First of all, thanks for holding this memorable and intriguing competition!\nAs others mentioned, there is no leak here and data is clean. This is marvelous thing isn't it?
\n\nI thought I got 1st place, but congrats @weiwei!
\n\nAlso, @sjv your approach is splendid than I expected! Thank you for motivating me on LB. It was really amazing time.
\n\nAnyway, I really enjoyed this competition. Again, thank you for holding this competition, and all kagglers!
\n\nHere is my git.
", "comments": [ { "id": 213934, "author_name": "ONODERA", "author_url": "", "post_date": "2017-08-15T14:52:59.410000", "content": "Here is my awful submission selection.
", "votes": 5, "replies": [ { "id": 213962, "author_name": "David K", "author_url": "", "post_date": "2017-08-15T15:21:31.660000", "content": "I wouldn't call that awful - it didn't cost you any rank. Rather a bit unlucky ;)\nAnd thank you for participating in the discussions, I'm sure you helped a lot of people (including myself).
", "votes": 3, "replies": [] }, { "id": 214312, "author_name": "Robert Goss ", "author_url": "", "post_date": "2017-08-16T13:59:41.257000", "content": "Congrats ONODERA! I really like your method of departmentalizing feature selection, may take that approach in the future.
\n\nDo you have any tips for new kagglers on how you selected your final submissions? Sadly my selections cost me nearly 60 places ;(.
", "votes": 0, "replies": [] }, { "id": 214325, "author_name": "ONODERA", "author_url": "", "post_date": "2017-08-16T14:44:41.857000", "content": "I'm not familiar with submission selecting so much. But if I select submission seriously, I'll evaluate it by CV.
\n\nEdit: Also it depends on structure of train test split.
", "votes": 0, "replies": [] } ] }, { "id": 214050, "author_name": "Fred Navruzov", "author_url": "", "post_date": "2017-08-15T20:33:29.923000", "content": "Very nice and steady top-3 performance for such a long time!\n
I also thought you absolutely won this competition until the very last moments.
Well done, it was nice to compete with you and to understand your hints!\n
Obviously, you're a \"top gun\"!\n
Great thanks and good luck!
What is your setup that has 300 GB RAM?
", "votes": 4, "replies": [] }, { "id": 213946, "author_name": "RobertoR", "author_url": "", "post_date": "2017-08-15T15:01:21.197000", "content": "Hey Onodera, thank you very much for your participation in the forum. It was really valuable. Congratulation for your second place.
", "votes": 1, "replies": [] }, { "id": 213940, "author_name": "Bradley Mitchell", "author_url": "", "post_date": "2017-08-15T14:55:48.657000", "content": "Thank you for posting this, ONODERA. I really like your approach to feature engineering - a relatively small number of high signal features. Congratulations!!!
", "votes": 1, "replies": [] }, { "id": 213939, "author_name": "zarak", "author_url": "", "post_date": "2017-08-15T14:55:44.113000", "content": "Thanks for posting your solution, and congratulations on a solo gold medal!
\n\nIt was fun to watch the intense battle between you and sjv, only for weiwei to come out of nowhere and take the lead at the last minute!
\n\nI was rooting for you because of your helpful posts on the forums, but I'm glad sjv also finished in the money given his really remarkable solution.
", "votes": 2, "replies": [] }, { "id": 420827, "author_name": "LongYin/杰少", "author_url": "", "post_date": "2018-11-14T07:13:55.537000", "content": "Your code is quite nice! I learned a lot from it.
\n\nWould you like to share the feature importance of your XGB model.
", "votes": 0, "replies": [] }, { "id": 419387, "author_name": "Vishal Singh", "author_url": "", "post_date": "2018-11-11T21:25:03.430000", "content": "Hi,\nDid you use last 3 orders of all the users for taining???
\n\nIf we use last 3 order of all the users then the dataset will be huge? where do you run the processes? \nPlease help!
", "votes": 0, "replies": [ { "id": 420049, "author_name": "ONODERA", "author_url": "", "post_date": "2018-11-13T01:30:03.830000", "content": "\n\n\nDid you use last 3 orders of all the users for taining???
\n
Yes, you need 200GB memory or so for running my model. Read README.md.
", "votes": 1, "replies": [] }, { "id": 432384, "author_name": "Eduardo Reis", "author_url": "", "post_date": "2018-12-03T19:42:05.687000", "content": "No way I could run that!
", "votes": 0, "replies": [] }, { "id": 432390, "author_name": "Eduardo Reis", "author_url": "", "post_date": "2018-12-03T19:46:27.250000", "content": "Would you have a youtube video for your ppt? Thanks for sharing the Repo.
", "votes": 0, "replies": [] } ] }, { "id": 341950, "author_name": "yu4u", "author_url": "", "post_date": "2018-06-12T15:07:08.917000", "content": "Tensai!!!
", "votes": 0, "replies": [] }, { "id": 259788, "author_name": "Jingangjing", "author_url": "", "post_date": "2017-12-19T01:45:40", "content": "It is amazing to see using xgboost alone can achieve 2nd place!
", "votes": 0, "replies": [] }, { "id": 214328, "author_name": "Serigne ", "author_url": "", "post_date": "2017-08-16T15:05:33.250000", "content": "Congrats ONODERA for your 2nd place., and thanks for sharing.
\n\nI really appreciated your lively approach and your sense of humour in discussion topics during the competition.
", "votes": 0, "replies": [] }, { "id": 214184, "author_name": "Mint", "author_url": "", "post_date": "2017-08-16T06:45:49.090000", "content": "I noticed you and a few others who've posted solutions broke the day in to segments, aka time zones. In terms of detecting whether a user is making an unusual order was binning the orders like that with hard cutoff better than a weighted moving average? Or was the binning method better because it let you easily create more features using it. I'm wondering since with time zones it seems user who orders at 11.30am and then 12 noon would be in separate zones.
", "votes": 0, "replies": [ { "id": 214200, "author_name": "ONODERA", "author_url": "", "post_date": "2017-08-16T07:53:24.297000", "content": "Good question, I don't think one time zone suffice.\nI should have added another one.
", "votes": 0, "replies": [] } ] }, { "id": 214117, "author_name": "LivingProgram", "author_url": "", "post_date": "2017-08-16T02:12:47.460000", "content": "Thanks for sharing! I was wondering how did you end up optimizing your hyperparameters for your final submission?
", "votes": 0, "replies": [ { "id": 214142, "author_name": "ONODERA", "author_url": "", "post_date": "2017-08-16T03:38:12.770000", "content": "Actually, I didn't think about hyperparameters so much. I only cared about necessary time.
", "votes": 0, "replies": [] } ] }, { "id": 214104, "author_name": "Sungjun, Kim", "author_url": "", "post_date": "2017-08-16T01:27:08.537000", "content": "Thank you for your sharing codes. You are so great!\nOne question, How did you select your 'special' features?
", "votes": 0, "replies": [ { "id": 214140, "author_name": "ONODERA", "author_url": "", "post_date": "2017-08-16T03:32:25.917000", "content": "I used feature importance of xgboost. We can get weight, gain and cover. I normalized those, sum, and then pick up from higher feature.
", "votes": 5, "replies": [] }, { "id": 214196, "author_name": "Sungjun, Kim", "author_url": "", "post_date": "2017-08-16T07:21:47.297000", "content": "Thank you for your advice! :)
", "votes": 0, "replies": [] } ] }, { "id": 214061, "author_name": "amasl", "author_url": "", "post_date": "2017-08-15T21:15:05.277000", "content": "Congrats and many thanks for your contribution!
", "votes": 0, "replies": [] }, { "id": 214052, "author_name": "YaGana Sheriff-Hussaini", "author_url": "", "post_date": "2017-08-15T20:38:50.773000", "content": "Congratulations @Onodera and thanks for posting your solution. Because this is my 1st involvement in this king of market basket analysis problem, I plan on using NN to build a model on this data so that I can explore and learn some more. \nSo I will be checking the top10 solutions including yours for all the different approaches to this problem.
\n\nAnd thank you also for participating in the discussions which made this competition much more interesting.
", "votes": 0, "replies": [] }, { "id": 214004, "author_name": "neverdies", "author_url": "", "post_date": "2017-08-15T17:42:46.847000", "content": "Thanks Onodera! You did an incredible job and shared lots of useful information during the competition. This is my fist competition and I learned a lot from your kagglers! It IS a great community! Congrats!
", "votes": 0, "replies": [] }, { "id": 213980, "author_name": "happycube", "author_url": "", "post_date": "2017-08-15T16:19:12.770000", "content": "Congrats and thanks for posting/sharing!
", "votes": 0, "replies": [] }, { "id": 214119, "author_name": "BigMorning", "author_url": "", "post_date": "2017-08-16T02:23:50.770000", "content": "Congrats! Thanks for sharing.
", "votes": 0, "replies": [] }, { "id": 214087, "author_name": "felii", "author_url": "", "post_date": "2017-08-15T23:24:38.970000", "content": "Congrats! Thanks for sharing.
", "votes": 0, "replies": [] }, { "id": 213972, "author_name": "plantsgo", "author_url": "", "post_date": "2017-08-15T15:58:54.807000", "content": "Congrats and thanks for sharing.
", "votes": 0, "replies": [] }, { "id": 213958, "author_name": "", "author_url": "", "post_date": "2017-08-15T15:15:37.643000", "content": "Thanks. You are still my hero.
", "votes": 0, "replies": [] }, { "id": 213935, "author_name": "beletecheneke", "author_url": "", "post_date": "2017-08-15T14:53:00.107000", "content": "Congrats and thanks for sharing.
", "votes": 0, "replies": [] } ], "comment_count": 33, "messages": [ { "id": 213928, "postDate": "2017-08-15T14:46:13.943Z", "content": "First of all, thanks for holding this memorable and intriguing competition!\nAs others mentioned, there is no leak here and data is clean. This is marvelous thing isn't it?
\n\nI thought I got 1st place, but congrats @weiwei!
\n\nAlso, @sjv your approach is splendid than I expected! Thank you for motivating me on LB. It was really amazing time.
\n\nAnyway, I really enjoyed this competition. Again, thank you for holding this competition, and all kagglers!
\n\nHere is my git.
", "rawMarkdown": "First of all, thanks for holding this memorable and intriguing competition!\nAs others mentioned, there is no leak here and data is clean. This is marvelous thing isn't it?\n\nI thought I got 1st place, but congrats @weiwei!\n\nAlso, @sjv your approach is splendid than I expected! Thank you for motivating me on LB. It was really amazing time.\n\nAnyway, I really enjoyed this competition. Again, thank you for holding this competition, and all kagglers!\n\nHere is my [git][1].\n\n\n [1]: https://github.com/KazukiOnodera/Instacart", "votes": 133 }, { "id": 213934, "postDate": "2017-08-15T14:52:59.410Z", "content": "Here is my awful submission selection.
", "rawMarkdown": "Here is my awful submission selection.", "votes": 5, "replies": [ { "id": 213962, "postDate": "2017-08-15T15:21:31.660Z", "content": "I wouldn't call that awful - it didn't cost you any rank. Rather a bit unlucky ;)\nAnd thank you for participating in the discussions, I'm sure you helped a lot of people (including myself).
", "rawMarkdown": "I wouldn't call that awful - it didn't cost you any rank. Rather a bit unlucky ;)\nAnd thank you for participating in the discussions, I'm sure you helped a lot of people (including myself).", "votes": 3 }, { "id": 214312, "postDate": "2017-08-16T13:59:41.257Z", "content": "Congrats ONODERA! I really like your method of departmentalizing feature selection, may take that approach in the future.
\n\nDo you have any tips for new kagglers on how you selected your final submissions? Sadly my selections cost me nearly 60 places ;(.
", "rawMarkdown": "Congrats ONODERA! I really like your method of departmentalizing feature selection, may take that approach in the future.\n\nDo you have any tips for new kagglers on how you selected your final submissions? Sadly my selections cost me nearly 60 places ;(. " }, { "id": 214325, "postDate": "2017-08-16T14:44:41.857Z", "content": "I'm not familiar with submission selecting so much. But if I select submission seriously, I'll evaluate it by CV.
\n\nEdit: Also it depends on structure of train test split.
", "rawMarkdown": "I'm not familiar with submission selecting so much. But if I select submission seriously, I'll evaluate it by CV.\n\nEdit: Also it depends on structure of train test split. " } ] }, { "id": 214050, "postDate": "2017-08-15T20:33:29.923Z", "content": "Very nice and steady top-3 performance for such a long time!\n
I also thought you absolutely won this competition until the very last moments.
Well done, it was nice to compete with you and to understand your hints!\n
Obviously, you're a \"top gun\"!\n
Great thanks and good luck!
What is your setup that has 300 GB RAM?
", "rawMarkdown": "What is your setup that has 300 GB RAM?", "votes": 4 }, { "id": 213946, "postDate": "2017-08-15T15:01:21.197Z", "content": "Hey Onodera, thank you very much for your participation in the forum. It was really valuable. Congratulation for your second place.
", "rawMarkdown": "\nHey Onodera, thank you very much for your participation in the forum. It was really valuable. Congratulation for your second place.", "votes": 1 }, { "id": 213940, "postDate": "2017-08-15T14:55:48.657Z", "content": "Thank you for posting this, ONODERA. I really like your approach to feature engineering - a relatively small number of high signal features. Congratulations!!!
", "rawMarkdown": "Thank you for posting this, ONODERA. I really like your approach to feature engineering - a relatively small number of high signal features. Congratulations!!!", "votes": 1 }, { "id": 213939, "postDate": "2017-08-15T14:55:44.113Z", "content": "Thanks for posting your solution, and congratulations on a solo gold medal!
\n\nIt was fun to watch the intense battle between you and sjv, only for weiwei to come out of nowhere and take the lead at the last minute!
\n\nI was rooting for you because of your helpful posts on the forums, but I'm glad sjv also finished in the money given his really remarkable solution.
", "rawMarkdown": "Thanks for posting your solution, and congratulations on a solo gold medal!\n\nIt was fun to watch the intense battle between you and sjv, only for weiwei to come out of nowhere and take the lead at the last minute!\n\nI was rooting for you because of your helpful posts on the forums, but I'm glad sjv also finished in the money given his really remarkable solution. ", "votes": 2 }, { "id": 420827, "postDate": "2018-11-14T07:13:55.537Z", "content": "Your code is quite nice! I learned a lot from it.
\n\nWould you like to share the feature importance of your XGB model.
", "rawMarkdown": "Your code is quite nice! I learned a lot from it.\n\nWould you like to share the feature importance of your XGB model. " }, { "id": 419387, "postDate": "2018-11-11T21:25:03.430Z", "content": "Hi,\nDid you use last 3 orders of all the users for taining???
\n\nIf we use last 3 order of all the users then the dataset will be huge? where do you run the processes? \nPlease help!
", "rawMarkdown": "Hi,\nDid you use last 3 orders of all the users for taining???\n\nIf we use last 3 order of all the users then the dataset will be huge? where do you run the processes? \nPlease help!", "replies": [ { "id": 420049, "postDate": "2018-11-13T01:30:03.830Z", "content": "\n\n\nDid you use last 3 orders of all the users for taining???
\n
Yes, you need 200GB memory or so for running my model. Read README.md.
", "rawMarkdown": "> Did you use last 3 orders of all the users for taining???\n\nYes, you need 200GB memory or so for running my model. Read README.md.", "votes": 1 }, { "id": 432384, "postDate": "2018-12-03T19:42:05.687Z", "content": "No way I could run that!
", "rawMarkdown": "No way I could run that!" }, { "id": 432390, "postDate": "2018-12-03T19:46:27.250Z", "content": "Would you have a youtube video for your ppt? Thanks for sharing the Repo.
", "rawMarkdown": "Would you have a youtube video for your ppt? Thanks for sharing the Repo." } ] }, { "id": 341950, "postDate": "2018-06-12T15:07:08.917Z", "content": "Tensai!!!
", "rawMarkdown": "Tensai!!!" }, { "id": 259788, "postDate": "2017-12-19T01:45:40Z", "content": "It is amazing to see using xgboost alone can achieve 2nd place!
", "rawMarkdown": "It is amazing to see using xgboost alone can achieve 2nd place!" }, { "id": 214328, "postDate": "2017-08-16T15:05:33.250Z", "content": "Congrats ONODERA for your 2nd place., and thanks for sharing.
\n\nI really appreciated your lively approach and your sense of humour in discussion topics during the competition.
", "rawMarkdown": "Congrats ONODERA for your 2nd place., and thanks for sharing. \n\nI really appreciated your lively approach and your sense of humour in discussion topics during the competition. " }, { "id": 214184, "postDate": "2017-08-16T06:45:49.090Z", "content": "I noticed you and a few others who've posted solutions broke the day in to segments, aka time zones. In terms of detecting whether a user is making an unusual order was binning the orders like that with hard cutoff better than a weighted moving average? Or was the binning method better because it let you easily create more features using it. I'm wondering since with time zones it seems user who orders at 11.30am and then 12 noon would be in separate zones.
", "rawMarkdown": "I noticed you and a few others who've posted solutions broke the day in to segments, aka time zones. In terms of detecting whether a user is making an unusual order was binning the orders like that with hard cutoff better than a weighted moving average? Or was the binning method better because it let you easily create more features using it. I'm wondering since with time zones it seems user who orders at 11.30am and then 12 noon would be in separate zones. ", "replies": [ { "id": 214200, "postDate": "2017-08-16T07:53:24.297Z", "content": "Good question, I don't think one time zone suffice.\nI should have added another one.
", "rawMarkdown": "Good question, I don't think one time zone suffice.\nI should have added another one." } ] }, { "id": 214117, "postDate": "2017-08-16T02:12:47.460Z", "content": "Thanks for sharing! I was wondering how did you end up optimizing your hyperparameters for your final submission?
", "rawMarkdown": "Thanks for sharing! I was wondering how did you end up optimizing your hyperparameters for your final submission?", "replies": [ { "id": 214142, "postDate": "2017-08-16T03:38:12.770Z", "content": "Actually, I didn't think about hyperparameters so much. I only cared about necessary time.
", "rawMarkdown": "Actually, I didn't think about hyperparameters so much. I only cared about necessary time." } ] }, { "id": 214104, "postDate": "2017-08-16T01:27:08.537Z", "content": "Thank you for your sharing codes. You are so great!\nOne question, How did you select your 'special' features?
", "rawMarkdown": "Thank you for your sharing codes. You are so great!\nOne question, How did you select your 'special' features?", "replies": [ { "id": 214140, "postDate": "2017-08-16T03:32:25.917Z", "content": "I used feature importance of xgboost. We can get weight, gain and cover. I normalized those, sum, and then pick up from higher feature.
", "rawMarkdown": "I used feature importance of xgboost. We can get weight, gain and cover. I normalized those, sum, and then pick up from higher feature.", "votes": 5 }, { "id": 214196, "postDate": "2017-08-16T07:21:47.297Z", "content": "Thank you for your advice! :)
", "rawMarkdown": "Thank you for your advice! :)" } ] }, { "id": 214061, "postDate": "2017-08-15T21:15:05.277Z", "content": "Congrats and many thanks for your contribution!
", "rawMarkdown": "Congrats and many thanks for your contribution!" }, { "id": 214052, "postDate": "2017-08-15T20:38:50.773Z", "content": "Congratulations @Onodera and thanks for posting your solution. Because this is my 1st involvement in this king of market basket analysis problem, I plan on using NN to build a model on this data so that I can explore and learn some more. \nSo I will be checking the top10 solutions including yours for all the different approaches to this problem.
\n\nAnd thank you also for participating in the discussions which made this competition much more interesting.
", "rawMarkdown": "Congratulations @Onodera and thanks for posting your solution. Because this is my 1st involvement in this king of market basket analysis problem, I plan on using NN to build a model on this data so that I can explore and learn some more. \nSo I will be checking the top10 solutions including yours for all the different approaches to this problem.\n\nAnd thank you also for participating in the discussions which made this competition much more interesting." }, { "id": 214004, "postDate": "2017-08-15T17:42:46.847Z", "content": "Thanks Onodera! You did an incredible job and shared lots of useful information during the competition. This is my fist competition and I learned a lot from your kagglers! It IS a great community! Congrats!
", "rawMarkdown": "Thanks Onodera! You did an incredible job and shared lots of useful information during the competition. This is my fist competition and I learned a lot from your kagglers! It IS a great community! Congrats!" }, { "id": 213980, "postDate": "2017-08-15T16:19:12.770Z", "content": "Congrats and thanks for posting/sharing!
", "rawMarkdown": "Congrats and thanks for posting/sharing!" }, { "id": 214119, "postDate": "2017-08-16T02:23:50.770Z", "content": "Congrats! Thanks for sharing.
", "rawMarkdown": "Congrats! Thanks for sharing." }, { "id": 214087, "postDate": "2017-08-15T23:24:38.970Z", "content": "Congrats! Thanks for sharing.
", "rawMarkdown": "Congrats! Thanks for sharing." }, { "id": 213972, "postDate": "2017-08-15T15:58:54.807Z", "content": "Congrats and thanks for sharing.
", "rawMarkdown": "Congrats and thanks for sharing." }, { "id": 213958, "postDate": "2017-08-15T15:15:37.643Z", "content": "Thanks. You are still my hero.
", "rawMarkdown": "Thanks. You are still my hero." }, { "id": 213935, "postDate": "2017-08-15T14:53:00.107Z", "content": "Congrats and thanks for sharing.
", "rawMarkdown": "Congrats and thanks for sharing." } ], "complete": true, "source": "live", "fetched_at": "2026-10-20T02:03:43Z", "competition_id": 6644, "competition_slug": "basket-analysis", "team_id": 673171, "solution_url": "/competitions/basket-analysis/writeups/keep-out-2nd-place-solution" }