As is discussed by many people, we often use KFold to get cv scores, but how to get scores when doing stacking? and How to tuning parameters of the second/third level estimators?
", "messages": [ { "id": "479406", "postDate": "02/27/2019 05:39:58", "content": "As is discussed by many people, we often use KFold to get cv scores, but how to get scores when doing stacking? and How to tuning parameters of the second/third level estimators?
", "rawMarkdown": "As is discussed by many people, we often use KFold to get cv scores, but how to get scores when doing stacking? and How to tuning parameters of the second/third level estimators?", "votes": null }, { "id": "479466", "postDate": "02/27/2019 06:57:05", "content": "Uhm. You can do with kfold too !
", "rawMarkdown": "Uhm. You can do with kfold too !", "votes": null }, { "id": "479693", "postDate": "02/27/2019 10:47:17", "content": "Maybe these can help! :)
\n\nhttp://blog.kaggle.com/2016/12/27/a-kagglers-guide-to-model-stacking-in-practice/\nhttps://www.kaggle.com/general/18793
", "rawMarkdown": "Maybe these can help! :)\n\nhttp://blog.kaggle.com/2016/12/27/a-kagglers-guide-to-model-stacking-in-practice/\nhttps://www.kaggle.com/general/18793", "votes": null }, { "id": "479751", "postDate": "02/27/2019 11:48:09", "content": "Thank you, and maybe the seed of kfold for the second level should be different from the first level kfold seed?
", "rawMarkdown": "Thank you, and maybe the seed of kfold for the second level should be different from the first level kfold seed?", "votes": null }, { "id": "479752", "postDate": "02/27/2019 11:48:26", "content": "Thank you, and I'll read it carefully.
", "rawMarkdown": "Thank you, and I'll read it carefully.", "votes": null }, { "id": "479888", "postDate": "02/27/2019 14:33:37", "content": "the first and very first thing to do is to create the validation scheme and write the fold assignments into the raw data CSV. It makes stacking & blending so much easier, since you can get the fold assignments right away from the dataframe at each stage of your pipeline.
\n\nFor stacking:
\n\nYou are training a model with your existing training set predictions as features, and the training target as target:
\n\nTraining Phase (try to keep the original K-fold validation scheme):
\n\nSubmission Phase:
\n\nThank you for your detailed and professional answers. Your answer has helped me a lot and solved my doubts.
", "rawMarkdown": "Thank you for your detailed and professional answers. Your answer has helped me a lot and solved my doubts.", "votes": null } ], "comments": [ { "id": 479466, "author_name": "sc0d3r", "author_url": "", "post_date": "02/27/2019 06:57:05", "content": "Uhm. You can do with kfold too !
", "votes": null, "replies": [ { "id": 479751, "author_name": "liewmanchoi", "author_url": "", "post_date": "02/27/2019 11:48:09", "content": "Thank you, and maybe the seed of kfold for the second level should be different from the first level kfold seed?
", "votes": null, "replies": [] } ] }, { "id": 479693, "author_name": "iamstark", "author_url": "", "post_date": "02/27/2019 10:47:17", "content": "Maybe these can help! :)
\n\nhttp://blog.kaggle.com/2016/12/27/a-kagglers-guide-to-model-stacking-in-practice/\nhttps://www.kaggle.com/general/18793
", "votes": null, "replies": [ { "id": 479752, "author_name": "liewmanchoi", "author_url": "", "post_date": "02/27/2019 11:48:26", "content": "Thank you, and I'll read it carefully.
", "votes": null, "replies": [] } ] }, { "id": 479888, "author_name": "kingychiu", "author_url": "", "post_date": "02/27/2019 14:33:37", "content": "the first and very first thing to do is to create the validation scheme and write the fold assignments into the raw data CSV. It makes stacking & blending so much easier, since you can get the fold assignments right away from the dataframe at each stage of your pipeline.
\n\nFor stacking:
\n\nYou are training a model with your existing training set predictions as features, and the training target as target:
\n\nTraining Phase (try to keep the original K-fold validation scheme):
\n\nSubmission Phase:
\n\nThank you for your detailed and professional answers. Your answer has helped me a lot and solved my doubts.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "479406": "As is discussed by many people, we often use KFold to get cv scores, but how to get scores when doing stacking? and How to tuning parameters of the second/third level estimators?", "479466": "Uhm. You can do with kfold too !", "479693": "Maybe these can help! :)\n\nhttp://blog.kaggle.com/2016/12/27/a-kagglers-guide-to-model-stacking-in-practice/\nhttps://www.kaggle.com/general/18793", "479751": "Thank you, and maybe the seed of kfold for the second level should be different from the first level kfold seed?", "479752": "Thank you, and I'll read it carefully.", "479888": "the first and very first thing to do is to create the validation scheme and **write the fold assignments into the raw data CSV.** It makes stacking & blending so much easier, since you can get the fold assignments right away from the dataframe at each stage of your pipeline.\n\n\n\nFor stacking:\n\nYou are training a model with your existing training set predictions as features, and the training target as target:\n\nTraining Phase (try to keep the original K-fold validation scheme):\n\n- Feature 1: Train Predictions of Model 1\n- Feature 2: Train Predictions of Model 2\n- Feature N: Train Predictions of Model N\n- Target: the Train Target\n\nSubmission Phase:\n\n- Feature 1: Submission of Model 1\n- Feature 2: Submission of Model 2\n- Feature N: Submission of Model N\n- Prediction: your stacking submission.", "480377": "Thank you for your detailed and professional answers. Your answer has helped me a lot and solved my doubts." }, "source": "meta" }