![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2513130%2F389bdd933bafe2216416e21e08fe617e%2F20190821151305.png?generation=1566371624681533&alt=media\")
Now our best single model's LB is -2.740, ensemble is -2.815. \n16 days ago our best single model is -2.474 but ensemble can reach to -2.598, is it normal?
", "messages": [ { "id": 604170, "postDate": "2019-08-21T05:37:43.010Z", "content": "Now our best single model's LB is -2.740, ensemble is -2.815. \n16 days ago our best single model is -2.474 but ensemble can reach to -2.598, is it normal?
", "rawMarkdown": "Now our best single model's LB is -2.740, ensemble is -2.815. \n16 days ago our best single model is -2.474 but ensemble can reach to -2.598, is it normal?", "votes": 7 }, { "id": 604417, "postDate": "2019-08-21T11:11:39.737Z", "content": "Ensemble is like Random Forest. Random Forest with simple tree, it maybe lead great improve(usually +0.2? if binary auc). Howerver, Random Forest with GBDT, it just little improve (usually +0.05? if binary auc. it is called random seed average).\nIf you get nice score, ensemble won't work well because the model is already fitted data, I think.
", "rawMarkdown": "Ensemble is like Random Forest. Random Forest with simple tree, it maybe lead great improve(usually +0.2? if binary auc). Howerver, Random Forest with GBDT, it just little improve (usually +0.05? if binary auc. it is called random seed average).\nIf you get nice score, ensemble won't work well because the model is already fitted data, I think.", "votes": 1 }, { "id": 604201, "postDate": "2019-08-21T06:22:17.840Z", "content": "I think it is normal that improving over base models becomes harder when the base models get better.
", "rawMarkdown": "I think it is normal that improving over base models becomes harder when the base models get better.\n", "votes": 2, "replies": [ { "id": 604226, "postDate": "2019-08-21T07:11:16.003Z", "content": "Thanks, I feel at ease from your reply. ;)
", "rawMarkdown": "Thanks, I feel at ease from your reply. ;)", "votes": 1 } ] }, { "id": 604328, "postDate": "2019-08-21T09:18:24.513Z", "content": "I think that it really depends on what techniques already been used to the ensemble, if there's many lacking, probably very much.
", "rawMarkdown": "I think that it really depends on what techniques already been used to the ensemble, if there's many lacking, probably very much.", "replies": [ { "id": 604358, "postDate": "2019-08-21T09:52:43.957Z", "content": "what do you mean \"many lacking\"?
", "rawMarkdown": "what do you mean \"many lacking\"?" } ] }, { "id": 604223, "postDate": "2019-08-21T07:07:07.610Z", "content": "LightGBM? Could you share the cross-validation result for each type?
", "rawMarkdown": "LightGBM? Could you share the cross-validation result for each type?", "replies": [ { "id": 604228, "postDate": "2019-08-21T07:11:33.233Z", "content": "Definitely，GNN\n
look at here. https://www.kaggle.com/c/champs-scalar-coupling/discussion/98444#latest-602104
", "rawMarkdown": "look at here. https://www.kaggle.com/c/champs-scalar-coupling/discussion/98444#latest-602104", "votes": 1 } ] }, { "id": 604211, "postDate": "2019-08-21T06:36:09.943Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 604417, "author_name": "yCarbon", "author_url": "", "post_date": "2019-08-21T11:11:39.737000", "content": "Ensemble is like Random Forest. Random Forest with simple tree, it maybe lead great improve(usually +0.2? if binary auc). Howerver, Random Forest with GBDT, it just little improve (usually +0.05? if binary auc. it is called random seed average).\nIf you get nice score, ensemble won't work well because the model is already fitted data, I think.
", "votes": 1, "replies": [] }, { "id": 604201, "author_name": "CPMP", "author_url": "", "post_date": "2019-08-21T06:22:17.840000", "content": "I think it is normal that improving over base models becomes harder when the base models get better.
", "votes": 2, "replies": [ { "id": 604226, "author_name": "senkin13", "author_url": "", "post_date": "2019-08-21T07:11:16.003000", "content": "Thanks, I feel at ease from your reply. ;)
", "votes": 1, "replies": [] } ] }, { "id": 604328, "author_name": "Brun", "author_url": "", "post_date": "2019-08-21T09:18:24.513000", "content": "I think that it really depends on what techniques already been used to the ensemble, if there's many lacking, probably very much.
", "votes": 0, "replies": [ { "id": 604358, "author_name": "senkin13", "author_url": "", "post_date": "2019-08-21T09:52:43.957000", "content": "what do you mean \"many lacking\"?
", "votes": 0, "replies": [] } ] }, { "id": 604223, "author_name": "Aleksandr Kosolapov", "author_url": "", "post_date": "2019-08-21T07:07:07.610000", "content": "LightGBM? Could you share the cross-validation result for each type?
", "votes": 0, "replies": [ { "id": 604228, "author_name": "fwj7", "author_url": "", "post_date": "2019-08-21T07:11:33.233000", "content": "Definitely，GNN\n
look at here. https://www.kaggle.com/c/champs-scalar-coupling/discussion/98444#latest-602104
", "votes": 1, "replies": [] } ] }, { "id": 604211, "author_name": "", "author_url": "", "post_date": "2019-08-21T06:36:09.943000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "604170": "Now our best single model's LB is -2.740, ensemble is -2.815. \n16 days ago our best single model is -2.474 but ensemble can reach to -2.598, is it normal?", "604417": "Ensemble is like Random Forest. Random Forest with simple tree, it maybe lead great improve(usually +0.2? if binary auc). Howerver, Random Forest with GBDT, it just little improve (usually +0.05? if binary auc. it is called random seed average).\nIf you get nice score, ensemble won't work well because the model is already fitted data, I think.", "604201": "I think it is normal that improving over base models becomes harder when the base models get better.\n", "604328": "I think that it really depends on what techniques already been used to the ensemble, if there's many lacking, probably very much.", "604223": "LightGBM? Could you share the cross-validation result for each type?", "604211": "" } }