I am a new kaggler,I want to know how can I make cv,and how can I to ensemble the relult from different model?Thanks!Thank you very much!
", "messages": [ { "id": "568500", "postDate": "07/05/2019 02:57:15", "content": "I am a new kaggler,I want to know how can I make cv,and how can I to ensemble the relult from different model?Thanks!Thank you very much!
", "rawMarkdown": "I am a new kaggler,I want to know how can I make cv,and how can I to ensemble the relult from different model?Thanks!Thank you very much!", "votes": null }, { "id": "568518", "postDate": "07/05/2019 03:45:00", "content": "hello, I want to know, too.By the way, I want to know where are you from?
", "rawMarkdown": "hello, I want to know, too.By the way, I want to know where are you from?", "votes": null }, { "id": "568528", "postDate": "07/05/2019 04:22:05", "content": "Most simple way is averaging like (OOF1 + OOF2) / 2.
\nAfter that you can try weighted averaging and stacking.
\nKaggle ensemble guide is also helpful.
Make CV: \nSplit the data to K folds and train K models by using each fold as the validation dataset. Then you can get CV score by computing the average validation score of each fold.
\n\nEnsemble the result from different models:\nLoad each model and give out their corresponding predictions on the test dataset, then you can ensemble them.
", "rawMarkdown": "Make CV: \nSplit the data to K folds and train K models by using each fold as the validation dataset. Then you can get CV score by computing the average validation score of each fold.\n\nEnsemble the result from different models:\nLoad each model and give out their corresponding predictions on the test dataset, then you can ensemble them.", "votes": null }, { "id": "568539", "postDate": "07/05/2019 04:45:29", "content": "I usually just taking average of the model outputs (the logits, not the final labels), like [0.12, 1.23, 2.34, 3.45, 4.56] not [0, 1, 2, 3, 4], then if you use classification do argmax if you use regression do something with the threshold.
", "rawMarkdown": "I usually just taking average of the model outputs (the logits, not the final labels), like [0.12, 1.23, 2.34, 3.45, 4.56] not [0, 1, 2, 3, 4], then if you use classification do argmax if you use regression do something with the threshold.", "votes": null }, { "id": "615637", "postDate": "09/02/2019 07:42:23", "content": "@jionie any ideas on ensembling regression models? a few simple methods don't seem right. We can't average predictions as 1 output and 3 output will give a 2 output which is absurd. How effective would averaging raw predictions and using same optimized threshold on validation set during training be? What if one model predicts 1.2 and the other 3.9? Is averaging the most effective here?
", "rawMarkdown": "jionie any ideas on ensembling regression models? a few simple methods don't seem right. We can't average predictions as 1 output and 3 output will give a 2 output which is absurd. How effective would averaging raw predictions and using same optimized threshold on validation set during training be? What if one model predicts 1.2 and the other 3.9? Is averaging the most effective here?", "votes": null } ], "comments": [ { "id": 568518, "author_name": "dataml999", "author_url": "", "post_date": "07/05/2019 03:45:00", "content": "hello, I want to know, too.By the way, I want to know where are you from?
", "votes": null, "replies": [] }, { "id": 568528, "author_name": "takuok", "author_url": "", "post_date": "07/05/2019 04:22:05", "content": "Most simple way is averaging like (OOF1 + OOF2) / 2.
\nAfter that you can try weighted averaging and stacking.
\nKaggle ensemble guide is also helpful.
Make CV: \nSplit the data to K folds and train K models by using each fold as the validation dataset. Then you can get CV score by computing the average validation score of each fold.
\n\nEnsemble the result from different models:\nLoad each model and give out their corresponding predictions on the test dataset, then you can ensemble them.
", "votes": null, "replies": [] }, { "id": 568539, "author_name": "jionie", "author_url": "", "post_date": "07/05/2019 04:45:29", "content": "I usually just taking average of the model outputs (the logits, not the final labels), like [0.12, 1.23, 2.34, 3.45, 4.56] not [0, 1, 2, 3, 4], then if you use classification do argmax if you use regression do something with the threshold.
", "votes": null, "replies": [ { "id": 615637, "author_name": "yousof9", "author_url": "", "post_date": "09/02/2019 07:42:23", "content": "@jionie any ideas on ensembling regression models? a few simple methods don't seem right. We can't average predictions as 1 output and 3 output will give a 2 output which is absurd. How effective would averaging raw predictions and using same optimized threshold on validation set during training be? What if one model predicts 1.2 and the other 3.9? Is averaging the most effective here?
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "568500": "I am a new kaggler,I want to know how can I make cv,and how can I to ensemble the relult from different model?Thanks!Thank you very much!", "568518": "hello, I want to know, too.By the way, I want to know where are you from?", "568528": "Most simple way is averaging like (OOF1 + OOF2) / 2. \nAfter that you can try weighted averaging and stacking. \n[Kaggle ensemble guide](https://mlwave.com/kaggle-ensembling-guide/) is also helpful.", "568529": "Make CV: \nSplit the data to K folds and train K models by using each fold as the validation dataset. Then you can get CV score by computing the average validation score of each fold.\n\nEnsemble the result from different models:\nLoad each model and give out their corresponding predictions on the test dataset, then you can ensemble them.", "568539": "I usually just taking average of the model outputs (the logits, not the final labels), like [0.12, 1.23, 2.34, 3.45, 4.56] not [0, 1, 2, 3, 4], then if you use classification do argmax if you use regression do something with the threshold.", "615637": "jionie any ideas on ensembling regression models? a few simple methods don't seem right. We can't average predictions as 1 output and 3 output will give a 2 output which is absurd. How effective would averaging raw predictions and using same optimized threshold on validation set during training be? What if one model predicts 1.2 and the other 3.9? Is averaging the most effective here?" }, "source": "meta" }