Submitted a public notebook twice - the exact code, it scored 0.262 and 0.286. Is there too much randomness? Another competition where a seed can change the results drastically? A shakeup incoming?
", "messages": [ { "id": 3205116, "postDate": "2025-05-19T10:39:43.970Z", "content": "Submitted a public notebook twice - the exact code, it scored 0.262 and 0.286. Is there too much randomness? Another competition where a seed can change the results drastically? A shakeup incoming?
", "rawMarkdown": "Submitted a public notebook twice - the exact code, it scored 0.262 and 0.286. Is there too much randomness? Another competition where a seed can change the results drastically? A shakeup incoming? ", "votes": 1 }, { "id": 3205166, "postDate": "2025-05-19T11:41:48Z", "content": "I‘m wondering why an ensemble of 2 models is worse than both.
", "rawMarkdown": "I‘m wondering why an ensemble of 2 models is worse than both.", "votes": 2, "replies": [ { "id": 3205170, "postDate": "2025-05-19T11:48:43.360Z", "content": "Same doubt, this is the first time witnessing something like that! Shouldn't it be the other way round 🤨
", "rawMarkdown": "Same doubt, this is the first time witnessing something like that! Shouldn't it be the other way round 🤨" }, { "id": 3205229, "postDate": "2025-05-19T13:55:13.867Z", "content": "You can't ensemble model like a traditional data model. Think about the characteristics of this competition and take a more appropriate approach to ensemble.
", "rawMarkdown": " You can't ensemble model like a traditional data model. Think about the characteristics of this competition and take a more appropriate approach to ensemble.", "replies": [ { "id": 3205246, "postDate": "2025-05-19T14:17:12.847Z", "content": "You mean ensembling via traditional ways ain't gonna give better results? Even when they're pretty diverse.
", "rawMarkdown": "You mean ensembling via traditional ways ain't gonna give better results? Even when they're pretty diverse." }, { "id": 3205252, "postDate": "2025-05-19T14:39:25.563Z", "content": "\n\nYou can't ensemble model like a traditional data model.
\n
It is a biological issue, not just a statistical one.
\n\n\nToo much Randomness?
\n
Maybe it is the reason, and the randomness varies within a defined range. Find a way to break through it.
", "rawMarkdown": ">You can't ensemble model like a traditional data model. \n\nIt is a biological issue, not just a statistical one.\n\n>Too much Randomness?\n\nMaybe it is [the reason](https://www.kaggle.com/code/youhanlee/boltz-1-inference-submission/comments#3195235), and the randomness varies within a defined range. Find a way to break through it.\n\n\n\n", "votes": 1 }, { "id": 3205638, "postDate": "2025-05-20T07:41:50.427Z", "content": "Simple voting or weighted averaging methods do not necessarily improve the results, even when there are significant differences between models. Switching to an ensemble mothod that better matches the requirements can effectively boost scores, even if the differences between models are small.
", "rawMarkdown": "Simple voting or weighted averaging methods do not necessarily improve the results, even when there are significant differences between models. Switching to an ensemble mothod that better matches the requirements can effectively boost scores, even if the differences between models are small." } ] }, { "id": 3207135, "postDate": "2025-05-22T09:31:05.920Z", "content": "I'm not sure. As far as I know, there seems to be no theory that shows that the fusion model must be a positive improvement.
", "rawMarkdown": "I'm not sure. As far as I know, there seems to be no theory that shows that the fusion model must be a positive improvement.", "replies": [ { "id": 3207685, "postDate": "2025-05-23T06:02:09.407Z", "content": "I just combine two model and then can achieve 0.380 score, based on protenix to improve.
", "rawMarkdown": "I just combine two model and then can achieve 0.380 score, based on protenix to improve.", "votes": 1 }, { "id": 3207693, "postDate": "2025-05-23T06:21:30.940Z", "content": "I mean that integration may improve or decrease, and there is no research that guarantees that it will improve.
", "rawMarkdown": "I mean that integration may improve or decrease, and there is no research that guarantees that it will improve." }, { "id": 3207698, "postDate": "2025-05-23T06:28:24.800Z", "content": "That's indeed the case; there's no good solution. Perhaps the private leaderboard rankings will shake once they're released
", "rawMarkdown": "That's indeed the case; there's no good solution. Perhaps the private leaderboard rankings will shake once they're released", "votes": 1 }, { "id": 3207781, "postDate": "2025-05-23T07:55:34.117Z", "rawMarkdown": "", "isDeleted": true } ] } ] } ], "comments": [ { "id": 3205166, "author_name": "yuanzhe zhou", "author_url": "", "post_date": "2025-05-19T11:41:48", "content": "I‘m wondering why an ensemble of 2 models is worse than both.
", "votes": 2, "replies": [ { "id": 3205170, "author_name": "AC", "author_url": "", "post_date": "2025-05-19T11:48:43.360000", "content": "Same doubt, this is the first time witnessing something like that! Shouldn't it be the other way round 🤨
", "votes": 0, "replies": [] }, { "id": 3205229, "author_name": "Tan_T64", "author_url": "", "post_date": "2025-05-19T13:55:13.867000", "content": "You can't ensemble model like a traditional data model. Think about the characteristics of this competition and take a more appropriate approach to ensemble.
", "votes": 0, "replies": [ { "id": 3205246, "author_name": "AC", "author_url": "", "post_date": "2025-05-19T14:17:12.847000", "content": "You mean ensembling via traditional ways ain't gonna give better results? Even when they're pretty diverse.
", "votes": 0, "replies": [] }, { "id": 3205252, "author_name": "Timmy Juicehouse", "author_url": "", "post_date": "2025-05-19T14:39:25.563000", "content": "\n\nYou can't ensemble model like a traditional data model.
\n
It is a biological issue, not just a statistical one.
\n\n\nToo much Randomness?
\n
Maybe it is the reason, and the randomness varies within a defined range. Find a way to break through it.
", "votes": 1, "replies": [] }, { "id": 3205638, "author_name": "Tan_T64", "author_url": "", "post_date": "2025-05-20T07:41:50.427000", "content": "Simple voting or weighted averaging methods do not necessarily improve the results, even when there are significant differences between models. Switching to an ensemble mothod that better matches the requirements can effectively boost scores, even if the differences between models are small.
", "votes": 0, "replies": [] } ] }, { "id": 3207135, "author_name": "DaoHe Liu", "author_url": "", "post_date": "2025-05-22T09:31:05.920000", "content": "I'm not sure. As far as I know, there seems to be no theory that shows that the fusion model must be a positive improvement.
", "votes": 0, "replies": [ { "id": 3207685, "author_name": "", "author_url": "", "post_date": "2025-05-23T06:02:09.407000", "content": "I just combine two model and then can achieve 0.380 score, based on protenix to improve.
", "votes": 1, "replies": [] }, { "id": 3207693, "author_name": "DaoHe Liu", "author_url": "", "post_date": "2025-05-23T06:21:30.940000", "content": "I mean that integration may improve or decrease, and there is no research that guarantees that it will improve.
", "votes": 0, "replies": [] }, { "id": 3207698, "author_name": "", "author_url": "", "post_date": "2025-05-23T06:28:24.800000", "content": "That's indeed the case; there's no good solution. Perhaps the private leaderboard rankings will shake once they're released
", "votes": 1, "replies": [] }, { "id": 3207781, "author_name": "", "author_url": "", "post_date": "2025-05-23T07:55:34.117000", "content": "", "votes": 0, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "3205116": "Submitted a public notebook twice - the exact code, it scored 0.262 and 0.286. Is there too much randomness? Another competition where a seed can change the results drastically? A shakeup incoming? ", "3205166": "I‘m wondering why an ensemble of 2 models is worse than both." } }