Is it legal to make submission in a way like here:\nhttps://www.kaggle.com/c/aptos2019-blindness-detection/discussion/100740#latest-588167\nOr for the final ones I have to make submission by uploading models weights?
", "messages": [ { "id": "615507", "postDate": "09/02/2019 03:15:32", "content": "Is it legal to make submission in a way like here:\nhttps://www.kaggle.com/c/aptos2019-blindness-detection/discussion/100740#latest-588167\nOr for the final ones I have to make submission by uploading models weights?
", "rawMarkdown": "Is it legal to make submission in a way like here:\nhttps://www.kaggle.com/c/aptos2019-blindness-detection/discussion/100740#latest-588167\nOr for the final ones I have to make submission by uploading models weights?", "votes": null }, { "id": "615510", "postDate": "09/02/2019 03:18:10", "content": "The private leaderboard score of this kind of submission would be 0.
", "rawMarkdown": "The private leaderboard score of this kind of submission would be 0.", "votes": null }, { "id": "615638", "postDate": "09/02/2019 07:43:39", "content": "why?
", "rawMarkdown": "why?", "votes": null }, { "id": "615658", "postDate": "09/02/2019 08:07:13", "content": "because you don't have the private test set - the test images that we have (test_images.zip) is only a small part (around 15%) of the full test set; this means that you can't predict the full test set locally on your PC - all predictions MUST run in a kaggle notebook
\n\njust to make it a bit more clear:\n- when you make a notebook and click commit, it runs on the public test set (the same images that we have from the zip file) - let's say in 20 minutes\n- after that when you make a submission, kaggle takes the script, replaces the test images with the full test set (around 13000 images) and runs the script on all the data; this MUST finish in less than 9 hours - the time from \"commit\" is not important at all !
\n\nthe script which you referenced predicts all \"new\" images with class 0 - wiht the fillna(0); very likely the score would actually be even lower than 0 (you can test it with submitting all 0 for the public test set, the score would probably be also < 0)
", "rawMarkdown": "because you don't have the private test set - the test images that we have (test_images.zip) is only a small part (around 15%) of the full test set; this means that you can't predict the full test set locally on your PC - all predictions MUST run in a kaggle notebook\n\njust to make it a bit more clear:\n- when you make a notebook and click commit, it runs on the public test set (the same images that we have from the zip file) - let's say in 20 minutes\n- after that when you make a submission, kaggle takes the script, replaces the test images with the full test set (around 13000 images) and runs the script on all the data; this MUST finish in less than 9 hours - the time from \"commit\" is not important at all !\n\nthe script which you referenced predicts all \"new\" images with class 0 - wiht the fillna(0); very likely the score would actually be even lower than 0 (you can test it with submitting all 0 for the public test set, the score would probably be also < 0)", "votes": null } ], "comments": [ { "id": 615510, "author_name": "naivelamb", "author_url": "", "post_date": "09/02/2019 03:18:10", "content": "The private leaderboard score of this kind of submission would be 0.
", "votes": null, "replies": [ { "id": 615638, "author_name": "talalaev", "author_url": "", "post_date": "09/02/2019 07:43:39", "content": "why?
", "votes": null, "replies": [] }, { "id": 615658, "author_name": "steelrose", "author_url": "", "post_date": "09/02/2019 08:07:13", "content": "because you don't have the private test set - the test images that we have (test_images.zip) is only a small part (around 15%) of the full test set; this means that you can't predict the full test set locally on your PC - all predictions MUST run in a kaggle notebook
\n\njust to make it a bit more clear:\n- when you make a notebook and click commit, it runs on the public test set (the same images that we have from the zip file) - let's say in 20 minutes\n- after that when you make a submission, kaggle takes the script, replaces the test images with the full test set (around 13000 images) and runs the script on all the data; this MUST finish in less than 9 hours - the time from \"commit\" is not important at all !
\n\nthe script which you referenced predicts all \"new\" images with class 0 - wiht the fillna(0); very likely the score would actually be even lower than 0 (you can test it with submitting all 0 for the public test set, the score would probably be also < 0)
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "615507": "Is it legal to make submission in a way like here:\nhttps://www.kaggle.com/c/aptos2019-blindness-detection/discussion/100740#latest-588167\nOr for the final ones I have to make submission by uploading models weights?", "615510": "The private leaderboard score of this kind of submission would be 0.", "615638": "why?", "615658": "because you don't have the private test set - the test images that we have (test_images.zip) is only a small part (around 15%) of the full test set; this means that you can't predict the full test set locally on your PC - all predictions MUST run in a kaggle notebook\n\njust to make it a bit more clear:\n- when you make a notebook and click commit, it runs on the public test set (the same images that we have from the zip file) - let's say in 20 minutes\n- after that when you make a submission, kaggle takes the script, replaces the test images with the full test set (around 13000 images) and runs the script on all the data; this MUST finish in less than 9 hours - the time from \"commit\" is not important at all !\n\nthe script which you referenced predicts all \"new\" images with class 0 - wiht the fillna(0); very likely the score would actually be even lower than 0 (you can test it with submitting all 0 for the public test set, the score would probably be also < 0)" }, "source": "meta" }