Hi All,
\n\nCould you please post your best score of single model !
\n\nRegards,\nJeeban
", "messages": [ { "id": "92421", "postDate": "09/14/2015 06:09:40", "content": "Hi All,
\n\nCould you please post your best score of single model !
\n\nRegards,\nJeeban
", "rawMarkdown": "Hi All,\r\n\r\nCould you please post your best score of single model !\r\n\r\nRegards,\r\nJeeban", "votes": null }, { "id": "92426", "postDate": "09/14/2015 06:15:12", "content": "I got 0.980217 as LB score with RF model.
", "rawMarkdown": "I got **0.980217** as LB score with RF model.", "votes": null }, { "id": "92645", "postDate": "09/15/2015 02:59:45", "content": "For me it's 0.981396 using a RF model, but it's still untuned and only little feature engineering. So I would think there is much room for improvement.
", "rawMarkdown": "For me it's **0.981396** using a RF model, but it's still untuned and only little feature engineering. So I would think there is much room for improvement.", "votes": null }, { "id": "93721", "postDate": "09/30/2015 17:41:10", "content": "0.982832 LB score without cv.
", "rawMarkdown": "**0.982832** LB score without cv.", "votes": null }, { "id": "93722", "postDate": "09/30/2015 18:19:57", "content": "0.983176 is now my best single model LB score (with similar cv value). But I can't get beyond that, even ensembling is pulling me down. Any advice, since your doing quite good on the Leaderboard?
\n\nThanks
", "rawMarkdown": "**0.983176** is now my best single model LB score (with similar cv value). But I can't get beyond that, even ensembling is pulling me down. Any advice, since your doing quite good on the Leaderboard?\r\n\r\nThanks", "votes": null }, { "id": "93742", "postDate": "09/30/2015 21:14:58", "content": "Actually I struggled with an other problem, namely my cv scores were on the sky, but my LB was deep under the ground, thus this is my first real own achievement after a long time "expedition" to solve this problem, but I'm quite happy to managed to figure it out. Now I have 12 days to build an ensemble to climb upper. \nI propose you to combine your single model with one of the "scripted" ensembles, I guess you will be able to improve your score. Let's talk few days later further discussion.\nGood luck!
", "rawMarkdown": "Actually I struggled with an other problem, namely my cv scores were on the sky, but my LB was deep under the ground, thus this is my first real own achievement after a long time \"expedition\" to solve this problem, but I'm quite happy to managed to figure it out. Now I have 12 days to build an ensemble to climb upper. \r\nI propose you to combine your single model with one of the \"scripted\" ensembles, I guess you will be able to improve your score. Let's talk few days later further discussion.\r\nGood luck!", "votes": null }, { "id": "95559", "postDate": "10/09/2015 05:56:51", "content": "I've got 0.983355 on LB with a single RF with feature engineering (stacking). The CV score is very similar: 0.983672. Applying XGBoost on the same dataset gives me CV score 0.985007, but it does not pass the agreement test :(
", "rawMarkdown": "I've got 0.983355 on LB with a single RF with feature engineering (stacking). The CV score is very similar: 0.983672. Applying XGBoost on the same dataset gives me CV score 0.985007, but it does not pass the agreement test :(", "votes": null }, { "id": "95901", "postDate": "10/12/2015 22:04:13", "content": "0.991990 on LB with a single UGBC model, no tricks, just feature engineering, CV score: 0.993.
", "rawMarkdown": "**0.991990** on LB with a single UGBC model, no tricks, just feature engineering, CV score: 0.993.", "votes": null }, { "id": "95929", "postDate": "10/13/2015 00:37:46", "content": "I got CV score 0.993884 with a single XGBoost. Ensembling it with RF and UGBC gave me CV score 0.994635 . This model didn't pass the correlation test by a narrow margin, but it did when I ensembled it with the modified version of the public script (LB score 0.987529). My final private LB score is 0.994894.
", "rawMarkdown": "I got CV score 0.993884 with a single XGBoost. Ensembling it with RF and UGBC gave me CV score 0.994635 . This model didn't pass the correlation test by a narrow margin, but it did when I ensembled it with the modified version of the public script (LB score 0.987529). My final private LB score is 0.994894.", "votes": null }, { "id": "95931", "postDate": "10/13/2015 00:47:28", "content": "My score with a single model, only with a single validation, not used the entire training (set because of lack of time), without CV 0.9935 and was fully aligned with LB. I have had no time. I have had no problem with the KS, but the cvm and time. :)
", "rawMarkdown": "My score with a single model, only with a single validation, not used the entire training (set because of lack of time), **without CV** **0.9935** and was fully aligned with LB. I have had no time. I have had no problem with the KS, but the cvm and time. :)", "votes": null } ], "comments": [ { "id": 92426, "author_name": "jeeban", "author_url": "", "post_date": "09/14/2015 06:15:12", "content": "I got 0.980217 as LB score with RF model.
", "votes": null, "replies": [] }, { "id": 92645, "author_name": "paso84", "author_url": "", "post_date": "09/15/2015 02:59:45", "content": "For me it's 0.981396 using a RF model, but it's still untuned and only little feature engineering. So I would think there is much room for improvement.
", "votes": null, "replies": [] }, { "id": 93721, "author_name": "", "author_url": "", "post_date": "09/30/2015 17:41:10", "content": "0.982832 LB score without cv.
", "votes": null, "replies": [] }, { "id": 93722, "author_name": "paso84", "author_url": "", "post_date": "09/30/2015 18:19:57", "content": "0.983176 is now my best single model LB score (with similar cv value). But I can't get beyond that, even ensembling is pulling me down. Any advice, since your doing quite good on the Leaderboard?
\n\nThanks
", "votes": null, "replies": [] }, { "id": 93742, "author_name": "", "author_url": "", "post_date": "09/30/2015 21:14:58", "content": "Actually I struggled with an other problem, namely my cv scores were on the sky, but my LB was deep under the ground, thus this is my first real own achievement after a long time "expedition" to solve this problem, but I'm quite happy to managed to figure it out. Now I have 12 days to build an ensemble to climb upper. \nI propose you to combine your single model with one of the "scripted" ensembles, I guess you will be able to improve your score. Let's talk few days later further discussion.\nGood luck!
", "votes": null, "replies": [] }, { "id": 95559, "author_name": "anton79", "author_url": "", "post_date": "10/09/2015 05:56:51", "content": "I've got 0.983355 on LB with a single RF with feature engineering (stacking). The CV score is very similar: 0.983672. Applying XGBoost on the same dataset gives me CV score 0.985007, but it does not pass the agreement test :(
", "votes": null, "replies": [] }, { "id": 95901, "author_name": "sionek", "author_url": "", "post_date": "10/12/2015 22:04:13", "content": "0.991990 on LB with a single UGBC model, no tricks, just feature engineering, CV score: 0.993.
", "votes": null, "replies": [] }, { "id": 95929, "author_name": "anton79", "author_url": "", "post_date": "10/13/2015 00:37:46", "content": "I got CV score 0.993884 with a single XGBoost. Ensembling it with RF and UGBC gave me CV score 0.994635 . This model didn't pass the correlation test by a narrow margin, but it did when I ensembled it with the modified version of the public script (LB score 0.987529). My final private LB score is 0.994894.
", "votes": null, "replies": [] }, { "id": 95931, "author_name": "", "author_url": "", "post_date": "10/13/2015 00:47:28", "content": "My score with a single model, only with a single validation, not used the entire training (set because of lack of time), without CV 0.9935 and was fully aligned with LB. I have had no time. I have had no problem with the KS, but the cvm and time. :)
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "92421": "Hi All,\r\n\r\nCould you please post your best score of single model !\r\n\r\nRegards,\r\nJeeban", "92426": "I got **0.980217** as LB score with RF model.", "92645": "For me it's **0.981396** using a RF model, but it's still untuned and only little feature engineering. So I would think there is much room for improvement.", "93721": "**0.982832** LB score without cv.", "93722": "**0.983176** is now my best single model LB score (with similar cv value). But I can't get beyond that, even ensembling is pulling me down. Any advice, since your doing quite good on the Leaderboard?\r\n\r\nThanks", "93742": "Actually I struggled with an other problem, namely my cv scores were on the sky, but my LB was deep under the ground, thus this is my first real own achievement after a long time \"expedition\" to solve this problem, but I'm quite happy to managed to figure it out. Now I have 12 days to build an ensemble to climb upper. \r\nI propose you to combine your single model with one of the \"scripted\" ensembles, I guess you will be able to improve your score. Let's talk few days later further discussion.\r\nGood luck!", "95559": "I've got 0.983355 on LB with a single RF with feature engineering (stacking). The CV score is very similar: 0.983672. Applying XGBoost on the same dataset gives me CV score 0.985007, but it does not pass the agreement test :(", "95901": "**0.991990** on LB with a single UGBC model, no tricks, just feature engineering, CV score: 0.993.", "95929": "I got CV score 0.993884 with a single XGBoost. Ensembling it with RF and UGBC gave me CV score 0.994635 . This model didn't pass the correlation test by a narrow margin, but it did when I ensembled it with the modified version of the public script (LB score 0.987529). My final private LB score is 0.994894.", "95931": "My score with a single model, only with a single validation, not used the entire training (set because of lack of time), **without CV** **0.9935** and was fully aligned with LB. I have had no time. I have had no problem with the KS, but the cvm and time. :)" }, "source": "meta" }