Just realize one thing. The competition allows any freely and publicly available external datasets but the test data is masked with timestamp. So how can we apply external data on test data?
", "messages": [ { "id": "3209391", "postDate": "05/25/2025 17:48:17", "content": "Just realize one thing. The competition allows any freely and publicly available external datasets but the test data is masked with timestamp. So how can we apply external data on test data?
", "rawMarkdown": "Just realize one thing. The competition allows *any freely and publicly available external datasets* but the test data is masked with timestamp. So how can we apply external data on test data?", "votes": null }, { "id": "3209419", "postDate": "05/25/2025 18:59:36", "content": "Perhaps the external data can be used for predicting the timestamp 😀 @yw2735
", "rawMarkdown": "Perhaps the external data can be used for predicting the timestamp 😀 @yw2735", "votes": null }, { "id": "3209612", "postDate": "05/26/2025 04:55:20", "content": "That's insane 🤣
", "rawMarkdown": "That's insane 🤣", "votes": null }, { "id": "3218665", "postDate": "06/06/2025 14:27:46", "content": "Is there a way we can clarify the reasoning behind the masking/shuffling of time-stamps with the host? Presumably there's a reason they chose this over online data-feeding (as in the JS comp for example), maybe to encourage us to produce a very low-latency row->row regression model? Does anyone know if/when such a model would ever be used in production?
\nI'm doing this competition to learn about modeling in the QR space, so I want to avoid doing anything unrealistic like reconstructing time-stamps and therefore leaking future data into past. But I also don't understand how you can use future data in any other way, if the time-stamps are masked, and rows are shuffled ðŸ«
\nBasically, it'd be nice to actually know the motivation behind what we're doing here.
", "rawMarkdown": "Is there a way we can clarify the reasoning behind the masking/shuffling of time-stamps with the host? Presumably there's a reason they chose this over online data-feeding (as in the JS comp for example), maybe to encourage us to produce a very low-latency row->row regression model? Does anyone know if/when such a model would ever be used in production?\n\nI'm doing this competition to learn about modeling in the QR space, so I want to avoid doing anything unrealistic like reconstructing time-stamps and therefore leaking future data into past. But I also don't understand how you can use future data in any other way, if the time-stamps are masked, and rows are shuffled 🫠\n\nBasically, it'd be nice to actually know the motivation behind what we're doing here.", "votes": null }, { "id": "3218675", "postDate": "06/06/2025 14:53:09", "content": "I have no idea why they choose community competition with offline evaluation. But the biggest problem of not masking timestamp in test data is that anyone can easily use future features to predict target now and will cause very high scores.
", "rawMarkdown": "I have no idea why they choose community competition with offline evaluation. But the biggest problem of not masking timestamp in test data is that anyone can easily use future features to predict target now and will cause very high scores.", "votes": null } ], "comments": [ { "id": 3209419, "author_name": "ravi20076", "author_url": "", "post_date": "05/25/2025 18:59:36", "content": "Perhaps the external data can be used for predicting the timestamp 😀 @yw2735
", "votes": null, "replies": [ { "id": 3209612, "author_name": "yw2735", "author_url": "", "post_date": "05/26/2025 04:55:20", "content": "That's insane 🤣
", "votes": null, "replies": [] } ] }, { "id": 3218665, "author_name": "sarahjeffreson", "author_url": "", "post_date": "06/06/2025 14:27:46", "content": "Is there a way we can clarify the reasoning behind the masking/shuffling of time-stamps with the host? Presumably there's a reason they chose this over online data-feeding (as in the JS comp for example), maybe to encourage us to produce a very low-latency row->row regression model? Does anyone know if/when such a model would ever be used in production?
\nI'm doing this competition to learn about modeling in the QR space, so I want to avoid doing anything unrealistic like reconstructing time-stamps and therefore leaking future data into past. But I also don't understand how you can use future data in any other way, if the time-stamps are masked, and rows are shuffled ðŸ«
\nBasically, it'd be nice to actually know the motivation behind what we're doing here.
", "votes": null, "replies": [ { "id": 3218675, "author_name": "yw2735", "author_url": "", "post_date": "06/06/2025 14:53:09", "content": "I have no idea why they choose community competition with offline evaluation. But the biggest problem of not masking timestamp in test data is that anyone can easily use future features to predict target now and will cause very high scores.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "3209391": "Just realize one thing. The competition allows *any freely and publicly available external datasets* but the test data is masked with timestamp. So how can we apply external data on test data?", "3209419": "Perhaps the external data can be used for predicting the timestamp 😀 @yw2735", "3209612": "That's insane 🤣", "3218665": "Is there a way we can clarify the reasoning behind the masking/shuffling of time-stamps with the host? Presumably there's a reason they chose this over online data-feeding (as in the JS comp for example), maybe to encourage us to produce a very low-latency row->row regression model? Does anyone know if/when such a model would ever be used in production?\n\nI'm doing this competition to learn about modeling in the QR space, so I want to avoid doing anything unrealistic like reconstructing time-stamps and therefore leaking future data into past. But I also don't understand how you can use future data in any other way, if the time-stamps are masked, and rows are shuffled 🫠\n\nBasically, it'd be nice to actually know the motivation behind what we're doing here.", "3218675": "I have no idea why they choose community competition with offline evaluation. But the biggest problem of not masking timestamp in test data is that anyone can easily use future features to predict target now and will cause very high scores." }, "source": "meta" }