Hey,
\nIn the public dataset, there are 2 \"test\" fragments, which are just subsets of the train set. These 2 \"test\" fragments get replaced with the actual test fragments during submission.
\nI just want to be sure, there are only 2 actual test fragments and not more, right? And the actual 2 test fragments are also called \"a\" and \"b\", right?
", "messages": [ { "id": "2286111", "postDate": "06/03/2023 08:22:04", "content": "Hey,
\nIn the public dataset, there are 2 \"test\" fragments, which are just subsets of the train set. These 2 \"test\" fragments get replaced with the actual test fragments during submission.
\nI just want to be sure, there are only 2 actual test fragments and not more, right? And the actual 2 test fragments are also called \"a\" and \"b\", right?
", "rawMarkdown": "Hey,\n\nIn the public dataset, there are 2 \"test\" fragments, which are just subsets of the train set. These 2 \"test\" fragments get replaced with the actual test fragments during submission. \n\nI just want to be sure, there are only 2 actual test fragments and not more, right? And the actual 2 test fragments are also called \"a\" and \"b\", right?", "votes": null }, { "id": "2286268", "postDate": "06/03/2023 10:48:23", "content": "I think just the test fragments get changed, keeping the paths the same (as 'a' and 'b'), and by that logic, the number of actual test images must also be 2.
", "rawMarkdown": "I think just the test fragments get changed, keeping the paths the same (as 'a' and 'b'), and by that logic, the number of actual test images must also be 2.", "votes": null }, { "id": "2286675", "postDate": "06/03/2023 16:35:14", "content": "Thank you I had the same question! I've been digging for a while but couldn't find a 100% convincing answer about what is going on.
\nThere definitely are test/a and test/b folders when I submit, but as far as I can tell the fragments don't get replaced at all and are just the usual train/1 fragment split into 2 parts. You can see and even plot information about the a and b test fragments in the logs and the downloadable output of the submission, and it looks exactly like the train/1 fragment.
\nIs it that when you enable one of the \"persistance options\" (right below the submit button) the test set does not get swapped with a hidden test set? I'm really confused, as being able to get info about the test set used for the LB would just make the LB pointless. Fairly new to Kaggle and new to this competition so it might be something obvious I'm missingš¤
When you click on submit, the notebook is actually run in parallel 2 times. One run is the one that you can see with the fake test data and one is the actual submission or which no log and output is visible.
", "rawMarkdown": "When you click on submit, the notebook is actually run in parallel 2 times. One run is the one that you can see with the fake test data and one is the actual submission or which no log and output is visible.", "votes": null }, { "id": "2286682", "postDate": "06/03/2023 16:44:57", "content": "Ahh that makes a lot of sense, thank you very much!
", "rawMarkdown": "Ahh that makes a lot of sense, thank you very much!", "votes": null }, { "id": "2287470", "postDate": "06/04/2023 14:00:03", "content": "Here is some more info. The hidden fragments are of similar size to the dummy \"test\" fragments.
", "rawMarkdown": "[Here](https://www.kaggle.com/competitions/vesuvius-challenge-ink-detection/discussion/401514#2221857) is some more info. The hidden fragments are of similar size to the dummy \"test\" fragments.", "votes": null } ], "comments": [ { "id": 2286268, "author_name": "thearijitdas", "author_url": "", "post_date": "06/03/2023 10:48:23", "content": "I think just the test fragments get changed, keeping the paths the same (as 'a' and 'b'), and by that logic, the number of actual test images must also be 2.
", "votes": null, "replies": [] }, { "id": 2286675, "author_name": "danieleroncaglioni", "author_url": "", "post_date": "06/03/2023 16:35:14", "content": "Thank you I had the same question! I've been digging for a while but couldn't find a 100% convincing answer about what is going on.
\nThere definitely are test/a and test/b folders when I submit, but as far as I can tell the fragments don't get replaced at all and are just the usual train/1 fragment split into 2 parts. You can see and even plot information about the a and b test fragments in the logs and the downloadable output of the submission, and it looks exactly like the train/1 fragment.
\nIs it that when you enable one of the \"persistance options\" (right below the submit button) the test set does not get swapped with a hidden test set? I'm really confused, as being able to get info about the test set used for the LB would just make the LB pointless. Fairly new to Kaggle and new to this competition so it might be something obvious I'm missingš¤
When you click on submit, the notebook is actually run in parallel 2 times. One run is the one that you can see with the fake test data and one is the actual submission or which no log and output is visible.
", "votes": null, "replies": [ { "id": 2286682, "author_name": "danieleroncaglioni", "author_url": "", "post_date": "06/03/2023 16:44:57", "content": "Ahh that makes a lot of sense, thank you very much!
", "votes": null, "replies": [] } ] } ] }, { "id": 2287470, "author_name": "alexz0", "author_url": "", "post_date": "06/04/2023 14:00:03", "content": "Here is some more info. The hidden fragments are of similar size to the dummy \"test\" fragments.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2286111": "Hey,\n\nIn the public dataset, there are 2 \"test\" fragments, which are just subsets of the train set. These 2 \"test\" fragments get replaced with the actual test fragments during submission. \n\nI just want to be sure, there are only 2 actual test fragments and not more, right? And the actual 2 test fragments are also called \"a\" and \"b\", right?", "2286268": "I think just the test fragments get changed, keeping the paths the same (as 'a' and 'b'), and by that logic, the number of actual test images must also be 2.", "2286675": "Thank you I had the same question! I've been digging for a while but couldn't find a 100% convincing answer about what is going on.\nThere definitely are test/a and test/b folders when I submit, but as far as I can tell the fragments don't get replaced at all and are just the usual train/1 fragment split into 2 parts. You can see and even plot information about the a and b test fragments in the logs and the downloadable output of the submission, and it looks exactly like the train/1 fragment.\nIs it that when you enable one of the \"persistance options\" (right below the submit button) the test set does not get swapped with a hidden test set? I'm really confused, as being able to get info about the test set used for the LB would just make the LB pointless. Fairly new to Kaggle and new to this competition so it might be something obvious I'm missingš¤", "2286680": "When you click on submit, the notebook is actually run in parallel 2 times. One run is the one that you can see with the fake test data and one is the actual submission or which no log and output is visible.", "2286682": "Ahh that makes a lot of sense, thank you very much!", "2287470": "[Here](https://www.kaggle.com/competitions/vesuvius-challenge-ink-detection/discussion/401514#2221857) is some more info. The hidden fragments are of similar size to the dummy \"test\" fragments." }, "source": "meta" }