I am currently getting 0.512 without an ensemble model. How much were you able to push the score on public leaderboard without an ensemble?
", "messages": [ { "id": "265271", "postDate": "01/05/2018 02:41:21", "content": "I am currently getting 0.512 without an ensemble model. How much were you able to push the score on public leaderboard without an ensemble?
", "rawMarkdown": "I am currently getting 0.512 without an ensemble model. How much were you able to push the score on public leaderboard without an ensemble?", "votes": null }, { "id": "265277", "postDate": "01/05/2018 03:15:05", "content": "You can definitely achieve 0.508 public score with a single model. With some simple bagging the score could go below that.
", "rawMarkdown": "You can definitely achieve 0.508 public score with a single model. With some simple bagging the score could go below that.", "votes": null }, { "id": "265289", "postDate": "01/05/2018 04:08:05", "content": "Thanks Ceshine! That helps :)
", "rawMarkdown": "Thanks Ceshine! That helps :)", "votes": null }, { "id": "265290", "postDate": "01/05/2018 04:09:19", "content": "thanks!
", "rawMarkdown": "thanks!", "votes": null }, { "id": "265366", "postDate": "01/05/2018 11:15:01", "content": "What about model? The best of mine is lgb
", "rawMarkdown": "What about model? The best of mine is lgb", "votes": null }, { "id": "265617", "postDate": "01/06/2018 04:00:42", "content": "Yeah.. same here! Best so far is LGBM
", "rawMarkdown": "Yeah.. same here! Best so far is LGBM", "votes": null }, { "id": "265745", "postDate": "01/06/2018 14:57:51", "content": "Wow! One model 0.508. Waiting for your FE skill post or param-tune skill post after the final deadline.
", "rawMarkdown": "Wow! One model 0.508. Waiting for your FE skill post or param-tune skill post after the final deadline.", "votes": null }, { "id": "265939", "postDate": "01/07/2018 06:14:16", "content": "My LGB gives me 0.505, while my NN is 0.506. (PS:due to some reason, I have no spare time for the rest following days, so I will not continue, good luck to everyone).
", "rawMarkdown": "My LGB gives me 0.505, while my NN is 0.506. (PS:due to some reason, I have no spare time for the rest following days, so I will not continue, good luck to everyone).", "votes": null }, { "id": "265987", "postDate": "01/07/2018 10:59:13", "content": "good job.Did you use categorial features?
", "rawMarkdown": "good job.Did you use categorial features?", "votes": null }, { "id": "266015", "postDate": "01/07/2018 13:29:32", "content": "Of course! These features are quite helpful for the model. For LGB and NN, you should employ different pre-processing approach for these features. You can refer to our Top 2 solution for Porto here. \n https://github.com/kelelexu/kaggle-porto-seguro
", "rawMarkdown": "Of course! These features are quite helpful for the model. For LGB and NN, you should employ different pre-processing approach for these features. You can refer to our Top 2 solution for Porto here. \n https://github.com/kelelexu/kaggle-porto-seguro", "votes": null }, { "id": "266287", "postDate": "01/08/2018 11:27:07", "content": "local cv setting matters also. I moved my NN from 0.511 to 0.508 just by changing local cv setting.
", "rawMarkdown": "local cv setting matters also. I moved my NN from 0.511 to 0.508 just by changing local cv setting.", "votes": null }, { "id": "266504", "postDate": "01/09/2018 01:08:33", "content": "so, for this kind of forecasting problem, the cv day/date matters ? pick a date more similar like 8.16, so we can get better forecast ?
", "rawMarkdown": "so, for this kind of forecasting problem, the cv day/date matters ? pick a date more similar like 8.16, so we can get better forecast ?", "votes": null }, { "id": "266564", "postDate": "01/09/2018 04:28:12", "content": "In any competition local CV setting is key. Deciding on which data you perform your local validation, and on which data you build the models used for submission creation is key. There is no general rule to find the best one. Ceshine Lee has shared a kernel with a quite interesting Cv setting, but there could be others.
", "rawMarkdown": "In any competition local CV setting is key. Deciding on which data you perform your local validation, and on which data you build the models used for submission creation is key. There is no general rule to find the best one. Ceshine Lee has shared a [kernel with a quite interesting Cv setting][1], but there could be others.\n\n\n [1]: https://www.kaggle.com/ceshine/lgbm-starter", "votes": null }, { "id": "266692", "postDate": "01/09/2018 12:49:23", "content": "Thank you!I got better score by categorial features.
", "rawMarkdown": "Thank you!I got better score by categorial features.", "votes": null }, { "id": "266947", "postDate": "01/10/2018 07:14:22", "content": "LGBM works best for me too!
", "rawMarkdown": "LGBM works best for me too!", "votes": null } ], "comments": [ { "id": 265277, "author_name": "ceshine", "author_url": "", "post_date": "01/05/2018 03:15:05", "content": "You can definitely achieve 0.508 public score with a single model. With some simple bagging the score could go below that.
", "votes": null, "replies": [ { "id": 265289, "author_name": "kiranganji99", "author_url": "", "post_date": "01/05/2018 04:08:05", "content": "Thanks Ceshine! That helps :)
", "votes": null, "replies": [] }, { "id": 265290, "author_name": "", "author_url": "", "post_date": "01/05/2018 04:09:19", "content": "thanks!
", "votes": null, "replies": [] }, { "id": 265745, "author_name": "yyqing", "author_url": "", "post_date": "01/06/2018 14:57:51", "content": "Wow! One model 0.508. Waiting for your FE skill post or param-tune skill post after the final deadline.
", "votes": null, "replies": [] }, { "id": 266287, "author_name": "cpmpml", "author_url": "", "post_date": "01/08/2018 11:27:07", "content": "local cv setting matters also. I moved my NN from 0.511 to 0.508 just by changing local cv setting.
", "votes": null, "replies": [] }, { "id": 266504, "author_name": "yyqing", "author_url": "", "post_date": "01/09/2018 01:08:33", "content": "so, for this kind of forecasting problem, the cv day/date matters ? pick a date more similar like 8.16, so we can get better forecast ?
", "votes": null, "replies": [] }, { "id": 266564, "author_name": "cpmpml", "author_url": "", "post_date": "01/09/2018 04:28:12", "content": "In any competition local CV setting is key. Deciding on which data you perform your local validation, and on which data you build the models used for submission creation is key. There is no general rule to find the best one. Ceshine Lee has shared a kernel with a quite interesting Cv setting, but there could be others.
", "votes": null, "replies": [] } ] }, { "id": 265366, "author_name": "kesjien", "author_url": "", "post_date": "01/05/2018 11:15:01", "content": "What about model? The best of mine is lgb
", "votes": null, "replies": [ { "id": 265617, "author_name": "kiranganji99", "author_url": "", "post_date": "01/06/2018 04:00:42", "content": "Yeah.. same here! Best so far is LGBM
", "votes": null, "replies": [] }, { "id": 266947, "author_name": "minhao", "author_url": "", "post_date": "01/10/2018 07:14:22", "content": "LGBM works best for me too!
", "votes": null, "replies": [] } ] }, { "id": 265939, "author_name": "kelexu", "author_url": "", "post_date": "01/07/2018 06:14:16", "content": "My LGB gives me 0.505, while my NN is 0.506. (PS:due to some reason, I have no spare time for the rest following days, so I will not continue, good luck to everyone).
", "votes": null, "replies": [ { "id": 265987, "author_name": "senkin13", "author_url": "", "post_date": "01/07/2018 10:59:13", "content": "good job.Did you use categorial features?
", "votes": null, "replies": [] }, { "id": 266015, "author_name": "kelexu", "author_url": "", "post_date": "01/07/2018 13:29:32", "content": "Of course! These features are quite helpful for the model. For LGB and NN, you should employ different pre-processing approach for these features. You can refer to our Top 2 solution for Porto here. \n https://github.com/kelelexu/kaggle-porto-seguro
", "votes": null, "replies": [] }, { "id": 266692, "author_name": "senkin13", "author_url": "", "post_date": "01/09/2018 12:49:23", "content": "Thank you!I got better score by categorial features.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "265271": "I am currently getting 0.512 without an ensemble model. How much were you able to push the score on public leaderboard without an ensemble?", "265277": "You can definitely achieve 0.508 public score with a single model. With some simple bagging the score could go below that.", "265289": "Thanks Ceshine! That helps :)", "265290": "thanks!", "265366": "What about model? The best of mine is lgb", "265617": "Yeah.. same here! Best so far is LGBM", "265745": "Wow! One model 0.508. Waiting for your FE skill post or param-tune skill post after the final deadline.", "265939": "My LGB gives me 0.505, while my NN is 0.506. (PS:due to some reason, I have no spare time for the rest following days, so I will not continue, good luck to everyone).", "265987": "good job.Did you use categorial features?", "266015": "Of course! These features are quite helpful for the model. For LGB and NN, you should employ different pre-processing approach for these features. You can refer to our Top 2 solution for Porto here. \n https://github.com/kelelexu/kaggle-porto-seguro", "266287": "local cv setting matters also. I moved my NN from 0.511 to 0.508 just by changing local cv setting.", "266504": "so, for this kind of forecasting problem, the cv day/date matters ? pick a date more similar like 8.16, so we can get better forecast ?", "266564": "In any competition local CV setting is key. Deciding on which data you perform your local validation, and on which data you build the models used for submission creation is key. There is no general rule to find the best one. Ceshine Lee has shared a [kernel with a quite interesting Cv setting][1], but there could be others.\n\n\n [1]: https://www.kaggle.com/ceshine/lgbm-starter", "266692": "Thank you!I got better score by categorial features.", "266947": "LGBM works best for me too!" }, "source": "meta" }