I've been very intrigued by how well KNN has done in this competition. It's the first time I've seen that algorithm dominate, at least in the Kaggle scripts. I've been learning a lot, but haven't found R KNN packages that rival what is coming out of python sklearn. Are there any good KNN packages that are efficient and handle Manhattan distance?
\n\nI've tried:
\n\nI've been very intrigued by how well KNN has done in this competition. It's the first time I've seen that algorithm dominate, at least in the Kaggle scripts. I've been learning a lot, but haven't found R KNN packages that rival what is coming out of python sklearn. Are there any good KNN packages that are efficient and handle Manhattan distance?
\n\nI've tried:
\n\nCheck out RANN.L1.
\n\nIt doesn't have the features you find in python, but it gets the job done.
", "rawMarkdown": "Check out [RANN.L1][1].\r\n\r\nIt doesn't have the features you find in python, but it gets the job done.\r\n\r\n\r\n [1]: https://cran.r-project.org/web/packages/RANN.L1/RANN.L1.pdf", "votes": null }, { "id": "125469", "postDate": "06/29/2016 18:47:56", "content": "I have used weighted knn classification using the kknn package with some good success. It allows general Minkowski distance and various weighting kernals. A nice R package for this competition to experiment with.
", "rawMarkdown": "I have used weighted knn classification using the kknn package with some good success. It allows general Minkowski distance and various weighting kernals. A nice R package for this competition to experiment with.", "votes": null }, { "id": "125800", "postDate": "07/02/2016 20:22:45", "content": "Thanks for the responses.
\n\n@SK ... this was exactly what I needed. Although it is bare bones, it plugged into the functions I'd written for the FNN results. Thanks!
\n\n@Khurram - I wish I had discovered this earlier. This looks like a very robust package, and will probably be my goto for KNN and R in the future. There's a bit to look into here, and I probably won't get to it before contest end. Thanks for sharing.
", "rawMarkdown": "Thanks for the responses.\r\n\r\n@SK ... this was exactly what I needed. Although it is bare bones, it plugged into the functions I'd written for the FNN results. Thanks!\r\n\r\n@Khurram - I wish I had discovered this earlier. This looks like a very robust package, and will probably be my goto for KNN and R in the future. There's a bit to look into here, and I probably won't get to it before contest end. Thanks for sharing.", "votes": null } ], "comments": [ { "id": 125449, "author_name": "optimism", "author_url": "", "post_date": "06/29/2016 15:15:09", "content": "Check out RANN.L1.
\n\nIt doesn't have the features you find in python, but it gets the job done.
", "votes": null, "replies": [] }, { "id": 125469, "author_name": "knadeem", "author_url": "", "post_date": "06/29/2016 18:47:56", "content": "I have used weighted knn classification using the kknn package with some good success. It allows general Minkowski distance and various weighting kernals. A nice R package for this competition to experiment with.
", "votes": null, "replies": [] }, { "id": 125800, "author_name": "datadave", "author_url": "", "post_date": "07/02/2016 20:22:45", "content": "Thanks for the responses.
\n\n@SK ... this was exactly what I needed. Although it is bare bones, it plugged into the functions I'd written for the FNN results. Thanks!
\n\n@Khurram - I wish I had discovered this earlier. This looks like a very robust package, and will probably be my goto for KNN and R in the future. There's a bit to look into here, and I probably won't get to it before contest end. Thanks for sharing.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "125400": "I've been very intrigued by how well KNN has done in this competition. It's the first time I've seen that algorithm dominate, at least in the Kaggle scripts. I've been learning a lot, but haven't found R KNN packages that rival what is coming out of python sklearn. Are there any good KNN packages that are efficient and handle Manhattan distance?\r\n\r\nI've tried:\r\n\r\n - FNN::knn (Euclidiean only) \r\n - knnGarden::knnVCN (slow)\r\n - scrime::gknn (can't handle more than 10 classes)", "125449": "Check out [RANN.L1][1].\r\n\r\nIt doesn't have the features you find in python, but it gets the job done.\r\n\r\n\r\n [1]: https://cran.r-project.org/web/packages/RANN.L1/RANN.L1.pdf", "125469": "I have used weighted knn classification using the kknn package with some good success. It allows general Minkowski distance and various weighting kernals. A nice R package for this competition to experiment with.", "125800": "Thanks for the responses.\r\n\r\n@SK ... this was exactly what I needed. Although it is bare bones, it plugged into the functions I'd written for the FNN results. Thanks!\r\n\r\n@Khurram - I wish I had discovered this earlier. This looks like a very robust package, and will probably be my goto for KNN and R in the future. There's a bit to look into here, and I probably won't get to it before contest end. Thanks for sharing." }, "source": "meta" }