First of all thanks for the interesting competition!
\nIs there a 12h runtime limit like usually? For some NN architectures that might be tough. I.e. with TabNet it might be really hard to get this done (especially when the prettrainer should be used as well).
", "messages": [ { "id": "2601543", "postDate": "01/14/2024 13:56:47", "content": "First of all thanks for the interesting competition!
\nIs there a 12h runtime limit like usually? For some NN architectures that might be tough. I.e. with TabNet it might be really hard to get this done (especially when the prettrainer should be used as well).
", "rawMarkdown": "First of all thanks for the interesting competition!\n\nIs there a 12h runtime limit like usually? For some NN architectures that might be tough. I.e. with TabNet it might be really hard to get this done (especially when the prettrainer should be used as well).", "votes": null }, { "id": "2601588", "postDate": "01/14/2024 14:26:41", "content": "@thomasmeiner i haven’t tried this as yet, but I opine that splitting the 45 datasets into smaller chunks (like 5 datasets X 9 splits) could be considered. Is this feasible @thomasmeiner?
", "rawMarkdown": "thomasmeiner i haven’t tried this as yet, but I opine that splitting the 45 datasets into smaller chunks (like 5 datasets X 9 splits) could be considered. Is this feasible @thomasmeiner?", "votes": null }, { "id": "2601605", "postDate": "01/14/2024 14:39:40", "content": "You mean splitting and then concatenating the submission? That would work
", "rawMarkdown": "You mean splitting and then concatenating the submission? That would work", "votes": null }, { "id": "2601771", "postDate": "01/14/2024 16:57:22", "content": "Yes, one could train models on separate datasets and then append them to submit @thomasmeiner
", "rawMarkdown": "Yes, one could train models on separate datasets and then append them to submit @thomasmeiner", "votes": null }, { "id": "2601897", "postDate": "01/14/2024 18:58:23", "content": "The runtime is set for all Kaggle notebooks. I am affraid have no hand in this.
\nThis is not a code competition so ultimately you can concatenate different runs. You might not know this but you can link a notebook as a source for itself, so you can keep track of past runs and update them.
", "rawMarkdown": "The runtime is set for all Kaggle notebooks. I am affraid have no hand in this. \n\nThis is not a code competition so ultimately you can concatenate different runs. You might not know this but you can link a notebook as a source for itself, so you can keep track of past runs and update them.", "votes": null }, { "id": "2603265", "postDate": "01/15/2024 17:20:02", "content": "I was thinking the same thing. I'm currently trying to run 1 suite of datasets at a time. Seems to be working, except the second suit takes sooo much time..
", "rawMarkdown": "I was thinking the same thing. I'm currently trying to run 1 suite of datasets at a time. Seems to be working, except the second suit takes sooo much time..", "votes": null } ], "comments": [ { "id": 2601588, "author_name": "ravi20076", "author_url": "", "post_date": "01/14/2024 14:26:41", "content": "@thomasmeiner i haven’t tried this as yet, but I opine that splitting the 45 datasets into smaller chunks (like 5 datasets X 9 splits) could be considered. Is this feasible @thomasmeiner?
", "votes": null, "replies": [ { "id": 2601605, "author_name": "thomasmeiner", "author_url": "", "post_date": "01/14/2024 14:39:40", "content": "You mean splitting and then concatenating the submission? That would work
", "votes": null, "replies": [ { "id": 2601771, "author_name": "ravi20076", "author_url": "", "post_date": "01/14/2024 16:57:22", "content": "Yes, one could train models on separate datasets and then append them to submit @thomasmeiner
", "votes": null, "replies": [ { "id": 2603265, "author_name": "manujosephv", "author_url": "", "post_date": "01/15/2024 17:20:02", "content": "I was thinking the same thing. I'm currently trying to run 1 suite of datasets at a time. Seems to be working, except the second suit takes sooo much time..
", "votes": null, "replies": [] } ] } ] } ] }, { "id": 2601897, "author_name": "lucasmorin", "author_url": "", "post_date": "01/14/2024 18:58:23", "content": "The runtime is set for all Kaggle notebooks. I am affraid have no hand in this.
\nThis is not a code competition so ultimately you can concatenate different runs. You might not know this but you can link a notebook as a source for itself, so you can keep track of past runs and update them.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2601543": "First of all thanks for the interesting competition!\n\nIs there a 12h runtime limit like usually? For some NN architectures that might be tough. I.e. with TabNet it might be really hard to get this done (especially when the prettrainer should be used as well).", "2601588": "thomasmeiner i haven’t tried this as yet, but I opine that splitting the 45 datasets into smaller chunks (like 5 datasets X 9 splits) could be considered. Is this feasible @thomasmeiner?", "2601605": "You mean splitting and then concatenating the submission? That would work", "2601771": "Yes, one could train models on separate datasets and then append them to submit @thomasmeiner", "2601897": "The runtime is set for all Kaggle notebooks. I am affraid have no hand in this. \n\nThis is not a code competition so ultimately you can concatenate different runs. You might not know this but you can link a notebook as a source for itself, so you can keep track of past runs and update them.", "2603265": "I was thinking the same thing. I'm currently trying to run 1 suite of datasets at a time. Seems to be working, except the second suit takes sooo much time.." }, "source": "meta" }