Maybe I'm mistaken but I think that changing the type of current competition to \"kernels only\" will solve problem with leakage. Only kernels, no external data sources -> no leakage. Any thoughts about it?
", "messages": [ { "id": 680984, "postDate": "2019-11-25T13:34:31.047Z", "content": "People are going to copy paste leaks instead of reading them from csv file
", "rawMarkdown": "People are going to copy paste leaks instead of reading them from csv file", "votes": 6, "replies": [ { "id": 682162, "postDate": "2019-11-27T02:55:15.860Z", "content": "Do you have any suggestions to save this competition? @gunesevitan
", "rawMarkdown": "Do you have any suggestions to save this competition? @gunesevitan " }, { "id": 682218, "postDate": "2019-11-27T05:11:32.530Z", "content": "I already posted my suggestion here\nhttps://www.kaggle.com/c/ashrae-energy-prediction/discussion/118375
", "rawMarkdown": "I already posted my suggestion here\nhttps://www.kaggle.com/c/ashrae-energy-prediction/discussion/118375" } ] }, { "id": 680929, "postDate": "2019-11-25T12:29:41.760Z", "content": "I think that it is not the solution, because you have the access to test public/private data and can use it (not directly) while training the model even in kernels-only competition (if you assume that any site with leakage will remain in final test dataset).
", "rawMarkdown": "I think that it is not the solution, because you have the access to test public/private data and can use it (not directly) while training the model even in kernels-only competition (if you assume that any site with leakage will remain in final test dataset).", "votes": 1, "replies": [ { "id": 680948, "postDate": "2019-11-25T12:55:25.730Z", "content": "Do you mean hard coding of this dataset?
", "rawMarkdown": "Do you mean hard coding of this dataset?" }, { "id": 680959, "postDate": "2019-11-25T13:08:31.617Z", "rawMarkdown": "", "isDeleted": true }, { "id": 680987, "postDate": "2019-11-25T13:38:07.923Z", "content": "I mean any tuning that is based on leakage info, including indirect usage.
", "rawMarkdown": "I mean any tuning that is based on leakage info, including indirect usage.", "votes": 1 } ] }, { "id": 680888, "postDate": "2019-11-25T11:17:22.753Z", "content": "Maybe I'm mistaken but I think that changing the type of current competition to \"kernels only\" will solve problem with leakage. Only kernels, no external data sources -> no leakage. Any thoughts about it?
", "rawMarkdown": "Maybe I'm mistaken but I think that changing the type of current competition to \"kernels only\" will solve problem with leakage. Only kernels, no external data sources -> no leakage. Any thoughts about it?", "votes": 1 }, { "id": 682592, "postDate": "2019-11-27T16:36:41.910Z", "content": "I think it might be able to fix the issue to a high degree. If kaggle can somehow block web scraping libraries in their notebooks, most people would compete fairly. Kaggle can then reset the leaderboard, and then only people with enough time to copy each value of the test set to the kernel will have a good score, which I am sure almost no one has.
", "rawMarkdown": "I think it might be able to fix the issue to a high degree. If kaggle can somehow block web scraping libraries in their notebooks, most people would compete fairly. Kaggle can then reset the leaderboard, and then only people with enough time to copy each value of the test set to the kernel will have a good score, which I am sure almost no one has." } ], "comments": [ { "id": 680984, "author_name": "Gunes Evitan", "author_url": "", "post_date": "2019-11-25T13:34:31.047000", "content": "People are going to copy paste leaks instead of reading them from csv file
", "votes": 6, "replies": [ { "id": 682162, "author_name": "Shahules", "author_url": "", "post_date": "2019-11-27T02:55:15.860000", "content": "Do you have any suggestions to save this competition? @gunesevitan
", "votes": 0, "replies": [] }, { "id": 682218, "author_name": "Gunes Evitan", "author_url": "", "post_date": "2019-11-27T05:11:32.530000", "content": "I already posted my suggestion here\nhttps://www.kaggle.com/c/ashrae-energy-prediction/discussion/118375
", "votes": 0, "replies": [] } ] }, { "id": 680929, "author_name": "Dmitry Labazkin", "author_url": "", "post_date": "2019-11-25T12:29:41.760000", "content": "I think that it is not the solution, because you have the access to test public/private data and can use it (not directly) while training the model even in kernels-only competition (if you assume that any site with leakage will remain in final test dataset).
", "votes": 1, "replies": [ { "id": 680948, "author_name": "Kostiantyn Isaienkov", "author_url": "", "post_date": "2019-11-25T12:55:25.730000", "content": "Do you mean hard coding of this dataset?
", "votes": 0, "replies": [] }, { "id": 680959, "author_name": "", "author_url": "", "post_date": "2019-11-25T13:08:31.617000", "content": "", "votes": 0, "replies": [] }, { "id": 680987, "author_name": "Dmitry Labazkin", "author_url": "", "post_date": "2019-11-25T13:38:07.923000", "content": "I mean any tuning that is based on leakage info, including indirect usage.
", "votes": 1, "replies": [] } ] }, { "id": 682592, "author_name": "Ayush Goel", "author_url": "", "post_date": "2019-11-27T16:36:41.910000", "content": "I think it might be able to fix the issue to a high degree. If kaggle can somehow block web scraping libraries in their notebooks, most people would compete fairly. Kaggle can then reset the leaderboard, and then only people with enough time to copy each value of the test set to the kernel will have a good score, which I am sure almost no one has.
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "680984": "People are going to copy paste leaks instead of reading them from csv file", "680929": "I think that it is not the solution, because you have the access to test public/private data and can use it (not directly) while training the model even in kernels-only competition (if you assume that any site with leakage will remain in final test dataset).", "680888": "Maybe I'm mistaken but I think that changing the type of current competition to \"kernels only\" will solve problem with leakage. Only kernels, no external data sources -> no leakage. Any thoughts about it?", "682592": "I think it might be able to fix the issue to a high degree. If kaggle can somehow block web scraping libraries in their notebooks, most people would compete fairly. Kaggle can then reset the leaderboard, and then only people with enough time to copy each value of the test set to the kernel will have a good score, which I am sure almost no one has." } }