{
  "id": 78233,
  "title": "Would anyone share the CV & LB score before and after post-processing?",
  "url": "/competitions/elo-merchant-category-recommendation/discussion/78233",
  "author_name": "",
  "post_date": "2019-01-21T13:15:36.379479600Z",
  "votes": 12,
  "comment_count": 36,
  "views": 0,
  "content": "<p>Since I only do this competition 5 days ago and I have 5 fold CV: 3.648 ~ 3.646（I shuffle the random seed）, and the LB score is 3.688(I only submit twice, so I have no idea how random seed influences the LB score.) . But when I see this discussion: <a href=\"https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/73517\">https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/73517</a>, I see many kagglers' CV are similar to me.  </p>\n\n<p>So, I am a little confused about how much  can you gain  with post processing. Would you like to share your corresponding CV&amp;LB score before and after post-processing?</p>",
  "messages": [
    {
      "id": "459260",
      "postDate": "01/21/2019 13:15:36",
      "content": "<p>Since I only do this competition 5 days ago and I have 5 fold CV: 3.648 ~ 3.646（I shuffle the random seed）, and the LB score is 3.688(I only submit twice, so I have no idea how random seed influences the LB score.) . But when I see this discussion: <a href=\"https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/73517\">https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/73517</a>, I see many kagglers' CV are similar to me.  </p>\n\n<p>So, I am a little confused about how much  can you gain  with post processing. Would you like to share your corresponding CV&amp;LB score before and after post-processing?</p>",
      "rawMarkdown": "Since I only do this competition 5 days ago and I have 5 fold CV: 3.648 ~ 3.646（I shuffle the random seed）, and the LB score is 3.688(I only submit twice, so I have no idea how random seed influences the LB score.) . But when I see this discussion: https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/73517, I see many kagglers' CV are similar to me.  \n\nSo, I am a little confused about how much  can you gain  with post processing. Would you like to share your corresponding CV&amp;LB score before and after post-processing?",
      "votes": null
    },
    {
      "id": "459274",
      "postDate": "01/21/2019 13:44:01",
      "content": "<p>Hi LongYin,</p>\n\n<p>i would not say post processing but i did some ensemble and stacking that gave me boost of  .003 over my best single model</p>\n\n<p><a href=\"https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/77849\">https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/77849</a></p>",
      "rawMarkdown": "Hi LongYin,\n\ni would not say post processing but i did some ensemble and stacking that gave me boost of  .003 over my best single model\n\nhttps://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/77849",
      "votes": null
    },
    {
      "id": "459532",
      "postDate": "01/21/2019 23:42:07",
      "content": "<p>i try post processing(<a href=\"https://www.kaggle.com/waitingli/combining-your-model-with-a-model-without-outlier\">https://www.kaggle.com/waitingli/combining-your-model-with-a-model-without-outlier</a>) with my best single model. but score \n<code>3.688 -&amp;gt; 3.705</code> (CV : 3.652)</p>",
      "rawMarkdown": "i try post processing(https://www.kaggle.com/waitingli/combining-your-model-with-a-model-without-outlier) with my best single model. but score \n`3.688 -&gt; 3.705` (CV : 3.652)",
      "votes": null
    },
    {
      "id": "459533",
      "postDate": "01/21/2019 23:54:15",
      "content": "<p>Before post-processing CV: 3.655 LB: 3.689.  After post processing LB: 3.685.</p>",
      "rawMarkdown": "Before post-processing CV: 3.655 LB: 3.689.  After post processing LB: 3.685.",
      "votes": null
    },
    {
      "id": "459546",
      "postDate": "01/22/2019 00:48:24",
      "content": "<p>Thanks for sharing.</p>",
      "rawMarkdown": "Thanks for sharing.",
      "votes": null
    },
    {
      "id": "459547",
      "postDate": "01/22/2019 00:49:19",
      "content": "<p>Thanks for sharing. I will try that next time.</p>",
      "rawMarkdown": "Thanks for sharing. I will try that next time.",
      "votes": null
    },
    {
      "id": "459548",
      "postDate": "01/22/2019 00:49:40",
      "content": "<p>Thanks for sharing.</p>",
      "rawMarkdown": "Thanks for sharing.",
      "votes": null
    },
    {
      "id": "459556",
      "postDate": "01/22/2019 01:28:34",
      "content": "<p>Actually, post processing should get 4-6/1000 points imp</p>",
      "rawMarkdown": "Actually, post processing should get 4-6/1000 points imp",
      "votes": null
    },
    {
      "id": "459557",
      "postDate": "01/22/2019 01:29:50",
      "content": "<p>Seems i should do more ensemble and stacking</p>",
      "rawMarkdown": "Seems i should do more ensemble and stacking",
      "votes": null
    },
    {
      "id": "459563",
      "postDate": "01/22/2019 02:07:24",
      "content": "<p>May I ask what do you mean shuffle the random seed ?</p>",
      "rawMarkdown": "May I ask what do you mean shuffle the random seed ?",
      "votes": null
    },
    {
      "id": "459566",
      "postDate": "01/22/2019 02:13:03",
      "content": "<p>i did different post processing approach. <code>3.687 -&amp;gt; 3.684</code> but i think that it is overfitting in the LB :D</p>",
      "rawMarkdown": "i did different post processing approach. `3.687 -&gt; 3.684` but i think that it is overfitting in the LB :D",
      "votes": null
    },
    {
      "id": "459582",
      "postDate": "01/22/2019 03:23:52",
      "content": "<p>Like the example below.</p>\n\n<p>'''\nfrom sklearn.model_selection import StratifiedKFold,KFold</p>\n\n<p>kf = KFold(n_splits=5,shuffle=True,random_state=1)</p>\n\n<p>'''</p>",
      "rawMarkdown": "Like the example below.\n\n'''\nfrom sklearn.model_selection import StratifiedKFold,KFold\n\nkf = KFold(n_splits=5,shuffle=True,random_state=1)\n\n'''",
      "votes": null
    },
    {
      "id": "459589",
      "postDate": "01/22/2019 03:41:42",
      "content": "<p>@Manoj -\nHow many different model did u use?</p>",
      "rawMarkdown": "Manoj -\nHow many different model did u use?",
      "votes": null
    },
    {
      "id": "459591",
      "postDate": "01/22/2019 03:47:21",
      "content": "<p>Got it</p>",
      "rawMarkdown": "Got it",
      "votes": null
    },
    {
      "id": "459593",
      "postDate": "01/22/2019 03:49:39",
      "content": "<p>Hi Subikash,\nI have used 8 models all of them are lightgbm so far.</p>",
      "rawMarkdown": "Hi Subikash,\nI have used 8 models all of them are lightgbm so far.",
      "votes": null
    },
    {
      "id": "459626",
      "postDate": "01/22/2019 05:14:46",
      "content": "<p>Thanks Manoj. All these are having different features ?</p>",
      "rawMarkdown": "Thanks Manoj. All these are having different features ?",
      "votes": null
    },
    {
      "id": "459660",
      "postDate": "01/22/2019 06:28:41",
      "content": "<p>LB: 3.695 (CV : 3.65498) -&gt; 3.690</p>",
      "rawMarkdown": "LB: 3.695 (CV : 3.65498) -&gt; 3.690",
      "votes": null
    },
    {
      "id": "459662",
      "postDate": "01/22/2019 06:31:30",
      "content": "<p>btw, I tried to use the merchants.csv, but it leads to bad results. Does anybody know how to use it?</p>",
      "rawMarkdown": "btw, I tried to use the merchants.csv, but it leads to bad results. Does anybody know how to use it?",
      "votes": null
    },
    {
      "id": "459667",
      "postDate": "01/22/2019 06:40:45",
      "content": "<p>Top 80-90 features are common, rest i am choosing and dropping randomly. I still experimenting, I have not yet build a proper pipeline</p>",
      "rawMarkdown": "Top 80-90 features are common, rest i am choosing and dropping randomly. I still experimenting, I have not yet build a proper pipeline",
      "votes": null
    },
    {
      "id": "459673",
      "postDate": "01/22/2019 06:49:23",
      "content": "<p><a href=\"https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/77987#458060\">https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/77987#458060</a></p>",
      "rawMarkdown": "https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/77987#458060",
      "votes": null
    },
    {
      "id": "459851",
      "postDate": "01/22/2019 13:48:42",
      "content": "<p>You mean you can get nearly 3.658 without post processing, this is amazing.</p>",
      "rawMarkdown": "You mean you can get nearly 3.658 without post processing, this is amazing.",
      "votes": null
    },
    {
      "id": "460117",
      "postDate": "01/23/2019 02:58:16",
      "content": "<p>Yes，Stacking really helps.</p>",
      "rawMarkdown": "Yes，Stacking really helps.",
      "votes": null
    },
    {
      "id": "460189",
      "postDate": "01/23/2019 06:36:32",
      "content": "<p>Single Model LGB = LB: 3.681 (CV : 3.6688) \nAfter post processing = LB: 3.673</p>",
      "rawMarkdown": "Single Model LGB = LB: 3.681 (CV : 3.6688) \nAfter post processing = LB: 3.673",
      "votes": null
    },
    {
      "id": "460191",
      "postDate": "01/23/2019 06:39:33",
      "content": "<p>Single LGB means one lgb model  or 5-fold LGB?</p>",
      "rawMarkdown": "Single LGB means one lgb model  or 5-fold LGB?",
      "votes": null
    },
    {
      "id": "460375",
      "postDate": "01/23/2019 15:15:31",
      "content": "<p>5 fold LGB.</p>",
      "rawMarkdown": "5 fold LGB.",
      "votes": null
    },
    {
      "id": "460552",
      "postDate": "01/24/2019 00:22:11",
      "content": "<p>thanks</p>",
      "rawMarkdown": "thanks",
      "votes": null
    },
    {
      "id": "460651",
      "postDate": "01/24/2019 07:01:35",
      "content": "<p>your cv score is impressive. I think it shows a low risk of overfitting. </p>",
      "rawMarkdown": "your cv score is impressive. I think it shows a low risk of overfitting.",
      "votes": null
    },
    {
      "id": "460654",
      "postDate": "01/24/2019 07:02:36",
      "content": "<p>For  outlier post-processing, would anyone share that how you decide the threshold? 2207/201917?</p>",
      "rawMarkdown": "For  outlier post-processing, would anyone share that how you decide the threshold? 2207/201917?",
      "votes": null
    },
    {
      "id": "460791",
      "postDate": "01/24/2019 12:30:02",
      "content": "<p>Sadly I got a huge gap between CV and LB.\nCV: 3.64882 LB: 3.693</p>",
      "rawMarkdown": "Sadly I got a huge gap between CV and LB.\nCV: 3.64882 LB: 3.693",
      "votes": null
    },
    {
      "id": "461419",
      "postDate": "01/26/2019 01:31:31",
      "content": "<p>This is reasonable, now, my cv is 3.6395 and LB is 3.679. The rest is from the kernel part by combining with binary classification.</p>",
      "rawMarkdown": "This is reasonable, now, my cv is 3.6395 and LB is 3.679. The rest is from the kernel part by combining with binary classification.",
      "votes": null
    },
    {
      "id": "461849",
      "postDate": "01/27/2019 06:35:22",
      "content": "<p>Sadly I got a better  CV is 3.648 but LB is 3.687\nbest LB is 3.683 and CV is3.6504\nI don't know how to deal with it.</p>",
      "rawMarkdown": "Sadly I got a better  CV is 3.648 but LB is 3.687\nbest LB is 3.683 and CV is3.6504\nI don't know how to deal with it.",
      "votes": null
    },
    {
      "id": "461855",
      "postDate": "01/27/2019 07:05:43",
      "content": "<p>Thanks for sharing.</p>",
      "rawMarkdown": "Thanks for sharing.",
      "votes": null
    },
    {
      "id": "462841",
      "postDate": "01/29/2019 02:03:26",
      "content": "<p>@Longyin Hi Longyin! may u willing to have a talk with u ,then consider if team up with us </p>",
      "rawMarkdown": "Longyin Hi Longyin! may u willing to have a talk with u ,then consider if team up with us",
      "votes": null
    },
    {
      "id": "463030",
      "postDate": "01/29/2019 09:29:10",
      "content": "<p>Thanks for your kind invitation.  I have formed a team.  I am looking forward to having opportunities for future cooperation.</p>",
      "rawMarkdown": "Thanks for your kind invitation.  I have formed a team.  I am looking forward to having opportunities for future cooperation.",
      "votes": null
    },
    {
      "id": "463044",
      "postDate": "01/29/2019 09:54:21",
      "content": "<p>okay, thanks haha</p>",
      "rawMarkdown": "okay, thanks haha",
      "votes": null
    },
    {
      "id": "463986",
      "postDate": "01/31/2019 02:30:12",
      "content": "<p>Both you guys are excellent!</p>",
      "rawMarkdown": "Both you guys are excellent!",
      "votes": null
    },
    {
      "id": "464675",
      "postDate": "02/01/2019 09:24:35",
      "content": "<p>Without post processing.\nSingle model 9 fold CV score 3.648; LB score 3.680.\nBreak up of above score based on outliers\nScore only for Non-Outlier Targets: 1.826 \nScore only for Outlier Targets: 30.261</p>",
      "rawMarkdown": "Without post processing.\nSingle model 9 fold CV score 3.648; LB score 3.680.\nBreak up of above score based on outliers\nScore only for Non-Outlier Targets: 1.826 \nScore only for Outlier Targets: 30.261",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 459274,
      "author_name": "mks2192",
      "author_url": "",
      "post_date": "01/21/2019 13:44:01",
      "content": "<p>Hi LongYin,</p>\n\n<p>i would not say post processing but i did some ensemble and stacking that gave me boost of  .003 over my best single model</p>\n\n<p><a href=\"https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/77849\">https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/77849</a></p>",
      "votes": null,
      "replies": [
        {
          "id": 459546,
          "author_name": "longyin2",
          "author_url": "",
          "post_date": "01/22/2019 00:48:24",
          "content": "<p>Thanks for sharing.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 459557,
          "author_name": "codlife",
          "author_url": "",
          "post_date": "01/22/2019 01:29:50",
          "content": "<p>Seems i should do more ensemble and stacking</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 459589,
          "author_name": "subikashpal",
          "author_url": "",
          "post_date": "01/22/2019 03:41:42",
          "content": "<p>@Manoj -\nHow many different model did u use?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 459593,
          "author_name": "mks2192",
          "author_url": "",
          "post_date": "01/22/2019 03:49:39",
          "content": "<p>Hi Subikash,\nI have used 8 models all of them are lightgbm so far.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 459626,
          "author_name": "subikashpal",
          "author_url": "",
          "post_date": "01/22/2019 05:14:46",
          "content": "<p>Thanks Manoj. All these are having different features ?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 459667,
          "author_name": "mks2192",
          "author_url": "",
          "post_date": "01/22/2019 06:40:45",
          "content": "<p>Top 80-90 features are common, rest i am choosing and dropping randomly. I still experimenting, I have not yet build a proper pipeline</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 460117,
          "author_name": "longyin2",
          "author_url": "",
          "post_date": "01/23/2019 02:58:16",
          "content": "<p>Yes，Stacking really helps.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 459532,
      "author_name": "chocozzz",
      "author_url": "",
      "post_date": "01/21/2019 23:42:07",
      "content": "<p>i try post processing(<a href=\"https://www.kaggle.com/waitingli/combining-your-model-with-a-model-without-outlier\">https://www.kaggle.com/waitingli/combining-your-model-with-a-model-without-outlier</a>) with my best single model. but score \n<code>3.688 -&amp;gt; 3.705</code> (CV : 3.652)</p>",
      "votes": null,
      "replies": [
        {
          "id": 459547,
          "author_name": "longyin2",
          "author_url": "",
          "post_date": "01/22/2019 00:49:19",
          "content": "<p>Thanks for sharing. I will try that next time.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 459566,
          "author_name": "chocozzz",
          "author_url": "",
          "post_date": "01/22/2019 02:13:03",
          "content": "<p>i did different post processing approach. <code>3.687 -&amp;gt; 3.684</code> but i think that it is overfitting in the LB :D</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 459533,
      "author_name": "fatall",
      "author_url": "",
      "post_date": "01/21/2019 23:54:15",
      "content": "<p>Before post-processing CV: 3.655 LB: 3.689.  After post processing LB: 3.685.</p>",
      "votes": null,
      "replies": [
        {
          "id": 459548,
          "author_name": "longyin2",
          "author_url": "",
          "post_date": "01/22/2019 00:49:40",
          "content": "<p>Thanks for sharing.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 459556,
          "author_name": "codlife",
          "author_url": "",
          "post_date": "01/22/2019 01:28:34",
          "content": "<p>Actually, post processing should get 4-6/1000 points imp</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 459851,
          "author_name": "longyin2",
          "author_url": "",
          "post_date": "01/22/2019 13:48:42",
          "content": "<p>You mean you can get nearly 3.658 without post processing, this is amazing.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 459563,
      "author_name": "marcuslin",
      "author_url": "",
      "post_date": "01/22/2019 02:07:24",
      "content": "<p>May I ask what do you mean shuffle the random seed ?</p>",
      "votes": null,
      "replies": [
        {
          "id": 459582,
          "author_name": "longyin2",
          "author_url": "",
          "post_date": "01/22/2019 03:23:52",
          "content": "<p>Like the example below.</p>\n\n<p>'''\nfrom sklearn.model_selection import StratifiedKFold,KFold</p>\n\n<p>kf = KFold(n_splits=5,shuffle=True,random_state=1)</p>\n\n<p>'''</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 459591,
          "author_name": "marcuslin",
          "author_url": "",
          "post_date": "01/22/2019 03:47:21",
          "content": "<p>Got it</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 459660,
      "author_name": "lmdeliangmi",
      "author_url": "",
      "post_date": "01/22/2019 06:28:41",
      "content": "<p>LB: 3.695 (CV : 3.65498) -&gt; 3.690</p>",
      "votes": null,
      "replies": [
        {
          "id": 459662,
          "author_name": "lmdeliangmi",
          "author_url": "",
          "post_date": "01/22/2019 06:31:30",
          "content": "<p>btw, I tried to use the merchants.csv, but it leads to bad results. Does anybody know how to use it?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 459673,
          "author_name": "chocozzz",
          "author_url": "",
          "post_date": "01/22/2019 06:49:23",
          "content": "<p><a href=\"https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/77987#458060\">https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/77987#458060</a></p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 460189,
      "author_name": "subikashpal",
      "author_url": "",
      "post_date": "01/23/2019 06:36:32",
      "content": "<p>Single Model LGB = LB: 3.681 (CV : 3.6688) \nAfter post processing = LB: 3.673</p>",
      "votes": null,
      "replies": [
        {
          "id": 460191,
          "author_name": "longyin2",
          "author_url": "",
          "post_date": "01/23/2019 06:39:33",
          "content": "<p>Single LGB means one lgb model  or 5-fold LGB?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 460375,
          "author_name": "subikashpal",
          "author_url": "",
          "post_date": "01/23/2019 15:15:31",
          "content": "<p>5 fold LGB.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 460552,
          "author_name": "longyin2",
          "author_url": "",
          "post_date": "01/24/2019 00:22:11",
          "content": "<p>thanks</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 460651,
          "author_name": "lmdeliangmi",
          "author_url": "",
          "post_date": "01/24/2019 07:01:35",
          "content": "<p>your cv score is impressive. I think it shows a low risk of overfitting. </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 460654,
      "author_name": "lmdeliangmi",
      "author_url": "",
      "post_date": "01/24/2019 07:02:36",
      "content": "<p>For  outlier post-processing, would anyone share that how you decide the threshold? 2207/201917?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 460791,
      "author_name": "kingychiu",
      "author_url": "",
      "post_date": "01/24/2019 12:30:02",
      "content": "<p>Sadly I got a huge gap between CV and LB.\nCV: 3.64882 LB: 3.693</p>",
      "votes": null,
      "replies": [
        {
          "id": 461419,
          "author_name": "longyin2",
          "author_url": "",
          "post_date": "01/26/2019 01:31:31",
          "content": "<p>This is reasonable, now, my cv is 3.6395 and LB is 3.679. The rest is from the kernel part by combining with binary classification.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 461849,
      "author_name": "process531",
      "author_url": "",
      "post_date": "01/27/2019 06:35:22",
      "content": "<p>Sadly I got a better  CV is 3.648 but LB is 3.687\nbest LB is 3.683 and CV is3.6504\nI don't know how to deal with it.</p>",
      "votes": null,
      "replies": [
        {
          "id": 461855,
          "author_name": "longyin2",
          "author_url": "",
          "post_date": "01/27/2019 07:05:43",
          "content": "<p>Thanks for sharing.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 462841,
      "author_name": "laplaceplanet",
      "author_url": "",
      "post_date": "01/29/2019 02:03:26",
      "content": "<p>@Longyin Hi Longyin! may u willing to have a talk with u ,then consider if team up with us </p>",
      "votes": null,
      "replies": [
        {
          "id": 463030,
          "author_name": "longyin2",
          "author_url": "",
          "post_date": "01/29/2019 09:29:10",
          "content": "<p>Thanks for your kind invitation.  I have formed a team.  I am looking forward to having opportunities for future cooperation.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 463044,
          "author_name": "laplaceplanet",
          "author_url": "",
          "post_date": "01/29/2019 09:54:21",
          "content": "<p>okay, thanks haha</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 463986,
          "author_name": "codlife",
          "author_url": "",
          "post_date": "01/31/2019 02:30:12",
          "content": "<p>Both you guys are excellent!</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 464675,
      "author_name": "azacharia",
      "author_url": "",
      "post_date": "02/01/2019 09:24:35",
      "content": "<p>Without post processing.\nSingle model 9 fold CV score 3.648; LB score 3.680.\nBreak up of above score based on outliers\nScore only for Non-Outlier Targets: 1.826 \nScore only for Outlier Targets: 30.261</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "459260": "Since I only do this competition 5 days ago and I have 5 fold CV: 3.648 ~ 3.646（I shuffle the random seed）, and the LB score is 3.688(I only submit twice, so I have no idea how random seed influences the LB score.) . But when I see this discussion: https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/73517, I see many kagglers' CV are similar to me.  \n\nSo, I am a little confused about how much  can you gain  with post processing. Would you like to share your corresponding CV&amp;LB score before and after post-processing?",
    "459274": "Hi LongYin,\n\ni would not say post processing but i did some ensemble and stacking that gave me boost of  .003 over my best single model\n\nhttps://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/77849",
    "459532": "i try post processing(https://www.kaggle.com/waitingli/combining-your-model-with-a-model-without-outlier) with my best single model. but score \n`3.688 -&gt; 3.705` (CV : 3.652)",
    "459533": "Before post-processing CV: 3.655 LB: 3.689.  After post processing LB: 3.685.",
    "459546": "Thanks for sharing.",
    "459547": "Thanks for sharing. I will try that next time.",
    "459548": "Thanks for sharing.",
    "459556": "Actually, post processing should get 4-6/1000 points imp",
    "459557": "Seems i should do more ensemble and stacking",
    "459563": "May I ask what do you mean shuffle the random seed ?",
    "459566": "i did different post processing approach. `3.687 -&gt; 3.684` but i think that it is overfitting in the LB :D",
    "459582": "Like the example below.\n\n'''\nfrom sklearn.model_selection import StratifiedKFold,KFold\n\nkf = KFold(n_splits=5,shuffle=True,random_state=1)\n\n'''",
    "459589": "Manoj -\nHow many different model did u use?",
    "459591": "Got it",
    "459593": "Hi Subikash,\nI have used 8 models all of them are lightgbm so far.",
    "459626": "Thanks Manoj. All these are having different features ?",
    "459660": "LB: 3.695 (CV : 3.65498) -&gt; 3.690",
    "459662": "btw, I tried to use the merchants.csv, but it leads to bad results. Does anybody know how to use it?",
    "459667": "Top 80-90 features are common, rest i am choosing and dropping randomly. I still experimenting, I have not yet build a proper pipeline",
    "459673": "https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/77987#458060",
    "459851": "You mean you can get nearly 3.658 without post processing, this is amazing.",
    "460117": "Yes，Stacking really helps.",
    "460189": "Single Model LGB = LB: 3.681 (CV : 3.6688) \nAfter post processing = LB: 3.673",
    "460191": "Single LGB means one lgb model  or 5-fold LGB?",
    "460375": "5 fold LGB.",
    "460552": "thanks",
    "460651": "your cv score is impressive. I think it shows a low risk of overfitting.",
    "460654": "For  outlier post-processing, would anyone share that how you decide the threshold? 2207/201917?",
    "460791": "Sadly I got a huge gap between CV and LB.\nCV: 3.64882 LB: 3.693",
    "461419": "This is reasonable, now, my cv is 3.6395 and LB is 3.679. The rest is from the kernel part by combining with binary classification.",
    "461849": "Sadly I got a better  CV is 3.648 but LB is 3.687\nbest LB is 3.683 and CV is3.6504\nI don't know how to deal with it.",
    "461855": "Thanks for sharing.",
    "462841": "Longyin Hi Longyin! may u willing to have a talk with u ,then consider if team up with us",
    "463030": "Thanks for your kind invitation.  I have formed a team.  I am looking forward to having opportunities for future cooperation.",
    "463044": "okay, thanks haha",
    "463986": "Both you guys are excellent!",
    "464675": "Without post processing.\nSingle model 9 fold CV score 3.648; LB score 3.680.\nBreak up of above score based on outliers\nScore only for Non-Outlier Targets: 1.826 \nScore only for Outlier Targets: 30.261"
  },
  "source": "meta"
}