I was wondering how people are getting such low RMSE. I dedicated a lot of time to feature engineering, but without much improvement. This leads me to believe that a lot of the improvement occurs with post-processing such as outlier detection and ensembling different predictions. Can people post their lb scores without post-processing and how much gain occured with outlier detection and enembling. Thank you.
", "messages": [ { "id": 473894, "postDate": "2019-02-18T16:48:14.877Z", "content": "I was wondering how people are getting such low RMSE. I dedicated a lot of time to feature engineering, but without much improvement. This leads me to believe that a lot of the improvement occurs with post-processing such as outlier detection and ensembling different predictions. Can people post their lb scores without post-processing and how much gain occured with outlier detection and enembling. Thank you.
", "rawMarkdown": "I was wondering how people are getting such low RMSE. I dedicated a lot of time to feature engineering, but without much improvement. This leads me to believe that a lot of the improvement occurs with post-processing such as outlier detection and ensembling different predictions. Can people post their lb scores without post-processing and how much gain occured with outlier detection and enembling. Thank you.", "votes": 5 }, { "id": 474900, "postDate": "2019-02-20T00:47:12.393Z", "content": "My best single model has CV 3.637 LB 3.679\nI am worried big gap...
", "rawMarkdown": "My best single model has CV 3.637 LB 3.679\nI am worried big gap...", "votes": 1, "replies": [ { "id": 476809, "postDate": "2019-02-23T07:48:50.783Z", "content": "me too
", "rawMarkdown": "me too" } ] }, { "id": 475202, "postDate": "2019-02-20T12:32:34.717Z", "content": "How you define post-process?
", "rawMarkdown": "How you define post-process?", "votes": 2, "replies": [ { "id": 475595, "postDate": "2019-02-21T00:44:56.207Z", "content": "@marcuslin, my understanding of post-processing is to modify the target values according to some heuristics or replace target values with other values based on some criteria.
", "rawMarkdown": "@marcuslin, my understanding of post-processing is to modify the target values according to some heuristics or replace target values with other values based on some criteria. ", "votes": 2 } ] }, { "id": 475317, "postDate": "2019-02-20T15:57:51.793Z", "content": "My best single model LightGBM has CV 3.648 LB 3.685 and XGB CV 3.645 LB 3.685
\n\nhttps://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/80103#469182
", "rawMarkdown": "My best single model LightGBM has CV 3.648 LB 3.685 and XGB CV 3.645 LB 3.685\n\nhttps://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/80103#469182\n" }, { "id": 474278, "postDate": "2019-02-19T06:49:10.860Z", "content": "can we have a talk?
\n\nMy best single model CV:3.641x LB:3.687
", "rawMarkdown": "can we have a talk?\n\nMy best single model CV:3.641x LB:3.687" }, { "id": 474890, "postDate": "2019-02-20T00:27:30.193Z", "rawMarkdown": "", "votes": 1, "isDeleted": true }, { "id": 473934, "postDate": "2019-02-18T17:55:10.063Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 474900, "author_name": "hatunina", "author_url": "", "post_date": "2019-02-20T00:47:12.393000", "content": "My best single model has CV 3.637 LB 3.679\nI am worried big gap...
", "votes": 1, "replies": [ { "id": 476809, "author_name": "Link", "author_url": "", "post_date": "2019-02-23T07:48:50.783000", "content": "me too
", "votes": 0, "replies": [] } ] }, { "id": 475202, "author_name": "Marcus Lin", "author_url": "", "post_date": "2019-02-20T12:32:34.717000", "content": "How you define post-process?
", "votes": 2, "replies": [ { "id": 475595, "author_name": "FP", "author_url": "", "post_date": "2019-02-21T00:44:56.207000", "content": "@marcuslin, my understanding of post-processing is to modify the target values according to some heuristics or replace target values with other values based on some criteria.
", "votes": 2, "replies": [] } ] }, { "id": 475317, "author_name": "Manoj", "author_url": "", "post_date": "2019-02-20T15:57:51.793000", "content": "My best single model LightGBM has CV 3.648 LB 3.685 and XGB CV 3.645 LB 3.685
\n\nhttps://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/80103#469182
", "votes": 0, "replies": [] }, { "id": 474278, "author_name": "Kang0709", "author_url": "", "post_date": "2019-02-19T06:49:10.860000", "content": "can we have a talk?
\n\nMy best single model CV:3.641x LB:3.687
", "votes": 0, "replies": [] }, { "id": 474890, "author_name": "", "author_url": "", "post_date": "2019-02-20T00:27:30.193000", "content": "", "votes": 1, "replies": [] }, { "id": 473934, "author_name": "", "author_url": "", "post_date": "2019-02-18T17:55:10.063000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "473894": "I was wondering how people are getting such low RMSE. I dedicated a lot of time to feature engineering, but without much improvement. This leads me to believe that a lot of the improvement occurs with post-processing such as outlier detection and ensembling different predictions. Can people post their lb scores without post-processing and how much gain occured with outlier detection and enembling. Thank you.", "474900": "My best single model has CV 3.637 LB 3.679\nI am worried big gap...", "475202": "How you define post-process?", "475317": "My best single model LightGBM has CV 3.648 LB 3.685 and XGB CV 3.645 LB 3.685\n\nhttps://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/80103#469182\n", "474278": "can we have a talk?\n\nMy best single model CV:3.641x LB:3.687", "474890": "", "473934": "" } }