Would not it be fun to run a contest where Kagglers get a shot at detecting fingerprints of blends with rogue submission such as -2.163 from this contest?
\n\n*Update: Not meant as a witch hunting contest. The contest could be based on a synthetic dataset that mimics the scenario of a rogue submission used to make unauthorized blends.
", "messages": [ { "id": "605543", "postDate": "08/22/2019 13:24:22", "content": "Would not it be fun to run a contest where Kagglers get a shot at detecting fingerprints of blends with rogue submission such as -2.163 from this contest?
\n\n*Update: Not meant as a witch hunting contest. The contest could be based on a synthetic dataset that mimics the scenario of a rogue submission used to make unauthorized blends.
", "rawMarkdown": "Would not it be fun to run a contest where Kagglers get a shot at detecting fingerprints of blends with rogue submission such as -2.163 from this contest?\n\n***Update:** Not meant as a witch hunting contest. The contest could be based on a synthetic dataset that mimics the scenario of a rogue submission used to make unauthorized blends.", "votes": null }, { "id": "605555", "postDate": "08/22/2019 13:42:31", "content": "The good news is that you only need to care about those with a LB score better than that. And given the influence of blending drops quickly with a difference in score, it limits the number of candidate teams to 20 or so.
", "rawMarkdown": "The good news is that you only need to care about those with a LB score better than that. And given the influence of blending drops quickly with a difference in score, it limits the number of candidate teams to 20 or so.", "votes": null }, { "id": "605557", "postDate": "08/22/2019 13:48:24", "content": "You know, I'd suggest focusing on improving your own score instead of witch hunting.
", "rawMarkdown": "You know, I'd suggest focusing on improving your own score instead of witch hunting.", "votes": null }, { "id": "605563", "postDate": "08/22/2019 13:53:43", "content": "Almost anybody can take advantage of such high scoring submission. Maybe not -3, but -2.5 could still benefit. Basically, the whole LB, with a couple of exceptions, could be affected.
", "rawMarkdown": "Almost anybody can take advantage of such high scoring submission. Maybe not -3, but -2.5 could still benefit. Basically, the whole LB, with a couple of exceptions, could be affected.", "votes": null }, { "id": "605566", "postDate": "08/22/2019 13:55:51", "content": "I am not talking about witch hunting or naming names. The contest could be based on synthetic data mimicking what happened here.
", "rawMarkdown": "I am not talking about witch hunting or naming names. The contest could be based on synthetic data mimicking what happened here.", "votes": null }, { "id": "605569", "postDate": "08/22/2019 13:59:16", "content": "I'll make a kernel after competition ends with a couple of blends and the leaked file, you can have a go at guessing which used it.
", "rawMarkdown": "I'll make a kernel after competition ends with a couple of blends and the leaked file, you can have a go at guessing which used it.", "votes": null }, { "id": "605585", "postDate": "08/22/2019 14:12:52", "content": "\n\n\n-2.5 could still benefit.
\n
Not from my experience. But you obviously have other source of info here.
", "rawMarkdown": "> -2.5 could still benefit. \n\nNot from my experience. But you obviously have other source of info here.", "votes": null }, { "id": "605588", "postDate": "08/22/2019 14:15:28", "content": "just a wild guess based on how crazily well subs blend ~ -1.8 level.
", "rawMarkdown": "just a wild guess based on how crazily well subs blend ~ -1.8 level.", "votes": null }, { "id": "606251", "postDate": "08/23/2019 11:20:46", "content": "I hope kaggle has some some logs,whenever someone downloads a file from a kernel output. If that is being captured, then it is easy to track who downloaded the file and if they used it.
", "rawMarkdown": "I hope kaggle has some some logs,whenever someone downloads a file from a kernel output. If that is being captured, then it is easy to track who downloaded the file and if they used it.", "votes": null }, { "id": "606299", "postDate": "08/23/2019 11:56:14", "content": "Downloading a file and using it are different things. Expecially considering that a warning not to use the file appeared several hours after the kernel.\nP. S. I didn't even see that kernel - just in case.
", "rawMarkdown": "Downloading a file and using it are different things. Expecially considering that a warning not to use the file appeared several hours after the kernel.\nP. S. I didn't even see that kernel - just in case.", "votes": null }, { "id": "606359", "postDate": "08/23/2019 13:33:58", "content": "\n\n\nP. S. I didn't even see that kernel - just in case.
\n
Nice try too ;)
", "rawMarkdown": "> P. S. I didn't even see that kernel - just in case.\n\nNice try too ;)", "votes": null }, { "id": "606412", "postDate": "08/23/2019 14:45:03", "content": "Well if it is possible to Identify the users who downloaded it(if not available the viewed it) then one can check if the users had a huge jump in their submissions score\nDisallowing that submission or updating the rules so that people using it will be removed from LB if it is being used as final submission can be done right ? It is just that it is some more unplanned work for kaggle probably.
", "rawMarkdown": "Well if it is possible to Identify the users who downloaded it(if not available the viewed it) then one can check if the users had a huge jump in their submissions score\nDisallowing that submission or updating the rules so that people using it will be removed from LB if it is being used as final submission can be done right ? It is just that it is some more unplanned work for kaggle probably.", "votes": null }, { "id": "606423", "postDate": "08/23/2019 14:59:02", "content": "If you read their post thats exactly what they are doing..
", "rawMarkdown": "If you read their post thats exactly what they are doing..", "votes": null }, { "id": "606439", "postDate": "08/23/2019 15:26:21", "content": "Oh, my bad didn't read that 😅
", "rawMarkdown": "Oh, my bad didn't read that 😅", "votes": null } ], "comments": [ { "id": 605555, "author_name": "cpmpml", "author_url": "", "post_date": "08/22/2019 13:42:31", "content": "The good news is that you only need to care about those with a LB score better than that. And given the influence of blending drops quickly with a difference in score, it limits the number of candidate teams to 20 or so.
", "votes": null, "replies": [ { "id": 605563, "author_name": "stocks", "author_url": "", "post_date": "08/22/2019 13:53:43", "content": "Almost anybody can take advantage of such high scoring submission. Maybe not -3, but -2.5 could still benefit. Basically, the whole LB, with a couple of exceptions, could be affected.
", "votes": null, "replies": [] }, { "id": 605585, "author_name": "cpmpml", "author_url": "", "post_date": "08/22/2019 14:12:52", "content": "\n\n\n-2.5 could still benefit.
\n
Not from my experience. But you obviously have other source of info here.
", "votes": null, "replies": [] }, { "id": 605588, "author_name": "stocks", "author_url": "", "post_date": "08/22/2019 14:15:28", "content": "just a wild guess based on how crazily well subs blend ~ -1.8 level.
", "votes": null, "replies": [] }, { "id": 606251, "author_name": "shubhinderrana", "author_url": "", "post_date": "08/23/2019 11:20:46", "content": "I hope kaggle has some some logs,whenever someone downloads a file from a kernel output. If that is being captured, then it is easy to track who downloaded the file and if they used it.
", "votes": null, "replies": [] }, { "id": 606299, "author_name": "artgor", "author_url": "", "post_date": "08/23/2019 11:56:14", "content": "Downloading a file and using it are different things. Expecially considering that a warning not to use the file appeared several hours after the kernel.\nP. S. I didn't even see that kernel - just in case.
", "votes": null, "replies": [] }, { "id": 606359, "author_name": "cpmpml", "author_url": "", "post_date": "08/23/2019 13:33:58", "content": "\n\n\nP. S. I didn't even see that kernel - just in case.
\n
Nice try too ;)
", "votes": null, "replies": [] }, { "id": 606412, "author_name": "shubhinderrana", "author_url": "", "post_date": "08/23/2019 14:45:03", "content": "Well if it is possible to Identify the users who downloaded it(if not available the viewed it) then one can check if the users had a huge jump in their submissions score\nDisallowing that submission or updating the rules so that people using it will be removed from LB if it is being used as final submission can be done right ? It is just that it is some more unplanned work for kaggle probably.
", "votes": null, "replies": [] }, { "id": 606423, "author_name": "julianmukaj", "author_url": "", "post_date": "08/23/2019 14:59:02", "content": "If you read their post thats exactly what they are doing..
", "votes": null, "replies": [] }, { "id": 606439, "author_name": "shubhinderrana", "author_url": "", "post_date": "08/23/2019 15:26:21", "content": "Oh, my bad didn't read that 😅
", "votes": null, "replies": [] } ] }, { "id": 605557, "author_name": "artgor", "author_url": "", "post_date": "08/22/2019 13:48:24", "content": "You know, I'd suggest focusing on improving your own score instead of witch hunting.
", "votes": null, "replies": [ { "id": 605566, "author_name": "stocks", "author_url": "", "post_date": "08/22/2019 13:55:51", "content": "I am not talking about witch hunting or naming names. The contest could be based on synthetic data mimicking what happened here.
", "votes": null, "replies": [] } ] }, { "id": 605569, "author_name": "julianmukaj", "author_url": "", "post_date": "08/22/2019 13:59:16", "content": "I'll make a kernel after competition ends with a couple of blends and the leaked file, you can have a go at guessing which used it.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "605543": "Would not it be fun to run a contest where Kagglers get a shot at detecting fingerprints of blends with rogue submission such as -2.163 from this contest?\n\n***Update:** Not meant as a witch hunting contest. The contest could be based on a synthetic dataset that mimics the scenario of a rogue submission used to make unauthorized blends.", "605555": "The good news is that you only need to care about those with a LB score better than that. And given the influence of blending drops quickly with a difference in score, it limits the number of candidate teams to 20 or so.", "605557": "You know, I'd suggest focusing on improving your own score instead of witch hunting.", "605563": "Almost anybody can take advantage of such high scoring submission. Maybe not -3, but -2.5 could still benefit. Basically, the whole LB, with a couple of exceptions, could be affected.", "605566": "I am not talking about witch hunting or naming names. The contest could be based on synthetic data mimicking what happened here.", "605569": "I'll make a kernel after competition ends with a couple of blends and the leaked file, you can have a go at guessing which used it.", "605585": "> -2.5 could still benefit. \n\nNot from my experience. But you obviously have other source of info here.", "605588": "just a wild guess based on how crazily well subs blend ~ -1.8 level.", "606251": "I hope kaggle has some some logs,whenever someone downloads a file from a kernel output. If that is being captured, then it is easy to track who downloaded the file and if they used it.", "606299": "Downloading a file and using it are different things. Expecially considering that a warning not to use the file appeared several hours after the kernel.\nP. S. I didn't even see that kernel - just in case.", "606359": "> P. S. I didn't even see that kernel - just in case.\n\nNice try too ;)", "606412": "Well if it is possible to Identify the users who downloaded it(if not available the viewed it) then one can check if the users had a huge jump in their submissions score\nDisallowing that submission or updating the rules so that people using it will be removed from LB if it is being used as final submission can be done right ? It is just that it is some more unplanned work for kaggle probably.", "606423": "If you read their post thats exactly what they are doing..", "606439": "Oh, my bad didn't read that 😅" }, "source": "meta" }