Hello, I'm trying to beat the VFold_human_expert_baseline score, and I understand that we need to consider the temporal cutoff date (which is safely before 2022-05-27).
\nIf I have correctly understood the rule, it means that I should not use any training data that comes after this temporal cutoff date.
\nGiven that, there are quite a few RNA sequences in train_sequences.csv and train_sequences.v2.csv that we are actually not allowed to use for training.
\nIn that case, why were these sequences provided as part of the training set, even if it should not be used for solution? Is this just for whom trying with multiple models trained with various cutoff dates? (I'm pointing that inferencing 2024-06 RNA can be inferenced with 2024-05 uploaded model)
Hello, I'm trying to beat the VFold_human_expert_baseline score, and I understand that we need to consider the temporal cutoff date (which is safely before 2022-05-27).
\nIf I have correctly understood the rule, it means that I should not use any training data that comes after this temporal cutoff date.
\nGiven that, there are quite a few RNA sequences in train_sequences.csv and train_sequences.v2.csv that we are actually not allowed to use for training.
\nIn that case, why were these sequences provided as part of the training set, even if it should not be used for solution? Is this just for whom trying with multiple models trained with various cutoff dates? (I'm pointing that inferencing 2024-06 RNA can be inferenced with 2024-05 uploaded model)
\n\nFuture data phase. Your selected submissions will be run against a completely new private test set generated after the end of the model training phases.
\n
I believe \"the end of the model training phases\" is May 29, 2025, which is when the competition closes.
\nSo, regardless of beating the human baseline, if your goal is to win the competition, all training data provided will be safe.
\nBut another, arguably very important question, is how relevant the additional data provided are to RNA science?
", "rawMarkdown": ">Future data phase. Your selected submissions will be run against a completely new private test set generated after the end of the model training phases.\n\nI believe \"the end of the model training phases\" is May 29, 2025, which is when the competition closes.\n\nSo, regardless of beating the human baseline, if your goal is to win the competition, all training data provided will be safe.\n\nBut another, arguably very important question, is [how relevant](https://www.kaggle.com/competitions/stanford-rna-3d-folding/discussion/575109#3187485) the additional data provided are to RNA science?", "votes": null }, { "id": "3187785", "postDate": "04/26/2025 14:42:15", "content": "@doheon114 It's great that you'll be going for the Early Sharing prize!
\nSome suggestions for how to respect temporal_cutoff dates are here:
https://www.kaggle.com/competitions/stanford-rna-3d-folding/discussion/572096#3173463
", "rawMarkdown": "doheon114 It's great that you'll be going for the Early Sharing prize! \n\nSome suggestions for how to respect `temporal_cutoff` dates are here:\n\nhttps://www.kaggle.com/competitions/stanford-rna-3d-folding/discussion/572096#3173463", "votes": null }, { "id": "3188132", "postDate": "04/27/2025 05:11:09", "content": "According to what you've written, the training set's temporal cutoff deadline would be safe until May 16, 2024! Thank you very much.
", "rawMarkdown": "According to what you've written, the training set's temporal cutoff deadline would be safe until May 16, 2024! Thank you very much.", "votes": null } ], "comments": [ { "id": 3187656, "author_name": "revealer", "author_url": "", "post_date": "04/26/2025 10:55:53", "content": "\n\nFuture data phase. Your selected submissions will be run against a completely new private test set generated after the end of the model training phases.
\n
I believe \"the end of the model training phases\" is May 29, 2025, which is when the competition closes.
\nSo, regardless of beating the human baseline, if your goal is to win the competition, all training data provided will be safe.
\nBut another, arguably very important question, is how relevant the additional data provided are to RNA science?
", "votes": null, "replies": [] }, { "id": 3187785, "author_name": "rhijudas", "author_url": "", "post_date": "04/26/2025 14:42:15", "content": "@doheon114 It's great that you'll be going for the Early Sharing prize!
\nSome suggestions for how to respect temporal_cutoff dates are here:
https://www.kaggle.com/competitions/stanford-rna-3d-folding/discussion/572096#3173463
", "votes": null, "replies": [ { "id": 3188132, "author_name": "doheon114", "author_url": "", "post_date": "04/27/2025 05:11:09", "content": "According to what you've written, the training set's temporal cutoff deadline would be safe until May 16, 2024! Thank you very much.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "3187556": "Hello, I'm trying to beat the VFold_human_expert_baseline score, and I understand that we need to consider the temporal cutoff date (which is safely before 2022-05-27).\nIf I have correctly understood the rule, it means that I should not use any training data that comes after this temporal cutoff date.\nGiven that, there are quite a few RNA sequences in train_sequences.csv and train_sequences.v2.csv that we are actually not allowed to use for training.\nIn that case, why were these sequences provided as part of the training set, even if it should not be used for solution? Is this just for whom trying with multiple models trained with various cutoff dates? (I'm pointing that inferencing 2024-06 RNA can be inferenced with 2024-05 uploaded model)", "3187656": ">Future data phase. Your selected submissions will be run against a completely new private test set generated after the end of the model training phases.\n\nI believe \"the end of the model training phases\" is May 29, 2025, which is when the competition closes.\n\nSo, regardless of beating the human baseline, if your goal is to win the competition, all training data provided will be safe.\n\nBut another, arguably very important question, is [how relevant](https://www.kaggle.com/competitions/stanford-rna-3d-folding/discussion/575109#3187485) the additional data provided are to RNA science?", "3187785": "doheon114 It's great that you'll be going for the Early Sharing prize! \n\nSome suggestions for how to respect `temporal_cutoff` dates are here:\n\nhttps://www.kaggle.com/competitions/stanford-rna-3d-folding/discussion/572096#3173463", "3188132": "According to what you've written, the training set's temporal cutoff deadline would be safe until May 16, 2024! Thank you very much." }, "source": "meta" }