To Bosch and Kaggle Administrators,
\n\nWill this competition be reset or will it continue as is? I don't mind either way, but it's probably better to know sooner versus later.
\n\nThanks,
\n\nMike
", "messages": [ { "id": "138473", "postDate": "10/09/2016 01:52:24", "content": "To Bosch and Kaggle Administrators,
\n\nWill this competition be reset or will it continue as is? I don't mind either way, but it's probably better to know sooner versus later.
\n\nThanks,
\n\nMike
", "rawMarkdown": "To Bosch and Kaggle Administrators,\r\n\r\nWill this competition be reset or will it continue as is? I don't mind either way, but it's probably better to know sooner versus later.\r\n\r\nThanks,\r\n\r\nMike", "votes": null }, { "id": "138476", "postDate": "10/09/2016 02:43:50", "content": "I'd also like to know if it's acceptable to create features by using combined train+test data.
\n\nWhat Faron just posted appears to be much less effective if diffing Id's is performed on train and test separately.
", "rawMarkdown": "I'd also like to know if it's acceptable to create features by using combined train+test data. \r\n\r\nWhat Faron just posted appears to be *much* less effective if diffing Id's is performed on train and test separately.", "votes": null }, { "id": "138513", "postDate": "10/09/2016 09:43:03", "content": "Mike.
\n\nWhy do you ask this now?
\n\nYou had found the leak quite some time ago if I can judge from your score.
\n\nThe fact that the leak just became known to the wider community, does not change the fact that the leak was there and some people had found it plus others had admitted it.
\n\nOn the contrary (and since the competition has not been already reset ) I think now it makes more sense to compete as a competitor ( because the client would not have received the value of an unleaky solution anyway)
\n\nI don't think the competition will be reset based on what has happened in the past.
\n\nIrrespective of this, I also think the competition should be reset. It is the timing of this post that surprises me a bit and the fact that it comes from someone that most probably had found the leak?
", "rawMarkdown": "Mike. \r\n\r\nWhy do you ask this now? \r\n\r\nYou had found the leak quite some time ago if I can judge from your score.\r\n\r\nThe fact that the leak just became known to the wider community, does not change the fact that the leak was there and some people had found it plus others had admitted it.\r\n\r\nOn the contrary (and since the competition has not been already reset ) I think now it makes more sense to compete as a competitor ( because the client would not have received the value of an unleaky solution anyway) \r\n\r\n I don't think the competition will be reset based on what has happened in the past. \r\n\r\nIrrespective of this, I also think the competition should be reset. It is the timing of this post that surprises me a bit and the fact that it comes from someone that most probably had found the leak?", "votes": null }, { "id": "138516", "postDate": "10/09/2016 09:47:44", "content": "In the future, there should be something at the beginning of the competition that states whether IDs can be modelled on.
\n\n@Marios, the leak can be modelled without using IDs or row orders at all, it's possible that Mike found it this route.
", "rawMarkdown": "In the future, there should be something at the beginning of the competition that states whether IDs can be modelled on.\r\n\r\n@Marios, the leak can be modelled without using IDs or row orders at all, it's possible that Mike found it this route.", "votes": null }, { "id": "138519", "postDate": "10/09/2016 09:51:18", "content": "I don't think an administrator exists for this competition as I received no answer to my rule clarification ages ago because I hated how train and test were split
", "rawMarkdown": "I don't think an administrator exists for this competition as I received no answer to my rule clarification ages ago because I hated how train and test were split", "votes": null }, { "id": "138522", "postDate": "10/09/2016 09:58:43", "content": "[quote=anokas;138516]
\n\n@Marios, the leak can be modelled without using IDs or row orders at all, it's possible that Mike found it this route.
\n\n[/quote]
\n\nI know and you can get to a 0.4 model using a Raspberry Pi
\n\nI will not dwell more on this.
", "rawMarkdown": "[quote=anokas;138516]\r\n\r\n@Marios, the leak can be modelled without using IDs or row orders at all, it's possible that Mike found it this route.\r\n\r\n[/quote]\r\n\r\nI know and you can get to a [0.4 model using a Raspberry Pi][1] \r\n\r\nI will not dwell more on this.\r\n\r\n\r\n [1]: https://www.kaggle.com/c/bosch-production-line-performance/forums/t/24065/the-magical-feature-from-lb-0-3-to-0-4/138220#post138220", "votes": null }, { "id": "138541", "postDate": "10/09/2016 13:17:42", "content": "Hi everyone,
\n\nIn this competition, there are strange features which improve our LB score.
\nI don't know it is leak or not. We need more information about dataset.
When there is a mismatch between what the competition organizer wants and what we think they want, then that competition becomes a haunted one. This is definitely hurting
", "rawMarkdown": "When there is a mismatch between what the competition organizer wants and what we think they want, then that competition becomes a haunted one. This is definitely hurting", "votes": null }, { "id": "138563", "postDate": "10/09/2016 15:39:58", "content": "I am not using the "leak" found by Faron. You can get above 0.4 without using any Id based features.
\n\n[quote=Μαριος Μιχαηλιδης KazAnova;138513]
\n\nMike.
\n\nWhy do you ask this now?
\n\nYou had found the leak quite some time ago if I can judge from your score.
\n\nThe fact that the leak just became known to the wider community, does not change the fact that the leak was there and some people had found it plus others had admitted it.
\n\nOn the contrary (and since the competition has not been already reset ) I think now it makes more sense to compete as a competitor ( because the client would not have received the value of an unleaky solution anyway)
\n\nI don't think the competition will be reset based on what has happened in the past.
\n\nIrrespective of this, I also think the competition should be reset. It is the timing of this post that surprises me a bit and the fact that it comes from someone that most probably had found the leak?
\n\n[/quote]
", "rawMarkdown": "I am not using the \"leak\" found by Faron. You can get above 0.4 without using any Id based features. \r\n\r\n[quote=Μαριος Μιχαηλιδης KazAnova;138513]\r\n\r\nMike. \r\n\r\nWhy do you ask this now? \r\n\r\nYou had found the leak quite some time ago if I can judge from your score.\r\n\r\nThe fact that the leak just became known to the wider community, does not change the fact that the leak was there and some people had found it plus others had admitted it.\r\n\r\nOn the contrary (and since the competition has not been already reset ) I think now it makes more sense to compete as a competitor ( because the client would not have received the value of an unleaky solution anyway) \r\n\r\n I don't think the competition will be reset based on what has happened in the past. \r\n\r\nIrrespective of this, I also think the competition should be reset. It is the timing of this post that surprises me a bit and the fact that it comes from someone that most probably had found the leak? \r\n\r\n[/quote]", "votes": null }, { "id": "138564", "postDate": "10/09/2016 15:46:08", "content": "@Mike, so that means you are using the leak in some indirect sense?
", "rawMarkdown": "Mike, so that means you are using the leak in some indirect sense?", "votes": null }, { "id": "138580", "postDate": "10/09/2016 16:20:13", "content": "[quote=Μαριος Μιχαηλιδης KazAnova;138522]
\n\nI know and you can get to a 0.4 model using a Raspberry Pi
\n\n[/quote]
\n\nNice One there !!! Raspberry Pi, Can't stop laughing...
", "rawMarkdown": "[quote=Μαριος Μιχαηλιδης KazAnova;138522]\r\n\r\n\r\nI know and you can get to a [0.4 model using a Raspberry Pi][1] \r\n\r\n\r\n [1]: https://www.kaggle.com/c/bosch-production-line-performance/forums/t/24065/the-magical-feature-from-lb-0-3-to-0-4/138220#post138220\r\n\r\n[/quote]\r\n\r\nNice One there !!! Raspberry Pi, Can't stop laughing...", "votes": null }, { "id": "138593", "postDate": "10/09/2016 17:25:14", "content": "With all regards, but this seems a discussion about terms, not about the underlying data. The observations are related to objects that travel through a netwerk of stations with some 1700+ different routes. We are missing information about the nature of the parts that are being produced, but it is logical to assume that like parts will follow like routes. Some (type of) parts will probably be more difficult to produce than others:
\n\n# -- Stations and lines L1 L2\n# |---------- L0 --------|**|-|*------- L3 ----------*\n# "0---------1---------2---------3-xxx-----4---------5-"\n# "0123456789012345678901234567890123456789012345678901"\n# 1: 0000000000000000000000001010011010001100000000000000 98 231 0.4242424\n# 2: 0000000000000000000000001010011010010100000000000000 85 194 0.4381443\nA second observation is that stations can become more vulnerable to error over time, so it is completely logical to assume the existence of some dependency between the visitor of a station and its pre- or successor. Hence near observations are (at least a bit) dependent and any real life numbering scheme for the id's (such as product serial numbers) will consequently introduce id's that carry more information than a random unique number.
\n\nThat's not leakage, that's a characteristic of the problem.
", "rawMarkdown": "With all regards, but this seems a discussion about terms, not about the underlying data. The observations are related to objects that travel through a netwerk of stations with some 1700+ different routes. We are missing information about the nature of the parts that are being produced, but it is logical to assume that like parts will follow like routes. Some (type of) parts will probably be more difficult to produce than others:\r\n\r\n\r\n # -- Stations and lines L1 L2\r\n # |---------- L0 --------|**|-|*------- L3 ----------*\r\n # \"0---------1---------2---------3-xxx-----4---------5-\"\r\n # \"0123456789012345678901234567890123456789012345678901\"\r\n # 1: 0000000000000000000000001010011010001100000000000000 98 231 0.4242424\r\n # 2: 0000000000000000000000001010011010010100000000000000 85 194 0.4381443\r\n\r\n\r\nA second observation is that stations can become more vulnerable to error over time, so it is completely logical to assume the existence of some dependency between the visitor of a station and its pre- or successor. Hence near observations are (at least a bit) dependent and any real life numbering scheme for the id's (such as product serial numbers) will _consequently_ introduce id's that carry more information than a random unique number. \r\n\r\nThat's not leakage, that's a characteristic of the problem.", "votes": null }, { "id": "138622", "postDate": "10/09/2016 21:03:53", "content": "I had time to play with the data set only yesterday, and my current score of 0.17 is just a result of a couple of if-else clauses, with no ML models involved.
\n\nI chose not to call it out, simply because I hadn't yet worked on it further to score 0.40+ to prove my point.
\n\nFaron & Kazanova saved me the trouble, I send them my thanks and respects.
\n\nPlease stop coming up with scenarios about how the "magic features" might indeed be helpful to Bosch. The data set is just poorly prepared.
\n\nI don't care if the competition is reset or not, as long as the decision is made soon.
", "rawMarkdown": "I had time to play with the data set only yesterday, and my current score of 0.17 is just a result of a couple of if-else clauses, with no ML models involved.\r\n\r\nI chose not to call it out, simply because I hadn't yet worked on it further to score 0.40+ to prove my point.\r\n\r\nFaron & Kazanova saved me the trouble, I send them my thanks and respects.\r\n\r\nPlease stop coming up with scenarios about how the \"magic features\" might indeed be helpful to Bosch. The data set is just poorly prepared.\r\n\r\nI don't care if the competition is reset or not, as long as the decision is made soon.", "votes": null }, { "id": "138649", "postDate": "10/10/2016 05:32:49", "content": "PrakashSubedi - if Kazanova (or any Grandmaster for that matter) says they can get .4 on an rpi, I believe them.
\n\nI could probably only get to about .34 with what I currently know, and with some time to write an efficient CSV parser so it wouldn't take too long to run ;)
", "rawMarkdown": "PrakashSubedi - if Kazanova (or any Grandmaster for that matter) says they can get .4 on an rpi, I believe them.\r\n\r\nI could probably only get to about .34 with what I currently know, and with some time to write an efficient CSV parser so it wouldn't take too long to run ;)", "votes": null }, { "id": "138651", "postDate": "10/10/2016 05:36:06", "content": "well i thought that was just a fun.. My bad !!! didn't read anokas post sayimg he actually intend to build a model to run on Pi.. would be interesting to see that..
", "rawMarkdown": "well i thought that was just a fun.. My bad !!! didn't read anokas post sayimg he actually intend to build a model to run on Pi.. would be interesting to see that..", "votes": null }, { "id": "138735", "postDate": "10/10/2016 17:14:37", "content": "[quote=PrakashSubedi;138651]
\n\nwell i thought that was just a fun.. My bad !!! didn't read anokas post sayimg he actually intend to build a model to run on Pi.. would be interesting to see that..
\n\n[/quote]
\n\nYeah, I'm working on it right now :D
\n\nShould be interesting to see if I can actually get a reasonable score, quite a fun little project.
", "rawMarkdown": "[quote=PrakashSubedi;138651]\r\n\r\nwell i thought that was just a fun.. My bad !!! didn't read anokas post sayimg he actually intend to build a model to run on Pi.. would be interesting to see that..\r\n\r\n[/quote]\r\n\r\nYeah, I'm working on it right now :D\r\n\r\nShould be interesting to see if I can actually get a reasonable score, quite a fun little project.", "votes": null }, { "id": "138746", "postDate": "10/10/2016 18:24:27", "content": "Keep calm and carry on. There are currently no plans to reset this competition.
", "rawMarkdown": "Keep calm and carry on. There are currently no plans to reset this competition.", "votes": null }, { "id": "138773", "postDate": "10/10/2016 20:56:34", "content": "Hi everbody,
\n\nIt has not been the first time this year. Neither the second. Problems related to data sets have been recurrent here on Kaggle.
\n\nSorry, I will not waste my time finding problems in data sets again. It is not only related to get points in the ranking. It is about our time. Either you contest trying to find problems or you do not have chance in the challenge. Am I right? It is discouraging.
\n\nGood luck for you guys.
\n\nRegards
", "rawMarkdown": "Hi everbody,\r\n\r\nIt has not been the first time this year. Neither the second. Problems related to data sets have been recurrent here on Kaggle.\r\n\r\nSorry, I will not waste my time finding problems in data sets again. It is not only related to get points in the ranking. It is about our time. Either you contest trying to find problems or you do not have chance in the challenge. Am I right? It is discouraging. \r\n\r\nGood luck for you guys.\r\n\r\nRegards", "votes": null }, { "id": "138777", "postDate": "10/10/2016 21:09:00", "content": "This was my first time participating in Kaggle competition and I find it total waste of time! Kaggle brings lots of smart people together to solve a stupid problem. Even after recognizing the problem, they are not willing to even simply apologize.
", "rawMarkdown": "This was my first time participating in Kaggle competition and I find it total waste of time! Kaggle brings lots of smart people together to solve a stupid problem. Even after recognizing the problem, they are not willing to even simply apologize.", "votes": null }, { "id": "138924", "postDate": "10/11/2016 18:07:00", "content": "Acoording to new kernels, it is not a leak. It is part of the problem.
\n\nWe need to be careful when using the word 'leak'.
\n\nThen, let's work guys...
", "rawMarkdown": "Acoording to new kernels, it is not a leak. It is part of the problem.\r\n\r\nWe need to be careful when using the word 'leak'.\r\n\r\nThen, let's work guys...", "votes": null }, { "id": "138928", "postDate": "10/11/2016 18:20:31", "content": "[quote=Humberto Brandão;138924]
\n\nAcoording to new kernels, it is not a leak. It is part of the problem.
\n\nWe need to be careful when using the word 'leak'.
\n\nThen, let's work guys...
\n\n[/quote]
\n\nWhat do you mean by 'new kernels'?
", "rawMarkdown": "[quote=Humberto Brandão;138924]\r\n\r\nAcoording to new kernels, it is not a leak. It is part of the problem.\r\n\r\nWe need to be careful when using the word 'leak'.\r\n\r\nThen, let's work guys...\r\n\r\n[/quote]\r\n\r\nWhat do you mean by 'new kernels'?", "votes": null }, { "id": "138932", "postDate": "10/11/2016 18:26:59", "content": "@anokas - I'm curious - can you get to .9 AUC without processing train+test together?
\n\nAnd I checked for new kernels, and I don't see anything rabbit-related not based on Faron's code.
", "rawMarkdown": "anokas - I'm curious - can you get to .9 AUC without processing train+test together?\r\n\r\nAnd I checked for new kernels, and I don't see anything rabbit-related not based on Faron's code.", "votes": null }, { "id": "138935", "postDate": "10/11/2016 18:29:29", "content": "[quote=happycube;138932]
\n\n@anokas - I'm curious - can you get to .9 AUC without processing train+test together?
\n\nAnd I checked for new kernels, and I don't see anything rabbit-related not based on Faron's code.
\n\n[/quote]
\n\nMy response to that is "I think so", but I'm not sure because I haven't tried running a model on it. But I'm relatively confident. Our super features actually had low AUC, but high MCC if I remember correctly.
", "rawMarkdown": "[quote=happycube;138932]\r\n\r\n@anokas - I'm curious - can you get to .9 AUC without processing train+test together?\r\n\r\nAnd I checked for new kernels, and I don't see anything rabbit-related not based on Faron's code.\r\n\r\n[/quote]\r\n\r\nMy response to that is \"I think so\", but I'm not sure because I haven't tried running a model on it. But I'm relatively confident. Our super features actually had low AUC, but high MCC if I remember correctly.", "votes": null } ], "comments": [ { "id": 138476, "author_name": "happycube", "author_url": "", "post_date": "10/09/2016 02:43:50", "content": "I'd also like to know if it's acceptable to create features by using combined train+test data.
\n\nWhat Faron just posted appears to be much less effective if diffing Id's is performed on train and test separately.
", "votes": null, "replies": [] }, { "id": 138513, "author_name": "kazanova", "author_url": "", "post_date": "10/09/2016 09:43:03", "content": "Mike.
\n\nWhy do you ask this now?
\n\nYou had found the leak quite some time ago if I can judge from your score.
\n\nThe fact that the leak just became known to the wider community, does not change the fact that the leak was there and some people had found it plus others had admitted it.
\n\nOn the contrary (and since the competition has not been already reset ) I think now it makes more sense to compete as a competitor ( because the client would not have received the value of an unleaky solution anyway)
\n\nI don't think the competition will be reset based on what has happened in the past.
\n\nIrrespective of this, I also think the competition should be reset. It is the timing of this post that surprises me a bit and the fact that it comes from someone that most probably had found the leak?
", "votes": null, "replies": [] }, { "id": 138516, "author_name": "anokas", "author_url": "", "post_date": "10/09/2016 09:47:44", "content": "In the future, there should be something at the beginning of the competition that states whether IDs can be modelled on.
\n\n@Marios, the leak can be modelled without using IDs or row orders at all, it's possible that Mike found it this route.
", "votes": null, "replies": [] }, { "id": 138519, "author_name": "scirpus", "author_url": "", "post_date": "10/09/2016 09:51:18", "content": "I don't think an administrator exists for this competition as I received no answer to my rule clarification ages ago because I hated how train and test were split
", "votes": null, "replies": [] }, { "id": 138522, "author_name": "kazanova", "author_url": "", "post_date": "10/09/2016 09:58:43", "content": "[quote=anokas;138516]
\n\n@Marios, the leak can be modelled without using IDs or row orders at all, it's possible that Mike found it this route.
\n\n[/quote]
\n\nI know and you can get to a 0.4 model using a Raspberry Pi
\n\nI will not dwell more on this.
", "votes": null, "replies": [] }, { "id": 138541, "author_name": "toshik", "author_url": "", "post_date": "10/09/2016 13:17:42", "content": "Hi everyone,
\n\nIn this competition, there are strange features which improve our LB score.
\nI don't know it is leak or not. We need more information about dataset.
When there is a mismatch between what the competition organizer wants and what we think they want, then that competition becomes a haunted one. This is definitely hurting
", "votes": null, "replies": [] }, { "id": 138563, "author_name": "mikeskim", "author_url": "", "post_date": "10/09/2016 15:39:58", "content": "I am not using the "leak" found by Faron. You can get above 0.4 without using any Id based features.
\n\n[quote=Μαριος Μιχαηλιδης KazAnova;138513]
\n\nMike.
\n\nWhy do you ask this now?
\n\nYou had found the leak quite some time ago if I can judge from your score.
\n\nThe fact that the leak just became known to the wider community, does not change the fact that the leak was there and some people had found it plus others had admitted it.
\n\nOn the contrary (and since the competition has not been already reset ) I think now it makes more sense to compete as a competitor ( because the client would not have received the value of an unleaky solution anyway)
\n\nI don't think the competition will be reset based on what has happened in the past.
\n\nIrrespective of this, I also think the competition should be reset. It is the timing of this post that surprises me a bit and the fact that it comes from someone that most probably had found the leak?
\n\n[/quote]
", "votes": null, "replies": [] }, { "id": 138564, "author_name": "ascending", "author_url": "", "post_date": "10/09/2016 15:46:08", "content": "@Mike, so that means you are using the leak in some indirect sense?
", "votes": null, "replies": [] }, { "id": 138580, "author_name": "gullu39", "author_url": "", "post_date": "10/09/2016 16:20:13", "content": "[quote=Μαριος Μιχαηλιδης KazAnova;138522]
\n\nI know and you can get to a 0.4 model using a Raspberry Pi
\n\n[/quote]
\n\nNice One there !!! Raspberry Pi, Can't stop laughing...
", "votes": null, "replies": [] }, { "id": 138593, "author_name": "mrooijer", "author_url": "", "post_date": "10/09/2016 17:25:14", "content": "With all regards, but this seems a discussion about terms, not about the underlying data. The observations are related to objects that travel through a netwerk of stations with some 1700+ different routes. We are missing information about the nature of the parts that are being produced, but it is logical to assume that like parts will follow like routes. Some (type of) parts will probably be more difficult to produce than others:
\n\n# -- Stations and lines L1 L2\n# |---------- L0 --------|**|-|*------- L3 ----------*\n# "0---------1---------2---------3-xxx-----4---------5-"\n# "0123456789012345678901234567890123456789012345678901"\n# 1: 0000000000000000000000001010011010001100000000000000 98 231 0.4242424\n# 2: 0000000000000000000000001010011010010100000000000000 85 194 0.4381443\nA second observation is that stations can become more vulnerable to error over time, so it is completely logical to assume the existence of some dependency between the visitor of a station and its pre- or successor. Hence near observations are (at least a bit) dependent and any real life numbering scheme for the id's (such as product serial numbers) will consequently introduce id's that carry more information than a random unique number.
\n\nThat's not leakage, that's a characteristic of the problem.
", "votes": null, "replies": [] }, { "id": 138622, "author_name": "barisumog", "author_url": "", "post_date": "10/09/2016 21:03:53", "content": "I had time to play with the data set only yesterday, and my current score of 0.17 is just a result of a couple of if-else clauses, with no ML models involved.
\n\nI chose not to call it out, simply because I hadn't yet worked on it further to score 0.40+ to prove my point.
\n\nFaron & Kazanova saved me the trouble, I send them my thanks and respects.
\n\nPlease stop coming up with scenarios about how the "magic features" might indeed be helpful to Bosch. The data set is just poorly prepared.
\n\nI don't care if the competition is reset or not, as long as the decision is made soon.
", "votes": null, "replies": [] }, { "id": 138649, "author_name": "happycube", "author_url": "", "post_date": "10/10/2016 05:32:49", "content": "PrakashSubedi - if Kazanova (or any Grandmaster for that matter) says they can get .4 on an rpi, I believe them.
\n\nI could probably only get to about .34 with what I currently know, and with some time to write an efficient CSV parser so it wouldn't take too long to run ;)
", "votes": null, "replies": [] }, { "id": 138651, "author_name": "gullu39", "author_url": "", "post_date": "10/10/2016 05:36:06", "content": "well i thought that was just a fun.. My bad !!! didn't read anokas post sayimg he actually intend to build a model to run on Pi.. would be interesting to see that..
", "votes": null, "replies": [] }, { "id": 138735, "author_name": "anokas", "author_url": "", "post_date": "10/10/2016 17:14:37", "content": "[quote=PrakashSubedi;138651]
\n\nwell i thought that was just a fun.. My bad !!! didn't read anokas post sayimg he actually intend to build a model to run on Pi.. would be interesting to see that..
\n\n[/quote]
\n\nYeah, I'm working on it right now :D
\n\nShould be interesting to see if I can actually get a reasonable score, quite a fun little project.
", "votes": null, "replies": [] }, { "id": 138746, "author_name": "wcukierski", "author_url": "", "post_date": "10/10/2016 18:24:27", "content": "Keep calm and carry on. There are currently no plans to reset this competition.
", "votes": null, "replies": [] }, { "id": 138773, "author_name": "brandao", "author_url": "", "post_date": "10/10/2016 20:56:34", "content": "Hi everbody,
\n\nIt has not been the first time this year. Neither the second. Problems related to data sets have been recurrent here on Kaggle.
\n\nSorry, I will not waste my time finding problems in data sets again. It is not only related to get points in the ranking. It is about our time. Either you contest trying to find problems or you do not have chance in the challenge. Am I right? It is discouraging.
\n\nGood luck for you guys.
\n\nRegards
", "votes": null, "replies": [] }, { "id": 138777, "author_name": "mojtabaebrahimi85", "author_url": "", "post_date": "10/10/2016 21:09:00", "content": "This was my first time participating in Kaggle competition and I find it total waste of time! Kaggle brings lots of smart people together to solve a stupid problem. Even after recognizing the problem, they are not willing to even simply apologize.
", "votes": null, "replies": [] }, { "id": 138924, "author_name": "brandao", "author_url": "", "post_date": "10/11/2016 18:07:00", "content": "Acoording to new kernels, it is not a leak. It is part of the problem.
\n\nWe need to be careful when using the word 'leak'.
\n\nThen, let's work guys...
", "votes": null, "replies": [] }, { "id": 138928, "author_name": "anokas", "author_url": "", "post_date": "10/11/2016 18:20:31", "content": "[quote=Humberto Brandão;138924]
\n\nAcoording to new kernels, it is not a leak. It is part of the problem.
\n\nWe need to be careful when using the word 'leak'.
\n\nThen, let's work guys...
\n\n[/quote]
\n\nWhat do you mean by 'new kernels'?
", "votes": null, "replies": [] }, { "id": 138932, "author_name": "happycube", "author_url": "", "post_date": "10/11/2016 18:26:59", "content": "@anokas - I'm curious - can you get to .9 AUC without processing train+test together?
\n\nAnd I checked for new kernels, and I don't see anything rabbit-related not based on Faron's code.
", "votes": null, "replies": [] }, { "id": 138935, "author_name": "anokas", "author_url": "", "post_date": "10/11/2016 18:29:29", "content": "[quote=happycube;138932]
\n\n@anokas - I'm curious - can you get to .9 AUC without processing train+test together?
\n\nAnd I checked for new kernels, and I don't see anything rabbit-related not based on Faron's code.
\n\n[/quote]
\n\nMy response to that is "I think so", but I'm not sure because I haven't tried running a model on it. But I'm relatively confident. Our super features actually had low AUC, but high MCC if I remember correctly.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "138473": "To Bosch and Kaggle Administrators,\r\n\r\nWill this competition be reset or will it continue as is? I don't mind either way, but it's probably better to know sooner versus later.\r\n\r\nThanks,\r\n\r\nMike", "138476": "I'd also like to know if it's acceptable to create features by using combined train+test data. \r\n\r\nWhat Faron just posted appears to be *much* less effective if diffing Id's is performed on train and test separately.", "138513": "Mike. \r\n\r\nWhy do you ask this now? \r\n\r\nYou had found the leak quite some time ago if I can judge from your score.\r\n\r\nThe fact that the leak just became known to the wider community, does not change the fact that the leak was there and some people had found it plus others had admitted it.\r\n\r\nOn the contrary (and since the competition has not been already reset ) I think now it makes more sense to compete as a competitor ( because the client would not have received the value of an unleaky solution anyway) \r\n\r\n I don't think the competition will be reset based on what has happened in the past. \r\n\r\nIrrespective of this, I also think the competition should be reset. It is the timing of this post that surprises me a bit and the fact that it comes from someone that most probably had found the leak?", "138516": "In the future, there should be something at the beginning of the competition that states whether IDs can be modelled on.\r\n\r\n@Marios, the leak can be modelled without using IDs or row orders at all, it's possible that Mike found it this route.", "138519": "I don't think an administrator exists for this competition as I received no answer to my rule clarification ages ago because I hated how train and test were split", "138522": "[quote=anokas;138516]\r\n\r\n@Marios, the leak can be modelled without using IDs or row orders at all, it's possible that Mike found it this route.\r\n\r\n[/quote]\r\n\r\nI know and you can get to a [0.4 model using a Raspberry Pi][1] \r\n\r\nI will not dwell more on this.\r\n\r\n\r\n [1]: https://www.kaggle.com/c/bosch-production-line-performance/forums/t/24065/the-magical-feature-from-lb-0-3-to-0-4/138220#post138220", "138541": "Hi everyone,\r\n\r\nIn this competition, there are strange features which improve our LB score.