Hi, after my first experience on a big kaggle competition I'm a bit confused as to why Kagglers are allowed to use other people's submission files. Isn't it kind of defeating the point of the learning and competition ethos to just collect a whole load of public submission files and then blend them together?
\nI know on this competition the 0.96 public LB file that dropped in the last week turned out to do badly on the private LB but it could have been otherwise couldn't it and then hard workers, trying out ideas and struggling with code and getting models to train in the resource constraints are being cheated aren't they?
", "messages": [ { "id": "976157", "postDate": "08/18/2020 17:10:13", "content": "Hi, after my first experience on a big kaggle competition I'm a bit confused as to why Kagglers are allowed to use other people's submission files. Isn't it kind of defeating the point of the learning and competition ethos to just collect a whole load of public submission files and then blend them together?
\nI know on this competition the 0.96 public LB file that dropped in the last week turned out to do badly on the private LB but it could have been otherwise couldn't it and then hard workers, trying out ideas and struggling with code and getting models to train in the resource constraints are being cheated aren't they?
", "rawMarkdown": "Hi, after my first experience on a big kaggle competition I'm a bit confused as to why Kagglers are allowed to use other people's submission files. Isn't it kind of defeating the point of the learning and competition ethos to just collect a whole load of public submission files and then blend them together?\n\nI know on this competition the 0.96 public LB file that dropped in the last week turned out to do badly on the private LB but it could have been otherwise couldn't it and then hard workers, trying out ideas and struggling with code and getting models to train in the resource constraints are being cheated aren't they?", "votes": null }, { "id": "976166", "postDate": "08/18/2020 17:23:26", "content": "It's (near) effortless to blend public kernels. If it was an effective strategy (and in some comps maybe it is) kagglers could make a public blend as one of their submissions. The fact that top kagglers won't routinely include public blends is testament to the poor performance of these blends.
\nEach individual must choose what they get out of Kaggle, for those who are here to learn these blends do little to inhibit their learning. They are frustrating but if you don't like them simply ignore them and congratulate them on the odd times that they do well.
", "rawMarkdown": "It's (near) effortless to blend public kernels. If it was an effective strategy (and in some comps maybe it is) kagglers could make a public blend as one of their submissions. The fact that top kagglers won't routinely include public blends is testament to the poor performance of these blends. \n\nEach individual must choose what they get out of Kaggle, for those who are here to learn these blends do little to inhibit their learning. They are frustrating but if you don't like them simply ignore them and congratulate them on the odd times that they do well.", "votes": null }, { "id": "976282", "postDate": "08/18/2020 19:06:44", "content": "One can fork public notebook, run it and make his own submission.csv with no effort. So just disallowing use of others csvs didn't resolve the issue.
", "rawMarkdown": "One can fork public notebook, run it and make his own submission.csv with no effort. So just disallowing use of others csvs didn't resolve the issue.", "votes": null } ], "comments": [ { "id": 976166, "author_name": "fchmiel", "author_url": "", "post_date": "08/18/2020 17:23:26", "content": "It's (near) effortless to blend public kernels. If it was an effective strategy (and in some comps maybe it is) kagglers could make a public blend as one of their submissions. The fact that top kagglers won't routinely include public blends is testament to the poor performance of these blends.
\nEach individual must choose what they get out of Kaggle, for those who are here to learn these blends do little to inhibit their learning. They are frustrating but if you don't like them simply ignore them and congratulate them on the odd times that they do well.
", "votes": null, "replies": [] }, { "id": 976282, "author_name": "bfishh", "author_url": "", "post_date": "08/18/2020 19:06:44", "content": "One can fork public notebook, run it and make his own submission.csv with no effort. So just disallowing use of others csvs didn't resolve the issue.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "976157": "Hi, after my first experience on a big kaggle competition I'm a bit confused as to why Kagglers are allowed to use other people's submission files. Isn't it kind of defeating the point of the learning and competition ethos to just collect a whole load of public submission files and then blend them together?\n\nI know on this competition the 0.96 public LB file that dropped in the last week turned out to do badly on the private LB but it could have been otherwise couldn't it and then hard workers, trying out ideas and struggling with code and getting models to train in the resource constraints are being cheated aren't they?", "976166": "It's (near) effortless to blend public kernels. If it was an effective strategy (and in some comps maybe it is) kagglers could make a public blend as one of their submissions. The fact that top kagglers won't routinely include public blends is testament to the poor performance of these blends. \n\nEach individual must choose what they get out of Kaggle, for those who are here to learn these blends do little to inhibit their learning. They are frustrating but if you don't like them simply ignore them and congratulate them on the odd times that they do well.", "976282": "One can fork public notebook, run it and make his own submission.csv with no effort. So just disallowing use of others csvs didn't resolve the issue." }, "source": "meta" }