![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3768255%2Fbf1c344a91ba86c2f85b3894740596d4%2FKaggleSubmissionRules.png?generation=1589783383896995&alt=media\")
Thanks for your participation! As it's become clear that a significant number of the top finishers made use of external data that included the test images, we're revising our approach to identifying qualified candidates for the TPU Leaderboard prizes.
\n\nIf you would like to be considered, please make a submission to this form by May 25th 2020: https://www.kaggle.com/tpu-leaderboard-submission.
\n\nFor clarification, submissions to this form must meet these eligibility requirements:\n1. Your submission must generate predictions from machine learning code fully run on TPUs for both training and inference\n2. You must have used external data that did not use test images. This includes if you used any of the 5 (ImageNet, Open Images, Oxford, TensorFlow, iNaturalist) datasets. You would need to have cleansed those datasets of overlaps/duplicates of the test images.\n3. Your notebook(s) must be made public.
\n\nPlease do not submit if you do not meet all 3 of these criteria.
\n\nWe'll review and verify the code of submissions starting with the highest private leaderboard score and announce winners on the forum! The existing leaderboard will remain as-is, since this is a playground competition.
\n\nWe are also awarding \"TPU Star\" prizes if you have made helpful community contributions (notebooks and forum discussions/comments), even if you do not meet the Leaderboard prize criteria above.
", "messages": [ { "id": "849377", "postDate": "05/15/2020 17:49:21", "content": "Thanks for your participation! As it's become clear that a significant number of the top finishers made use of external data that included the test images, we're revising our approach to identifying qualified candidates for the TPU Leaderboard prizes.
\n\nIf you would like to be considered, please make a submission to this form by May 25th 2020: https://www.kaggle.com/tpu-leaderboard-submission.
\n\nFor clarification, submissions to this form must meet these eligibility requirements:\n1. Your submission must generate predictions from machine learning code fully run on TPUs for both training and inference\n2. You must have used external data that did not use test images. This includes if you used any of the 5 (ImageNet, Open Images, Oxford, TensorFlow, iNaturalist) datasets. You would need to have cleansed those datasets of overlaps/duplicates of the test images.\n3. Your notebook(s) must be made public.
\n\nPlease do not submit if you do not meet all 3 of these criteria.
\n\nWe'll review and verify the code of submissions starting with the highest private leaderboard score and announce winners on the forum! The existing leaderboard will remain as-is, since this is a playground competition.
\n\nWe are also awarding \"TPU Star\" prizes if you have made helpful community contributions (notebooks and forum discussions/comments), even if you do not meet the Leaderboard prize criteria above.
", "rawMarkdown": "Thanks for your participation! As it's become clear that a significant number of the top finishers made use of external data that included the test images, we're revising our approach to identifying qualified candidates for the TPU Leaderboard prizes.\n\nIf you would like to be considered, please make a submission to this form by **May 25th 2020**: [https://www.kaggle.com/tpu-leaderboard-submission](https://www.kaggle.com/tpu-leaderboard-submission).\n\nFor clarification, submissions to this form must meet these eligibility requirements:\n1. Your submission must generate predictions from machine learning code fully run on TPUs for both training and inference\n2. You must have used external data that did not use test images. This includes if you used any of the 5 (ImageNet, Open Images, Oxford, TensorFlow, iNaturalist) datasets. You would need to have cleansed those datasets of overlaps/duplicates of the test images.\n3. Your notebook(s) must be made public.\n\n**Please do not submit if you do not meet all 3 of these criteria.**\n\nWe'll review and verify the code of submissions starting with the highest private leaderboard score and announce winners on the forum! The existing leaderboard will remain as-is, since this is a playground competition.\n\n\n*We are also awarding [\"TPU Star\" prizes](https://www.kaggle.com/c/flower-classification-with-tpus/discussion/150908) if you have made helpful community contributions (notebooks and forum discussions/comments), even if you do not meet the Leaderboard prize criteria above.*", "votes": null }, { "id": "849549", "postDate": "05/15/2020 22:04:34", "content": "I do not agree at all with maintaining the current leaderboard positions. I think the fairest thing would be to cancel (blank) Private Leaderboard.
\n\nRegardless of whether the competition is playground, Kaggle is rewarding a final high private ranking for those who did not follow your directions to select only clean submissions.
\n\nMany of us stop working with external data sets after your data-posts (data-thread) days before the deadline and put aside the effort (and scoring) made with them.
", "rawMarkdown": "I do not agree at all with maintaining the current leaderboard positions. I think the fairest thing would be to cancel (blank) Private Leaderboard.\n\nRegardless of whether the competition is playground, Kaggle is rewarding a final high private ranking for those who did not follow your directions to select only clean submissions.\n\nMany of us stop working with external data sets after your data-posts (data-thread) days before the deadline and put aside the effort (and scoring) made with them.", "votes": null }, { "id": "851644", "postDate": "05/17/2020 19:27:24", "content": "My suggestion is to rescore the leaderboard using the images that were not included in any of the external datasets!
", "rawMarkdown": "My suggestion is to rescore the leaderboard using the images that were not included in any of the external datasets!", "votes": null }, { "id": "852059", "postDate": "05/18/2020 06:39:11", "content": "It seems that point number 2 of the current eligibility requirements may be unintentional unless I have understood it wrong. While the second half of the first sentence seems correct the first part indicates that using external data after excluding test images was mandatory.
\n\n
@haveri To win a TPU Leaderboard prize, if you used external data, then that external data must not use the test images. If you didn't use external data at all for training, then that's fine, as well.
", "rawMarkdown": "haveri To win a TPU Leaderboard prize, if you used external data, then that external data must not use the test images. If you didn't use external data at all for training, then that's fine, as well.", "votes": null } ], "comments": [ { "id": 849549, "author_name": "coreacasa", "author_url": "", "post_date": "05/15/2020 22:04:34", "content": "I do not agree at all with maintaining the current leaderboard positions. I think the fairest thing would be to cancel (blank) Private Leaderboard.
\n\nRegardless of whether the competition is playground, Kaggle is rewarding a final high private ranking for those who did not follow your directions to select only clean submissions.
\n\nMany of us stop working with external data sets after your data-posts (data-thread) days before the deadline and put aside the effort (and scoring) made with them.
", "votes": null, "replies": [] }, { "id": 851644, "author_name": "safavieh", "author_url": "", "post_date": "05/17/2020 19:27:24", "content": "My suggestion is to rescore the leaderboard using the images that were not included in any of the external datasets!
", "votes": null, "replies": [] }, { "id": 852059, "author_name": "haveri", "author_url": "", "post_date": "05/18/2020 06:39:11", "content": "It seems that point number 2 of the current eligibility requirements may be unintentional unless I have understood it wrong. While the second half of the first sentence seems correct the first part indicates that using external data after excluding test images was mandatory.
\n\n
@haveri To win a TPU Leaderboard prize, if you used external data, then that external data must not use the test images. If you didn't use external data at all for training, then that's fine, as well.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "849377": "Thanks for your participation! As it's become clear that a significant number of the top finishers made use of external data that included the test images, we're revising our approach to identifying qualified candidates for the TPU Leaderboard prizes.\n\nIf you would like to be considered, please make a submission to this form by **May 25th 2020**: [https://www.kaggle.com/tpu-leaderboard-submission](https://www.kaggle.com/tpu-leaderboard-submission).\n\nFor clarification, submissions to this form must meet these eligibility requirements:\n1. Your submission must generate predictions from machine learning code fully run on TPUs for both training and inference\n2. You must have used external data that did not use test images. This includes if you used any of the 5 (ImageNet, Open Images, Oxford, TensorFlow, iNaturalist) datasets. You would need to have cleansed those datasets of overlaps/duplicates of the test images.\n3. Your notebook(s) must be made public.\n\n**Please do not submit if you do not meet all 3 of these criteria.**\n\nWe'll review and verify the code of submissions starting with the highest private leaderboard score and announce winners on the forum! The existing leaderboard will remain as-is, since this is a playground competition.\n\n\n*We are also awarding [\"TPU Star\" prizes](https://www.kaggle.com/c/flower-classification-with-tpus/discussion/150908) if you have made helpful community contributions (notebooks and forum discussions/comments), even if you do not meet the Leaderboard prize criteria above.*", "849549": "I do not agree at all with maintaining the current leaderboard positions. I think the fairest thing would be to cancel (blank) Private Leaderboard.\n\nRegardless of whether the competition is playground, Kaggle is rewarding a final high private ranking for those who did not follow your directions to select only clean submissions.\n\nMany of us stop working with external data sets after your data-posts (data-thread) days before the deadline and put aside the effort (and scoring) made with them.", "851644": "My suggestion is to rescore the leaderboard using the images that were not included in any of the external datasets!", "852059": "It seems that point number 2 of the current eligibility requirements may be unintentional unless I have understood it wrong. While the second half of the first sentence seems correct the first part indicates that using external data after excluding test images was mandatory.\n\n", "854291": "haveri To win a TPU Leaderboard prize, if you used external data, then that external data must not use the test images. If you didn't use external data at all for training, then that's fine, as well." }, "source": "meta" }