": 1
        },
        {
          "id": 470413,
          "postDate": "2019-02-12T22:54:55.920Z",
          "content": "<p>I have had lb scores off by 0.011 if the regularization parameters of xgboost are not used, so I think that tuning these parameters could make a big difference.  If you don't want to take the time to do it, there are packages out there to do it for you.  Optuna for Python and mlr for r.</p>",
          "rawMarkdown": "I have had lb scores off by 0.011 if the regularization parameters of xgboost are not used, so I think that tuning these parameters could make a big difference.  If you don't want to take the time to do it, there are packages out there to do it for you.  Optuna for Python and mlr for r."
        }
      ]
    },
    {
      "id": 470307,
      "postDate": "2019-02-12T18:33:44.320Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 470203,
      "author_name": "Mr Loke",
      "author_url": "",
      "post_date": "2019-02-12T15:26:50.387000",
      "content": "<p>NN result does not look consistent for me over 3.7</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 470133,
      "author_name": "Murat Korkmaz",
      "author_url": "",
      "post_date": "2019-02-12T12:52:49.947000",
      "content": "<p>I tried NN but didn't take good results from it. My best NN score was 3.707.</p>",
      "votes": 0,
      "replies": [
        {
          "id": 470134,
          "author_name": "SubikashPal",
          "author_url": "",
          "post_date": "2019-02-12T12:54:30.890000",
          "content": "<p>same kind of pain here also for NN.</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 470023,
      "author_name": "SubikashPal",
      "author_url": "",
      "post_date": "2019-02-12T08:32:17.867000",
      "content": "<p>even i am ready to hear from tops. I am only using LGB and XGB.</p>",
      "votes": 0,
      "replies": [
        {
          "id": 470024,
          "author_name": "Manoj",
          "author_url": "",
          "post_date": "2019-02-12T08:35:08.267000",
          "content": "<p>Hi Subikash,\ndid you find any difference in the performance of LGB and XGB for the same data set ?</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 470025,
          "author_name": "SubikashPal",
          "author_url": "",
          "post_date": "2019-02-12T08:37:51.097000",
          "content": "<p>yah..theres a gap but it adds some point in stacking. XGB is performing poor. My top XGB is 3.686.</p>",
          "votes": 1,
          "replies": []
        },
        {
          "id": 470058,
          "author_name": "Manoj",
          "author_url": "",
          "post_date": "2019-02-12T10:20:10",
          "content": "<p>Are your saying your LGB on same data is 3.675 compared to XGB @ 3.686 ?</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 470060,
          "author_name": "Manoj",
          "author_url": "",
          "post_date": "2019-02-12T10:22:00.110000",
          "content": "<p>In my case XGB is @ 3.686 and LGB is also @3.686\nHow did you optimized your hyper parameters ?</p>",
          "votes": 1,
          "replies": []
        },
        {
          "id": 470099,
          "author_name": "FP",
          "author_url": "",
          "post_date": "2019-02-12T11:21:08.177000",
          "content": "<p>I am curious to know too if you don't mind sharing, <a href=\"/subikashpal\">@subikashpal</a>. Basically both my LB and CV rmse don't change much after adding new features, although they show high importance in the lgb importance plot. I tried parameter tuning with optuna but the LB rmse gets worse. </p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 470111,
          "author_name": "SubikashPal",
          "author_url": "",
          "post_date": "2019-02-12T11:59:50.697000",
          "content": "<p>I am not spending time on parameter tuning. for XGB , i am taking it from some kernel. nothing changed from there. Sorry, i couldn't recollect which kernel.</p>\n\n<p>LGB - parameters also adopted from some kernel. </p>\n\n<p>Nothing fancy in terms of parameter tuning.</p>",
          "votes": 1,
          "replies": []
        },
        {
          "id": 470413,
          "author_name": "Kyle Peters",
          "author_url": "",
          "post_date": "2019-02-12T22:54:55.920000",
          "content": "<p>I have had lb scores off by 0.011 if the regularization parameters of xgboost are not used, so I think that tuning these parameters could make a big difference.  If you don't want to take the time to do it, there are packages out there to do it for you.  Optuna for Python and mlr for r.</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 470307,
      "author_name": "",
      "author_url": "",
      "post_date": "2019-02-12T18:33:44.320000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "469965": "Anybody using any other model apart from Lightgbm and Xgb\nMay be for diversification, can share some experience ?",
    "470203": "NN result does not look consistent for me over 3.7",
    "470133": "I tried NN but didn't take good results from it. My best NN score was 3.707.",
    "470023": "even i am ready to hear from tops. I am only using LGB and XGB.",
    "470307": ""
  }
}