I don't know how to deal with the multi-label problem.
", "messages": [ { "id": 454480, "postDate": "2019-01-11T17:44:21.010Z", "content": "I found a method that seemed to work well. When augmenting with fastai, you work with a matrix whose dimensions include number of augmentations. I took multiple models and combined them into one matrix by stacking in the augmentation dimension. Then I averaged in this dimension, after removing the bottom 75% of predictions.
", "rawMarkdown": "I found a method that seemed to work well. When augmenting with fastai, you work with a matrix whose dimensions include number of augmentations. I took multiple models and combined them into one matrix by stacking in the augmentation dimension. Then I averaged in this dimension, after removing the bottom 75% of predictions." }, { "id": 450938, "postDate": "2019-01-06T05:28:12.473Z", "content": "If you are a beginner like me, maybe the tutorial of random forest or XGBoost can help you...
\n\nAlso, the machine learning tutorial itself is also recommended. It is a project about house price predict.
", "rawMarkdown": "If you are a beginner like me, maybe the tutorial of [random forest](https://www.kaggle.com/dansbecker/random-forests) or [XGBoost](https://www.kaggle.com/dansbecker/xgboost) can help you... \n\nAlso, the [machine learning tutorial](https://www.kaggle.com/learn/machine-learning) itself is also recommended. It is a project about house price predict. ", "replies": [ { "id": 454394, "postDate": "2019-01-11T15:15:32.730Z", "content": "Thanks！
", "rawMarkdown": "Thanks！" } ] }, { "id": 450924, "postDate": "2019-01-06T04:43:30.243Z", "content": "Ensembling depends on what you ensamble so you can't just take otger people's methods. However, a few ones that are general enough :
\n\nMmmm that's about it, i guess. Other ones are specific
", "rawMarkdown": "Ensembling depends on what you ensamble so you can't just take otger people's methods. However, a few ones that are general enough :\n\n1. Averaging the prediction probability. If you have one model that is better than the others, do weighted average \n2. Accumulating the predictions. Just add any predictions over the threshold that do no appear yet\n\nMmmm that's about it, i guess. Other ones are specific \n", "replies": [ { "id": 454395, "postDate": "2019-01-11T15:16:29.790Z", "content": "Thanks a lot，however I was busy these days，and I‘ll learn it .
", "rawMarkdown": "Thanks a lot，however I was busy these days，and I‘ll learn it ." } ] }, { "id": 450915, "postDate": "2019-01-06T04:33:27.793Z", "content": "I don't know how to deal with the multi-label problem.
", "rawMarkdown": "I don't know how to deal with the multi-label problem." }, { "id": 450962, "postDate": "2019-01-06T06:55:06.173Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 454480, "author_name": "pete", "author_url": "", "post_date": "2019-01-11T17:44:21.010000", "content": "I found a method that seemed to work well. When augmenting with fastai, you work with a matrix whose dimensions include number of augmentations. I took multiple models and combined them into one matrix by stacking in the augmentation dimension. Then I averaged in this dimension, after removing the bottom 75% of predictions.
", "votes": 0, "replies": [] }, { "id": 450938, "author_name": "Wang Ci", "author_url": "", "post_date": "2019-01-06T05:28:12.473000", "content": "If you are a beginner like me, maybe the tutorial of random forest or XGBoost can help you...
\n\nAlso, the machine learning tutorial itself is also recommended. It is a project about house price predict.
", "votes": 0, "replies": [ { "id": 454394, "author_name": "spectre", "author_url": "", "post_date": "2019-01-11T15:15:32.730000", "content": "Thanks！
", "votes": 0, "replies": [] } ] }, { "id": 450924, "author_name": "Moshel", "author_url": "", "post_date": "2019-01-06T04:43:30.243000", "content": "Ensembling depends on what you ensamble so you can't just take otger people's methods. However, a few ones that are general enough :
\n\nMmmm that's about it, i guess. Other ones are specific
", "votes": 0, "replies": [ { "id": 454395, "author_name": "spectre", "author_url": "", "post_date": "2019-01-11T15:16:29.790000", "content": "Thanks a lot，however I was busy these days，and I‘ll learn it .
", "votes": 0, "replies": [] } ] }, { "id": 450962, "author_name": "", "author_url": "", "post_date": "2019-01-06T06:55:06.173000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "454480": "I found a method that seemed to work well. When augmenting with fastai, you work with a matrix whose dimensions include number of augmentations. I took multiple models and combined them into one matrix by stacking in the augmentation dimension. Then I averaged in this dimension, after removing the bottom 75% of predictions.", "450938": "If you are a beginner like me, maybe the tutorial of [random forest](https://www.kaggle.com/dansbecker/random-forests) or [XGBoost](https://www.kaggle.com/dansbecker/xgboost) can help you... \n\nAlso, the [machine learning tutorial](https://www.kaggle.com/learn/machine-learning) itself is also recommended. It is a project about house price predict. ", "450924": "Ensembling depends on what you ensamble so you can't just take otger people's methods. However, a few ones that are general enough :\n\n1. Averaging the prediction probability. If you have one model that is better than the others, do weighted average \n2. Accumulating the predictions. Just add any predictions over the threshold that do no appear yet\n\nMmmm that's about it, i guess. Other ones are specific \n", "450915": "I don't know how to deal with the multi-label problem.", "450962": "" } }