I am not writting this because i am not in gold zone ,but just sharing some observation by my one of team member.
\nFew days ago some thing bit strange was found, lb gold zone was almost deadlocked at 90.6 but suddenly 4 to 5 teams in no time came simultaneously to score at 90.7 and getting into gold range. Generally this happens if some one shares a public notebook capable of getting such score or some useful tips. I did a dig dint find any such notebook or tips published in that time frame.
\nSuspected range of fishy score position in lb
\n**from 6 to 17. **
\nI may be wrong about certain folks so please pardon me its not for you at all. . From past revelations of this sort some people might have learnt so they may be ensuring that just by making few sub they dont get very high in lb so they might have done regular submissions like others.
Once again i pardon if this is not a case. I just shared in case of such possibility upon finding unsual trend which may be coincidental also.
\nSo upto kaggle team for investigation about this
Regards
\nJaideep
I am not writting this because i am not in gold zone ,but just sharing some observation by my one of team member.
\nFew days ago some thing bit strange was found, lb gold zone was almost deadlocked at 90.6 but suddenly 4 to 5 teams in no time came simultaneously to score at 90.7 and getting into gold range. Generally this happens if some one shares a public notebook capable of getting such score or some useful tips. I did a dig dint find any such notebook or tips published in that time frame.
\nSuspected range of fishy score position in lb
\n**from 6 to 17. **
\nI may be wrong about certain folks so please pardon me its not for you at all. . From past revelations of this sort some people might have learnt so they may be ensuring that just by making few sub they dont get very high in lb so they might have done regular submissions like others.
Once again i pardon if this is not a case. I just shared in case of such possibility upon finding unsual trend which may be coincidental also.
\nSo upto kaggle team for investigation about this
Regards
\nJaideep
here are two examples from my experience
\nin one non-kaggle competition a team from in-the-money zone offered to merge. My position was very close to money so I felt we can pool it off together and I gladly accepted. All our ensembles were bad (a bit better than my score but worse than theirs). Our team improved a little thanks to what i contributed. Later, pandas.DataFrame.corr() revieled that all ensemble submissions were actually slightly adjusted submissions from my models. Also, I was surprised with the performance of late comers, as some managed to jump over us in 2-3 subs :).
On Alaska2 I was in gold zone 10 days before the end. A kaggle master contacted me and offered a merge. She also mentioned 8xV100. We said no. In the next two days I saw her merge with someone above us and suddenly quite a few participants jumped over us within hours, but also over them. That competition was very stable until that point and I felt something bad is going on. I did not have proof, so I stayed silent. I contacted her previous teammates and they all have the same story \"Yeah, she is the one with a lot of GPUs. We merged, but then she disappeared. She was very busy.\" She would activate only when other teammates made improvements.
So there you have it.
\nIs there a way to handle this. I don't see it and I accept such reality.
\nHere just to learn. Good luck to everyone.
", "rawMarkdown": "here are two examples from my experience\n\n1. in one non-kaggle competition a team from in-the-money zone offered to merge. My position was very close to money so I felt we can pool it off together and I gladly accepted. All our ensembles were bad (a bit better than my score but worse than theirs). Our team improved a little thanks to what i contributed. Later, pandas.DataFrame.corr() revieled that all ensemble submissions were actually slightly adjusted submissions from my models. Also, I was surprised with the performance of late comers, as some managed to jump over us in 2-3 subs :).\n\n2. On Alaska2 I was in gold zone 10 days before the end. A kaggle master contacted me and offered a merge. She also mentioned 8xV100. We said no. In the next two days I saw her merge with someone above us and suddenly quite a few participants jumped over us within hours, but also over them. That competition was very stable until that point and I felt something bad is going on. I did not have proof, so I stayed silent. I contacted her previous teammates and they all have the same story \"Yeah, she is the one with a lot of GPUs. We merged, but then she disappeared. She was very busy.\" She would activate only when other teammates made improvements. \n\nSo there you have it.\n\nIs there a way to handle this. I don't see it and I accept such reality. \n\nHere just to learn. Good luck to everyone.", "votes": 9, "replies": [ { "id": 1185254, "postDate": "2021-02-04T04:25:27.160Z", "content": "I see back to back placed lb teams with same score after this thread have started merging :) …
", "rawMarkdown": "I see back to back placed lb teams with same score after this thread have started merging :) ... " } ] }, { "id": 1185696, "postDate": "2021-02-04T09:45:14.473Z", "content": "Kaggle should give score upto 4 or 5 decimal point like 0.9046 or .90326. It will be easy to figure out which all are getting exact same score.
", "rawMarkdown": "Kaggle should give score upto 4 or 5 decimal point like 0.9046 or .90326. It will be easy to figure out which all are getting exact same score.", "votes": 8 }, { "id": 1184481, "postDate": "2021-02-03T14:47:23.170Z", "content": "This is code competition. Hopefully Kaggle will check, at least, all Gold inference notebooks on private LB and request the corresponding training code if necessary.
\nI would even agree we make public chosen inference submission notebooks for all the LB.
\n", "rawMarkdown": "This is code competition. Hopefully Kaggle will check, at least, all Gold inference notebooks on private LB and request the corresponding training code if necessary.\n\nI would even agree we make public chosen inference submission notebooks for all the LB. \n\ncc @juliaelliott @sohier", "votes": 3, "replies": [ { "id": 1185000, "postDate": "2021-02-03T23:38:24.397Z", "content": "I agree this opinion.
", "rawMarkdown": "I agree this opinion." } ] }, { "id": 1184325, "postDate": "2021-02-03T13:43:34.233Z", "content": "Something was going on here too.
", "rawMarkdown": "Something was going on [here](https://www.kaggle.com/c/rfcx-species-audio-detection/discussion/216003) too.", "votes": 3 }, { "id": 1184343, "postDate": "2021-02-03T13:51:20.647Z", "content": "I suspect that Private sharing is common in Chinese Competition Zones.
", "rawMarkdown": "I suspect that Private sharing is common in Chinese Competition Zones.", "votes": 1 }, { "id": 1185250, "postDate": "2021-02-04T04:20:12.187Z", "content": "I find see some consecutive ranked lb teams with same score have started to merge now … so is my doubt really look witch hunt to some folks ..
", "rawMarkdown": "I find see some consecutive ranked lb teams with same score have started to merge now ... so is my doubt really look witch hunt to some folks ..\n", "votes": 1, "replies": [ { "id": 1185304, "postDate": "2021-02-04T05:31:24.253Z", "content": "One of them just suscribed 2 months ago ^^
\nSome cheaters are so obvious
\nDon't worry, Kaggle will investigate their cases.
", "rawMarkdown": "One of them just suscribed 2 months ago ^^\n\nSome cheaters are so obvious \n\nDon't worry, Kaggle will investigate their cases. " }, { "id": 1185357, "postDate": "2021-02-04T06:04:37.573Z", "rawMarkdown": "", "isDeleted": true }, { "id": 1185363, "postDate": "2021-02-04T06:09:24.993Z", "content": "In HuBMAP I found a guy rank 22,just submit twice.He seems to join kaggle these days.
\nMy be he is a genius.
Working nearly two month and failed to achieve 905 vs join 2 day ago rank 22/960. I may consider go home and farming
", "rawMarkdown": "Working nearly two month and failed to achieve 905 vs join 2 day ago rank 22/960. I may consider go home and farming", "votes": 12 }, { "id": 1185424, "postDate": "2021-02-04T06:47:54.187Z", "content": "knowing the fact notebook sharing in private mode is detectable in kaggle so these folks share or buy code from other environments.. Not sure if kaggle has developed some better means to detect fishy submissions.
\nHope our words are reaching to ears of kaggle team for their notice and investigations
rank 20 and rank 22 in HuBMAP
", "rawMarkdown": "rank 20 and rank 22 in HuBMAP\n" } ] }, { "id": 1184431, "postDate": "2021-02-03T14:23:28.123Z", "content": "Has something to do with this ?
\nThe author says the guy who wanted to pay him, is also currently at the top of this LB
\nMay be the guy managed to pay another one at the top here.
", "rawMarkdown": "Has something to do with [this](https://www.kaggle.com/c/hubmap-kidney-segmentation/discussion/211212) ? \n\nThe author says the guy who wanted to pay him, is also currently at the top of this LB\n\nMay be the guy managed to pay another one at the top here. ", "votes": 2 }, { "id": 1184306, "postDate": "2021-02-03T13:34:51.060Z", "content": "@juliaelliott fyi
", "rawMarkdown": "@juliaelliott fyi", "votes": -1 }, { "id": 1184321, "postDate": "2021-02-03T13:41:36.873Z", "content": "can these witchhunts please stop. take this up with the mods
", "rawMarkdown": "can these witchhunts please stop. take this up with the mods", "votes": -6, "replies": [ { "id": 1184445, "postDate": "2021-02-03T14:29:58.080Z", "content": "I would agree if there were not too many cheaters who tried shamelessly to cheat in so many ways, particularly during the last 2 years, and who would ruin the credibility and reputation of kaggle if nothing is done.
", "rawMarkdown": "I would agree if there were not too many cheaters who tried shamelessly to cheat in so many ways, particularly during the last 2 years, and who would ruin the credibility and reputation of kaggle if nothing is done. ", "votes": 3 }, { "id": 1184471, "postDate": "2021-02-03T14:40:28.513Z", "content": "@proletheus m just citing a possibility baesd on past experience and recent observation..
\npeople try all means.
you are citing a possibility… and then spam it all over the forum. why do you need to say the same thing in the looking for a team thread?
", "rawMarkdown": "you are citing a possibility... and then spam it all over the forum. why do you need to say the same thing in the looking for a team thread?", "votes": -3 }, { "id": 1184540, "postDate": "2021-02-03T15:09:31.337Z", "content": "it was nothing intentional..just was pointing to Admin..
", "rawMarkdown": "it was nothing intentional..just was pointing to Admin.. " }, { "id": 1184543, "postDate": "2021-02-03T15:12:10.980Z", "content": "what..? are you in a haze? how can writing multiple comments not be intentional
", "rawMarkdown": "what..? are you in a haze? how can writing multiple comments not be intentional", "votes": -3 }, { "id": 1184588, "postDate": "2021-02-03T15:47:41.157Z", "content": "may be yes..i deleted now you can go and check…i ealized after posting better i put it as topic for awareness and alertness
", "rawMarkdown": "may be yes..i deleted now you can go and check...i ealized after posting better i put it as topic for awareness and alertness" } ] }, { "id": 1188306, "postDate": "2021-02-06T05:58:08.567Z", "content": "You shuld @ Kaggle starf.
", "rawMarkdown": "You shuld @ Kaggle starf." }, { "id": 1185150, "postDate": "2021-02-04T01:46:00.293Z", "content": "Aha,This is another interesting case.
", "rawMarkdown": "Aha,This is another interesting [case](https://www.kaggle.com/c/global-wheat-detection/discussion/161296).\n" }, { "id": 1184381, "postDate": "2021-02-03T14:07:26.493Z", "rawMarkdown": "", "votes": 1, "isDeleted": true, "replies": [ { "id": 1184418, "postDate": "2021-02-03T14:17:21.057Z", "rawMarkdown": "", "isDeleted": true }, { "id": 1184421, "postDate": "2021-02-03T14:17:47.190Z", "rawMarkdown": "", "isDeleted": true }, { "id": 1184423, "postDate": "2021-02-03T14:18:13.733Z", "content": "haha yes thanks :D
", "rawMarkdown": "haha yes thanks :D" } ] } ], "comments": [ { "id": 1184550, "author_name": "Miroslav Valan", "author_url": "", "post_date": "2021-02-03T15:16:28.863000", "content": "here are two examples from my experience
\nin one non-kaggle competition a team from in-the-money zone offered to merge. My position was very close to money so I felt we can pool it off together and I gladly accepted. All our ensembles were bad (a bit better than my score but worse than theirs). Our team improved a little thanks to what i contributed. Later, pandas.DataFrame.corr() revieled that all ensemble submissions were actually slightly adjusted submissions from my models. Also, I was surprised with the performance of late comers, as some managed to jump over us in 2-3 subs :).
On Alaska2 I was in gold zone 10 days before the end. A kaggle master contacted me and offered a merge. She also mentioned 8xV100. We said no. In the next two days I saw her merge with someone above us and suddenly quite a few participants jumped over us within hours, but also over them. That competition was very stable until that point and I felt something bad is going on. I did not have proof, so I stayed silent. I contacted her previous teammates and they all have the same story \"Yeah, she is the one with a lot of GPUs. We merged, but then she disappeared. She was very busy.\" She would activate only when other teammates made improvements.
So there you have it.
\nIs there a way to handle this. I don't see it and I accept such reality.
\nHere just to learn. Good luck to everyone.
", "votes": 9, "replies": [ { "id": 1185254, "author_name": "Jaideep", "author_url": "", "post_date": "2021-02-04T04:25:27.160000", "content": "I see back to back placed lb teams with same score after this thread have started merging :) …
", "votes": 0, "replies": [] } ] }, { "id": 1185696, "author_name": "Vicky Goyal", "author_url": "", "post_date": "2021-02-04T09:45:14.473000", "content": "Kaggle should give score upto 4 or 5 decimal point like 0.9046 or .90326. It will be easy to figure out which all are getting exact same score.
", "votes": 8, "replies": [] }, { "id": 1184481, "author_name": "Serigne ", "author_url": "", "post_date": "2021-02-03T14:47:23.170000", "content": "This is code competition. Hopefully Kaggle will check, at least, all Gold inference notebooks on private LB and request the corresponding training code if necessary.
\nI would even agree we make public chosen inference submission notebooks for all the LB.
\n", "votes": 3, "replies": [ { "id": 1185000, "author_name": "Yoshio Sugiyama", "author_url": "", "post_date": "2021-02-03T23:38:24.397000", "content": "I agree this opinion.
", "votes": 0, "replies": [] } ] }, { "id": 1184325, "author_name": "Dracarys", "author_url": "", "post_date": "2021-02-03T13:43:34.233000", "content": "Something was going on here too.
", "votes": 3, "replies": [] }, { "id": 1184343, "author_name": "HEEEEE", "author_url": "", "post_date": "2021-02-03T13:51:20.647000", "content": "I suspect that Private sharing is common in Chinese Competition Zones.
", "votes": 1, "replies": [] }, { "id": 1185250, "author_name": "Jaideep", "author_url": "", "post_date": "2021-02-04T04:20:12.187000", "content": "I find see some consecutive ranked lb teams with same score have started to merge now … so is my doubt really look witch hunt to some folks ..
", "votes": 1, "replies": [ { "id": 1185304, "author_name": "Serigne ", "author_url": "", "post_date": "2021-02-04T05:31:24.253000", "content": "One of them just suscribed 2 months ago ^^
\nSome cheaters are so obvious
\nDon't worry, Kaggle will investigate their cases.
", "votes": 0, "replies": [] }, { "id": 1185357, "author_name": "", "author_url": "", "post_date": "2021-02-04T06:04:37.573000", "content": "", "votes": 0, "replies": [] }, { "id": 1185363, "author_name": "HEEEEE", "author_url": "", "post_date": "2021-02-04T06:09:24.993000", "content": "In HuBMAP I found a guy rank 22,just submit twice.He seems to join kaggle these days.
\nMy be he is a genius.
Working nearly two month and failed to achieve 905 vs join 2 day ago rank 22/960. I may consider go home and farming
", "votes": 12, "replies": [] }, { "id": 1185424, "author_name": "Jaideep", "author_url": "", "post_date": "2021-02-04T06:47:54.187000", "content": "knowing the fact notebook sharing in private mode is detectable in kaggle so these folks share or buy code from other environments.. Not sure if kaggle has developed some better means to detect fishy submissions.
\nHope our words are reaching to ears of kaggle team for their notice and investigations
rank 20 and rank 22 in HuBMAP
", "votes": 0, "replies": [] } ] }, { "id": 1184431, "author_name": "Serigne ", "author_url": "", "post_date": "2021-02-03T14:23:28.123000", "content": "Has something to do with this ?
\nThe author says the guy who wanted to pay him, is also currently at the top of this LB
\nMay be the guy managed to pay another one at the top here.
", "votes": 2, "replies": [] }, { "id": 1184306, "author_name": "Jaideep", "author_url": "", "post_date": "2021-02-03T13:34:51.060000", "content": "@juliaelliott fyi
", "votes": -1, "replies": [] }, { "id": 1184321, "author_name": "Proletheus", "author_url": "", "post_date": "2021-02-03T13:41:36.873000", "content": "can these witchhunts please stop. take this up with the mods
", "votes": -6, "replies": [ { "id": 1184445, "author_name": "Serigne ", "author_url": "", "post_date": "2021-02-03T14:29:58.080000", "content": "I would agree if there were not too many cheaters who tried shamelessly to cheat in so many ways, particularly during the last 2 years, and who would ruin the credibility and reputation of kaggle if nothing is done.
", "votes": 3, "replies": [] }, { "id": 1184471, "author_name": "Jaideep", "author_url": "", "post_date": "2021-02-03T14:40:28.513000", "content": "@proletheus m just citing a possibility baesd on past experience and recent observation..
\npeople try all means.
you are citing a possibility… and then spam it all over the forum. why do you need to say the same thing in the looking for a team thread?
", "votes": -3, "replies": [] }, { "id": 1184540, "author_name": "Jaideep", "author_url": "", "post_date": "2021-02-03T15:09:31.337000", "content": "it was nothing intentional..just was pointing to Admin..
", "votes": 0, "replies": [] }, { "id": 1184543, "author_name": "Proletheus", "author_url": "", "post_date": "2021-02-03T15:12:10.980000", "content": "what..? are you in a haze? how can writing multiple comments not be intentional
", "votes": -3, "replies": [] }, { "id": 1184588, "author_name": "Jaideep", "author_url": "", "post_date": "2021-02-03T15:47:41.157000", "content": "may be yes..i deleted now you can go and check…i ealized after posting better i put it as topic for awareness and alertness
", "votes": 0, "replies": [] } ] }, { "id": 1188306, "author_name": "Zekun", "author_url": "", "post_date": "2021-02-06T05:58:08.567000", "content": "You shuld @ Kaggle starf.
", "votes": 0, "replies": [] }, { "id": 1185150, "author_name": "HAaHAa", "author_url": "", "post_date": "2021-02-04T01:46:00.293000", "content": "Aha,This is another interesting case.
", "votes": 0, "replies": [] }, { "id": 1184381, "author_name": "", "author_url": "", "post_date": "2021-02-03T14:07:26.493000", "content": "", "votes": 1, "replies": [ { "id": 1184418, "author_name": "", "author_url": "", "post_date": "2021-02-03T14:17:21.057000", "content": "", "votes": 0, "replies": [] }, { "id": 1184421, "author_name": "", "author_url": "", "post_date": "2021-02-03T14:17:47.190000", "content": "", "votes": 0, "replies": [] }, { "id": 1184423, "author_name": "Alexander Riedel", "author_url": "", "post_date": "2021-02-03T14:18:13.733000", "content": "haha yes thanks :D
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "1184299": "I am not writting this because i am not in gold zone ,but just sharing some observation by my one of team member.\n\nFew days ago some thing bit strange was found, lb gold zone was almost deadlocked at 90.6 but suddenly 4 to 5 teams in no time came simultaneously to score at 90.7 and getting into gold range. Generally this happens if some one shares a public notebook capable of getting such score or some useful tips. I did a dig dint find any such notebook or tips published in that time frame. \nSuspected range of fishy score position in lb\n**from 6 to 17. **\nI may be wrong about certain folks so please pardon me its not for you at all. . From past revelations of this sort some people might have learnt so they may be ensuring that just by making few sub they dont get very high in lb so they might have done regular submissions like others. \n\n\nOnce again i pardon if this is not a case. I just shared in case of such possibility upon finding unsual trend which may be coincidental also. \nSo upto kaggle team for investigation about this\n\n \nRegards\nJaideep", "1184550": "here are two examples from my experience\n\n1. in one non-kaggle competition a team from in-the-money zone offered to merge. My position was very close to money so I felt we can pool it off together and I gladly accepted. All our ensembles were bad (a bit better than my score but worse than theirs). Our team improved a little thanks to what i contributed. Later, pandas.DataFrame.corr() revieled that all ensemble submissions were actually slightly adjusted submissions from my models. Also, I was surprised with the performance of late comers, as some managed to jump over us in 2-3 subs :).\n\n2. On Alaska2 I was in gold zone 10 days before the end. A kaggle master contacted me and offered a merge. She also mentioned 8xV100. We said no. In the next two days I saw her merge with someone above us and suddenly quite a few participants jumped over us within hours, but also over them. That competition was very stable until that point and I felt something bad is going on. I did not have proof, so I stayed silent. I contacted her previous teammates and they all have the same story \"Yeah, she is the one with a lot of GPUs. We merged, but then she disappeared. She was very busy.\" She would activate only when other teammates made improvements. \n\nSo there you have it.\n\nIs there a way to handle this. I don't see it and I accept such reality. \n\nHere just to learn. Good luck to everyone.", "1185696": "Kaggle should give score upto 4 or 5 decimal point like 0.9046 or .90326. It will be easy to figure out which all are getting exact same score.", "1184481": "This is code competition. Hopefully Kaggle will check, at least, all Gold inference notebooks on private LB and request the corresponding training code if necessary.\n\nI would even agree we make public chosen inference submission notebooks for all the LB. \n\ncc @juliaelliott @sohier", "1184325": "Something was going on [here](https://www.kaggle.com/c/rfcx-species-audio-detection/discussion/216003) too.", "1184343": "I suspect that Private sharing is common in Chinese Competition Zones.", "1185250": "I find see some consecutive ranked lb teams with same score have started to merge now ... so is my doubt really look witch hunt to some folks ..\n", "1184431": "Has something to do with [this](https://www.kaggle.com/c/hubmap-kidney-segmentation/discussion/211212) ? \n\nThe author says the guy who wanted to pay him, is also currently at the top of this LB\n\nMay be the guy managed to pay another one at the top here. ", "1184306": "@juliaelliott fyi", "1184321": "can these witchhunts please stop. take this up with the mods", "1188306": "You shuld @ Kaggle starf.", "1185150": "Aha,This is another interesting [case](https://www.kaggle.com/c/global-wheat-detection/discussion/161296).\n", "1184381": "" } }