The last week is coming. Everyone will become crazy and the rank on the leaderboard may change a lot. Good luck everyone!
\n\nHow are you going to spend the final week? I will still trying new models and tuning parameters.
", "messages": [ { "id": "125532", "postDate": "06/30/2016 06:26:42", "content": "The last week is coming. Everyone will become crazy and the rank on the leaderboard may change a lot. Good luck everyone!
\n\nHow are you going to spend the final week? I will still trying new models and tuning parameters.
", "rawMarkdown": "The last week is coming. Everyone will become crazy and the rank on the leaderboard may change a lot. Good luck everyone!\r\n\r\nHow are you going to spend the final week? I will still trying new models and tuning parameters.", "votes": null }, { "id": "125542", "postDate": "06/30/2016 08:02:42", "content": "Due to this being a recruitment challenge, I think the last week will be fairly calm.
\n\nWe will not see the usual large improvements from teams merging their solutions, but definitely some gains from users assembling their previous submissions.
", "rawMarkdown": "Due to this being a recruitment challenge, I think the last week will be fairly calm. \r\n\r\nWe will not see the usual large improvements from teams merging their solutions, but definitely some gains from users assembling their previous submissions.", "votes": null }, { "id": "125558", "postDate": "06/30/2016 11:10:10", "content": "For the final week, low specs PC (6GB RAM, 2 cores) is going to finish its xgb job which is a total 10x24 hours of time, ETA is 7x24 hours, so for the rest 6 days I will miss 2x40 grid cells out of 20x40 ;))
\n\nAnd it's for unknown score, keep calm as fortunately next week is holiday season ;))
\n\nWish you all the best.
", "rawMarkdown": "For the final week, low specs PC (6GB RAM, 2 cores) is going to finish its xgb job which is a total 10x24 hours of time, ETA is 7x24 hours, so for the rest 6 days I will miss 2x40 grid cells out of 20x40 ;)) \r\n\r\nAnd it's for *unknown* score, keep calm as fortunately next week is holiday season ;)) \r\n\r\nWish you all the best.", "votes": null }, { "id": "125580", "postDate": "06/30/2016 14:51:27", "content": "@aldente I admire your patience, it would go on my nerves waiting a whole week for the decisive submission. ;-)
", "rawMarkdown": "aldente I admire your patience, it would go on my nerves waiting a whole week for the decisive submission. ;-)", "votes": null }, { "id": "125599", "postDate": "06/30/2016 19:18:03", "content": "For me, the last few days are about finishing work on another model--and then learning whether that model turns out to be beneficial for the ensemble, or remains a "nice try". Time permitting, I also want to do some fine tuning on existing models.
\n\nGood luck to all!
", "rawMarkdown": "For me, the last few days are about finishing work on another model--and then learning whether that model turns out to be beneficial for the ensemble, or remains a \"nice try\". Time permitting, I also want to do some fine tuning on existing models.\r\n\r\nGood luck to all!", "votes": null }, { "id": "125602", "postDate": "06/30/2016 19:57:07", "content": "@Frederik, exactly, also thinking about: will electricity go out, temperature of CPU cores ;))
\n\n@Andre, great achievement! Hope another chance for teaming up ;) I am still wondering how to ensemble different models (different grid size) efficiently & effectively in this case, considering hundred of classes. After due date hopefully you all will share the method ;)
", "rawMarkdown": "Frederik, exactly, also thinking about: will electricity go out, temperature of CPU cores ;)) \r\n\r\n@Andre, great achievement! Hope another chance for teaming up ;) I am still wondering how to [ensemble][1] different models (different grid size) efficiently & effectively in this case, considering hundred of classes. After due date hopefully you all will share the method ;) \r\n\r\n\r\n [1]: https://www.kaggle.com/c/facebook-v-predicting-check-ins/forums/t/21265/creating-ensemble-of-2-submissions", "votes": null }, { "id": "125880", "postDate": "07/04/2016 00:26:42", "content": "3 days left. Now it needs 0.604 to get into top 10. Everyone works very hard on improving their model!
", "rawMarkdown": "3 days left. Now it needs 0.604 to get into top 10. Everyone works very hard on improving their model!", "votes": null }, { "id": "125970", "postDate": "07/05/2016 05:10:36", "content": "Yes sir! For now, low specs PC has been finishing 569 cells out of 800 (20x40) of XGBoost grid (0.59546 LB). Sorted from smallest [row x class], now each cell needs 40-60 minutes, still 231 cells to go which is impossible ;))
\n\nWondering which is better (time & score), KNN or XGB?
", "rawMarkdown": "Yes sir! For now, low specs PC has been finishing 569 cells out of 800 (20x40) of XGBoost grid (0.59546 LB). Sorted from smallest [row x class], now each cell needs 40-60 minutes, still 231 cells to go which is impossible ;)) \r\n \r\nWondering which is better (time & score), KNN or XGB?", "votes": null }, { "id": "125971", "postDate": "07/05/2016 05:55:50", "content": "@aldente, my XGB model is about 2% better than KNN. But KNN is of course much faster than XGB.
", "rawMarkdown": "aldente, my XGB model is about 2% better than KNN. But KNN is of course much faster than XGB.", "votes": null }, { "id": "125982", "postDate": "07/05/2016 09:09:56", "content": "@Li Li, thanks! My XGB is also better than my KNN, I just wondered if someone could do KNN better than my XGB's LB score ;) Considering your LB score, I suspect your KNN score is better than my XGB.
", "rawMarkdown": "Li Li, thanks! My XGB is also better than my KNN, I just wondered if someone could do KNN better than my XGB's LB score ;) Considering your LB score, I suspect your KNN score is better than my XGB.", "votes": null } ], "comments": [ { "id": 125542, "author_name": "frederik", "author_url": "", "post_date": "06/30/2016 08:02:42", "content": "Due to this being a recruitment challenge, I think the last week will be fairly calm.
\n\nWe will not see the usual large improvements from teams merging their solutions, but definitely some gains from users assembling their previous submissions.
", "votes": null, "replies": [] }, { "id": 125558, "author_name": "aldente", "author_url": "", "post_date": "06/30/2016 11:10:10", "content": "For the final week, low specs PC (6GB RAM, 2 cores) is going to finish its xgb job which is a total 10x24 hours of time, ETA is 7x24 hours, so for the rest 6 days I will miss 2x40 grid cells out of 20x40 ;))
\n\nAnd it's for unknown score, keep calm as fortunately next week is holiday season ;))
\n\nWish you all the best.
", "votes": null, "replies": [] }, { "id": 125580, "author_name": "frederik", "author_url": "", "post_date": "06/30/2016 14:51:27", "content": "@aldente I admire your patience, it would go on my nerves waiting a whole week for the decisive submission. ;-)
", "votes": null, "replies": [] }, { "id": 125599, "author_name": "andrenaef", "author_url": "", "post_date": "06/30/2016 19:18:03", "content": "For me, the last few days are about finishing work on another model--and then learning whether that model turns out to be beneficial for the ensemble, or remains a "nice try". Time permitting, I also want to do some fine tuning on existing models.
\n\nGood luck to all!
", "votes": null, "replies": [] }, { "id": 125602, "author_name": "aldente", "author_url": "", "post_date": "06/30/2016 19:57:07", "content": "@Frederik, exactly, also thinking about: will electricity go out, temperature of CPU cores ;))
\n\n@Andre, great achievement! Hope another chance for teaming up ;) I am still wondering how to ensemble different models (different grid size) efficiently & effectively in this case, considering hundred of classes. After due date hopefully you all will share the method ;)
", "votes": null, "replies": [] }, { "id": 125880, "author_name": "aikinogard", "author_url": "", "post_date": "07/04/2016 00:26:42", "content": "3 days left. Now it needs 0.604 to get into top 10. Everyone works very hard on improving their model!
", "votes": null, "replies": [] }, { "id": 125970, "author_name": "aldente", "author_url": "", "post_date": "07/05/2016 05:10:36", "content": "Yes sir! For now, low specs PC has been finishing 569 cells out of 800 (20x40) of XGBoost grid (0.59546 LB). Sorted from smallest [row x class], now each cell needs 40-60 minutes, still 231 cells to go which is impossible ;))
\n\nWondering which is better (time & score), KNN or XGB?
", "votes": null, "replies": [] }, { "id": 125971, "author_name": "aikinogard", "author_url": "", "post_date": "07/05/2016 05:55:50", "content": "@aldente, my XGB model is about 2% better than KNN. But KNN is of course much faster than XGB.
", "votes": null, "replies": [] }, { "id": 125982, "author_name": "aldente", "author_url": "", "post_date": "07/05/2016 09:09:56", "content": "@Li Li, thanks! My XGB is also better than my KNN, I just wondered if someone could do KNN better than my XGB's LB score ;) Considering your LB score, I suspect your KNN score is better than my XGB.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "125532": "The last week is coming. Everyone will become crazy and the rank on the leaderboard may change a lot. Good luck everyone!\r\n\r\nHow are you going to spend the final week? I will still trying new models and tuning parameters.", "125542": "Due to this being a recruitment challenge, I think the last week will be fairly calm. \r\n\r\nWe will not see the usual large improvements from teams merging their solutions, but definitely some gains from users assembling their previous submissions.", "125558": "For the final week, low specs PC (6GB RAM, 2 cores) is going to finish its xgb job which is a total 10x24 hours of time, ETA is 7x24 hours, so for the rest 6 days I will miss 2x40 grid cells out of 20x40 ;)) \r\n\r\nAnd it's for *unknown* score, keep calm as fortunately next week is holiday season ;)) \r\n\r\nWish you all the best.", "125580": "aldente I admire your patience, it would go on my nerves waiting a whole week for the decisive submission. ;-)", "125599": "For me, the last few days are about finishing work on another model--and then learning whether that model turns out to be beneficial for the ensemble, or remains a \"nice try\". Time permitting, I also want to do some fine tuning on existing models.\r\n\r\nGood luck to all!", "125602": "Frederik, exactly, also thinking about: will electricity go out, temperature of CPU cores ;)) \r\n\r\n@Andre, great achievement! Hope another chance for teaming up ;) I am still wondering how to [ensemble][1] different models (different grid size) efficiently & effectively in this case, considering hundred of classes. After due date hopefully you all will share the method ;) \r\n\r\n\r\n [1]: https://www.kaggle.com/c/facebook-v-predicting-check-ins/forums/t/21265/creating-ensemble-of-2-submissions", "125880": "3 days left. Now it needs 0.604 to get into top 10. Everyone works very hard on improving their model!", "125970": "Yes sir! For now, low specs PC has been finishing 569 cells out of 800 (20x40) of XGBoost grid (0.59546 LB). Sorted from smallest [row x class], now each cell needs 40-60 minutes, still 231 cells to go which is impossible ;)) \r\n \r\nWondering which is better (time & score), KNN or XGB?", "125971": "aldente, my XGB model is about 2% better than KNN. But KNN is of course much faster than XGB.", "125982": "Li Li, thanks! My XGB is also better than my KNN, I just wondered if someone could do KNN better than my XGB's LB score ;) Considering your LB score, I suspect your KNN score is better than my XGB." }, "source": "meta" }