{ "id": 97705, "title": "Tip for tree boosting users [LB -1.6]", "url": "/competitions/champs-scalar-coupling/discussion/97705", "author_name": "Yi23333", "post_date": "2019-06-28T14:17:58.848000", "votes": 89, "comment_count": 21, "views": 0, "content": "

Hello, I just want to give a tip for those using algorithms like lightgbm or xgboost. As we know that the variance of different types are very different. For example, j coupling values of 1JHC has very large variance while 3JHH has relatively small variance. Therefore, the parameters for tree boosting algorithms such as the \"depth of trees\" need to be adjusted accordingly. This gives me a large boost.

", "messages": [ { "id": 563634, "postDate": "2019-06-28T14:17:58.850Z", "content": "

Hello, I just want to give a tip for those using algorithms like lightgbm or xgboost. As we know that the variance of different types are very different. For example, j coupling values of 1JHC has very large variance while 3JHH has relatively small variance. Therefore, the parameters for tree boosting algorithms such as the \"depth of trees\" need to be adjusted accordingly. This gives me a large boost.

", "rawMarkdown": "Hello, I just want to give a tip for those using algorithms like lightgbm or xgboost. As we know that the variance of different types are very different. For example, j coupling values of 1JHC has very large variance while 3JHH has relatively small variance. Therefore, the parameters for tree boosting algorithms such as the \"depth of trees\" need to be adjusted accordingly. This gives me a large boost.", "votes": 89 }, { "id": 563703, "postDate": "2019-06-28T15:34:56.250Z", "content": "

Good to know - thanks for the information!

", "rawMarkdown": "Good to know - thanks for the information!", "votes": 1 }, { "id": 595274, "postDate": "2019-08-09T03:23:14.137Z", "content": "

That's crazy. Really appreciate your help, thanks.

", "rawMarkdown": "That's crazy. Really appreciate your help, thanks. ", "votes": 2 }, { "id": 597485, "postDate": "2019-08-12T12:06:49.963Z", "content": "

What technique are you using for parameter tunning?

", "rawMarkdown": "What technique are you using for parameter tunning?\n" }, { "id": 596609, "postDate": "2019-08-11T01:17:02.867Z", "content": "

This is very useful, thank you for the tip!

", "rawMarkdown": "This is very useful, thank you for the tip!" }, { "id": 568227, "postDate": "2019-07-04T15:06:09.697Z", "content": "

Thanks for sharing , could you please also tell us what is 3JHH and 1JHC.

", "rawMarkdown": "Thanks for sharing , could you please also tell us what is 3JHH and 1JHC.", "replies": [ { "id": 568340, "postDate": "2019-07-04T18:28:56.183Z", "content": "

These are types of coupling constants we are trying to predict in this competition..

", "rawMarkdown": "These are types of coupling constants we are trying to predict in this competition.." } ] }, { "id": 563980, "postDate": "2019-06-28T21:01:17.947Z", "content": "

Similarly question to Kyle, may I know your oof MAE for 1JHC?

", "rawMarkdown": "Similarly question to Kyle, may I know your oof MAE for 1JHC?" }, { "id": 563743, "postDate": "2019-06-28T16:18:24.220Z", "content": "

How much of a score boost did it give you. And, should 1JHC or 3JHC have the larger tree depth?

", "rawMarkdown": "How much of a score boost did it give you. And, should 1JHC or 3JHC have the larger tree depth?", "replies": [ { "id": 564002, "postDate": "2019-06-28T21:36:49.050Z", "content": "

My experiment on 1JHN gives me a logmae boost of 0.3 for that type! I think 1JHC should have a relatively smaller depth than 3JHC because of the more severe overfitting problem.

", "rawMarkdown": "My experiment on 1JHN gives me a logmae boost of 0.3 for that type! I think 1JHC should have a relatively smaller depth than 3JHC because of the more severe overfitting problem.", "votes": 4 } ] }, { "id": 564250, "postDate": "2019-06-29T07:44:52.530Z", "rawMarkdown": "", "votes": -2, "isDeleted": true }, { "id": 568208, "postDate": "2019-07-04T14:44:53.883Z", "content": "

Thanks for you tip

", "rawMarkdown": "Thanks for you tip", "votes": 1 }, { "id": 567384, "postDate": "2019-07-03T12:32:52.890Z", "content": "

omg thanks!

", "rawMarkdown": "omg thanks!", "votes": 1 }, { "id": 566586, "postDate": "2019-07-02T10:52:39.457Z", "content": "

Thanks =) great tip

", "rawMarkdown": "Thanks =) great tip", "votes": 1 }, { "id": 566501, "postDate": "2019-07-02T08:45:52.690Z", "content": "

Thanks！

", "rawMarkdown": "Thanks！", "votes": 1 }, { "id": 565957, "postDate": "2019-07-01T15:55:51.877Z", "content": "

What a TIP! Thanks bro!

", "rawMarkdown": "What a TIP! Thanks bro!", "votes": 1 }, { "id": 565533, "postDate": "2019-07-01T05:03:49.033Z", "content": "

Thanks for sharing the information!

", "rawMarkdown": "Thanks for sharing the information!", "votes": 1 }, { "id": 565532, "postDate": "2019-07-01T05:01:05.917Z", "content": "

thanks!

", "rawMarkdown": "thanks!", "votes": 1 }, { "id": 564774, "postDate": "2019-06-30T00:59:35.790Z", "content": "

Thanks for the good information!

", "rawMarkdown": "Thanks for the good information!", "votes": 1 }, { "id": 564497, "postDate": "2019-06-29T14:00:12.550Z", "content": "

thanks for the information.

", "rawMarkdown": "thanks for the information.", "votes": 1 }, { "id": 610061, "postDate": "2019-08-28T11:25:55.320Z", "content": "

thanks!

", "rawMarkdown": "thanks!" }, { "id": 568811, "postDate": "2019-07-05T13:22:23.973Z", "content": "

Thank s for sharing!

", "rawMarkdown": "Thank s for sharing!" } ], "comments": [ { "id": 563703, "author_name": "Scirpus", "author_url": "", "post_date": "2019-06-28T15:34:56.250000", "content": "

Good to know - thanks for the information!

", "votes": 1, "replies": [] }, { "id": 595274, "author_name": "torch", "author_url": "", "post_date": "2019-08-09T03:23:14.137000", "content": "

That's crazy. Really appreciate your help, thanks.

", "votes": 2, "replies": [] }, { "id": 597485, "author_name": "Jaime Badiola", "author_url": "", "post_date": "2019-08-12T12:06:49.963000", "content": "

What technique are you using for parameter tunning?

", "votes": 0, "replies": [] }, { "id": 596609, "author_name": "Mike G", "author_url": "", "post_date": "2019-08-11T01:17:02.867000", "content": "

This is very useful, thank you for the tip!

", "votes": 0, "replies": [] }, { "id": 568227, "author_name": "gunish jha", "author_url": "", "post_date": "2019-07-04T15:06:09.697000", "content": "

Thanks for sharing , could you please also tell us what is 3JHH and 1JHC.

", "votes": 0, "replies": [ { "id": 568340, "author_name": "somnambWl", "author_url": "", "post_date": "2019-07-04T18:28:56.183000", "content": "

These are types of coupling constants we are trying to predict in this competition..

", "votes": 0, "replies": [] } ] }, { "id": 563980, "author_name": "Shuhao Cao", "author_url": "", "post_date": "2019-06-28T21:01:17.947000", "content": "

Similarly question to Kyle, may I know your oof MAE for 1JHC?

", "votes": 0, "replies": [] }, { "id": 563743, "author_name": "Kyle Peters", "author_url": "", "post_date": "2019-06-28T16:18:24.220000", "content": "

How much of a score boost did it give you. And, should 1JHC or 3JHC have the larger tree depth?

", "votes": 0, "replies": [ { "id": 564002, "author_name": "Yi23333", "author_url": "", "post_date": "2019-06-28T21:36:49.050000", "content": "

My experiment on 1JHN gives me a logmae boost of 0.3 for that type! I think 1JHC should have a relatively smaller depth than 3JHC because of the more severe overfitting problem.

", "votes": 4, "replies": [] } ] }, { "id": 564250, "author_name": "", "author_url": "", "post_date": "2019-06-29T07:44:52.530000", "content": "", "votes": -2, "replies": [] }, { "id": 568208, "author_name": "Tony Teng", "author_url": "", "post_date": "2019-07-04T14:44:53.883000", "content": "

Thanks for you tip

", "votes": 1, "replies": [] }, { "id": 567384, "author_name": "Beomki Yeo", "author_url": "", "post_date": "2019-07-03T12:32:52.890000", "content": "

omg thanks!

", "votes": 1, "replies": [] }, { "id": 566586, "author_name": "dexter6855", "author_url": "", "post_date": "2019-07-02T10:52:39.457000", "content": "

Thanks =) great tip

", "votes": 1, "replies": [] }, { "id": 566501, "author_name": "Mike Zhu", "author_url": "", "post_date": "2019-07-02T08:45:52.690000", "content": "

Thanks！

", "votes": 1, "replies": [] }, { "id": 565957, "author_name": "Adriano Avelar", "author_url": "", "post_date": "2019-07-01T15:55:51.877000", "content": "

What a TIP! Thanks bro!

", "votes": 1, "replies": [] }, { "id": 565533, "author_name": "Dondon2019", "author_url": "", "post_date": "2019-07-01T05:03:49.033000", "content": "

Thanks for sharing the information!

", "votes": 1, "replies": [] }, { "id": 565532, "author_name": "Yama", "author_url": "", "post_date": "2019-07-01T05:01:05.917000", "content": "

thanks!

", "votes": 1, "replies": [] }, { "id": 564774, "author_name": "kirikei", "author_url": "", "post_date": "2019-06-30T00:59:35.790000", "content": "

Thanks for the good information!

", "votes": 1, "replies": [] }, { "id": 564497, "author_name": "mochidan", "author_url": "", "post_date": "2019-06-29T14:00:12.550000", "content": "

thanks for the information.

", "votes": 1, "replies": [] }, { "id": 610061, "author_name": "RK", "author_url": "", "post_date": "2019-08-28T11:25:55.320000", "content": "

thanks!

", "votes": 0, "replies": [] }, { "id": 568811, "author_name": "AL", "author_url": "", "post_date": "2019-07-05T13:22:23.973000", "content": "

Thank s for sharing!

", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "563634": "Hello, I just want to give a tip for those using algorithms like lightgbm or xgboost. As we know that the variance of different types are very different. For example, j coupling values of 1JHC has very large variance while 3JHH has relatively small variance. Therefore, the parameters for tree boosting algorithms such as the \"depth of trees\" need to be adjusted accordingly. This gives me a large boost.", "563703": "Good to know - thanks for the information!", "595274": "That's crazy. Really appreciate your help, thanks. ", "597485": "What technique are you using for parameter tunning?\n", "596609": "This is very useful, thank you for the tip!", "568227": "Thanks for sharing , could you please also tell us what is 3JHH and 1JHC.", "563980": "Similarly question to Kyle, may I know your oof MAE for 1JHC?", "563743": "How much of a score boost did it give you. And, should 1JHC or 3JHC have the larger tree depth?", "564250": "", "568208": "Thanks for you tip", "567384": "omg thanks!", "566586": "Thanks =) great tip", "566501": "Thanks！", "565957": "What a TIP! Thanks bro!", "565533": "Thanks for sharing the information!", "565532": "thanks!", "564774": "Thanks for the good information!", "564497": "thanks for the information.", "610061": "thanks!", "568811": "Thank s for sharing!" } }