![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1109229%2F66509268fc35872c9e3faccb898a6f37%2FScreen%20Shot%202019-12-19%20at%2022.56.24.png?generation=1576767513275742&alt=media\")
Welcome to the TensorFlow 2.0 Question Answering competition.
\n\nIn this competition we're answering questions by identifying relevant excerpts from Wikipedia pages. Please make sure to check out our Evaluation page to see how our metric works, and how it differs from the standard Natural Questions metric. As always, there's pertinent information on the Data page re: data and prediction structure.
\n\nThe original TF 1.x Bert baseline for this dataset has been updated to work with TF 2.0, and is included as both a benchmark and a starter notebook. Please fork and improve it! Several utility scripts are included that open up some of the Bert back end if desired.
\n\nNote that this is an inference-only code competition - you should be doing most of your training offline, as GPU notebooks have a limit of only 3 hours.
\n\nPlease let us know if you have any questions. Best of luck!
", "messages": [ { "id": 660205, "postDate": "2019-10-28T21:37:25.963Z", "content": "Welcome to the TensorFlow 2.0 Question Answering competition.
\n\nIn this competition we're answering questions by identifying relevant excerpts from Wikipedia pages. Please make sure to check out our Evaluation page to see how our metric works, and how it differs from the standard Natural Questions metric. As always, there's pertinent information on the Data page re: data and prediction structure.
\n\nThe original TF 1.x Bert baseline for this dataset has been updated to work with TF 2.0, and is included as both a benchmark and a starter notebook. Please fork and improve it! Several utility scripts are included that open up some of the Bert back end if desired.
\n\nNote that this is an inference-only code competition - you should be doing most of your training offline, as GPU notebooks have a limit of only 3 hours.
\n\nPlease let us know if you have any questions. Best of luck!
", "rawMarkdown": "Welcome to the TensorFlow 2.0 Question Answering competition.\n\nIn this competition we're answering questions by identifying relevant excerpts from Wikipedia pages. Please make sure to check out our Evaluation page to see how our metric works, and how it differs from the standard Natural Questions metric. As always, there's pertinent information on the Data page re: data and prediction structure.\n\nThe original TF 1.x Bert baseline for this dataset has been updated to work with TF 2.0, and is included as both a benchmark and a [starter notebook](https://www.kaggle.com/philculliton/using-tensorflow-2-0-w-bert-on-nq). Please fork and improve it! Several utility scripts are included that open up some of the Bert back end if desired.\n\nNote that this is an [inference-only code competition](https://www.kaggle.com/c/tensorflow2-question-answering/overview/code-requirements) - you should be doing most of your training offline, as GPU notebooks have a limit of only 3 hours.\n\nPlease let us know if you have any questions. Best of luck!", "votes": 11 }, { "id": 682441, "postDate": "2019-11-27T12:39:52.630Z", "content": "Hi @philculliton and @juliaelliott
\n\nIt'd be really great if you listen to these shout-outs about sharing the official metric implementation. It's good for Kaggle to learn from its mistakes - the community can at least verify that the metric is calculated in a right transparent way. Otherwise the greatest part of the competition is about reverse-engineering.
", "rawMarkdown": "Hi @philculliton and @juliaelliott \n\nIt'd be really great if you listen to these shout-outs about sharing the official metric implementation. It's good for Kaggle to learn from its mistakes - the community can at least verify that the metric is calculated in a right transparent way. Otherwise the greatest part of the competition is about reverse-engineering. ", "votes": 6, "replies": [ { "id": 682703, "postDate": "2019-11-27T19:08:08.847Z", "content": "I second that. Quite some discussion posts requesting for metric clarification. 🙄 It just makes no sense to start modeling if that has not been clarified
", "rawMarkdown": "I second that. Quite some discussion posts requesting for metric clarification. 🙄 It just makes no sense to start modeling if that has not been clarified", "votes": 2 }, { "id": 682797, "postDate": "2019-11-27T21:30:45.003Z", "content": "Some more \"fun\" here.
", "rawMarkdown": "Some more \"fun\" [here](https://www.kaggle.com/c/tensorflow2-question-answering/discussion/119280). " } ] }, { "id": 709941, "postDate": "2020-01-04T05:13:56.387Z", "content": "Is the 3 hour GPU limit on submissions going to be for the 88% of the data that's private, the 12% of the data that is public, or for all the test data combined?
", "rawMarkdown": "Is the 3 hour GPU limit on submissions going to be for the 88% of the data that's private, the 12% of the data that is public, or for all the test data combined?", "votes": 1, "replies": [ { "id": 709971, "postDate": "2020-01-04T06:29:32.113Z", "content": "@ejmejm It applies to both. So you should prepare for your committed notebook to also have adequate time to re-run on the private test set (which is ~7 times the public test set).
", "rawMarkdown": "@ejmejm It applies to both. So you should prepare for your committed notebook to also have adequate time to re-run on the private test set (which is ~7 times the public test set).", "votes": 1 } ] }, { "id": 698675, "postDate": "2019-12-19T14:58:49.240Z", "content": "Hi @juliaelliott and @philculliton
\nI submit the same kernel twice, once successful and once prompt timeout. Is there any way to solve this problem?
@wochidadonggua You'll need to ensure your notebook can be successfully run when submitted against the private test set, which is considerably larger. Since your notebook's run is likely non-deterministic or if you've made any changes (even small) to the code, I imagine it could vary in runtime from run to run.
", "rawMarkdown": "@wochidadonggua You'll need to ensure your notebook can be successfully run when submitted against the private test set, which is considerably larger. Since your notebook's run is likely non-deterministic or if you've made any changes (even small) to the code, I imagine it could vary in runtime from run to run." } ] }, { "id": 660340, "postDate": "2019-10-29T02:34:46.397Z", "content": "How do I participate if all programming is done in C++? Do I just build and train my model in c++ then write a script in Python just to read the model and make inference? I am seeing more and more of these code competitions and they do not really allow for C++.
", "rawMarkdown": "How do I participate if all programming is done in C++? Do I just build and train my model in c++ then write a script in Python just to read the model and make inference? I am seeing more and more of these code competitions and they do not really allow for C++. ", "votes": 2 }, { "id": 719483, "postDate": "2020-01-15T14:36:44.980Z", "content": "Hi, @juliaelliott !
\n\nSorry if that's already been asked but it's not clear for our team how many Kaggle GPU credits we'll have.
\n\nI see these 30h reset on Saturday. Do we share these hours or it's 30h for each one of us?
", "rawMarkdown": "Hi, @juliaelliott !\n\nSorry if that's already been asked but it's not clear for our team how many Kaggle GPU credits we'll have. \n\nI see these 30h reset on Saturday. Do we share these hours or it's 30h for each one of us? ", "replies": [ { "id": 719566, "postDate": "2020-01-15T15:45:56.303Z", "content": "@kashnitsky The weekly quota is per person and global (across the Kaggle platform and not competition-specific), regardless of how you team.
", "rawMarkdown": "@kashnitsky The weekly quota is per person and global (across the Kaggle platform and not competition-specific), regardless of how you team.", "votes": 1 }, { "id": 719574, "postDate": "2020-01-15T15:54:47.487Z", "content": "Thanks for the clarification!
", "rawMarkdown": "Thanks for the clarification!" } ] }, { "id": 680537, "postDate": "2019-11-24T21:05:46.903Z", "content": "It should be officially made very clear that the acceptable coding Type could be either Notebook OR Script. As this is a Code Competition, rather than only saying Submissions to this competition must be made through Kaggle Notebooks in the Code Requirements section, it could be written as Submissions to this competition must be made through Kaggle Notebooks (Notebook or Script type) or something on those lines to make it more clear. Otherwise, having a button with the New Notebook label and then again given a choice between Notebook and Script, can be a little bit unclear at first especially for this type of competition.
@kishupro Thanks for your feedback. We have been using \"Notebook\" to cover both \"Notebook or Script\" types as a general term. But agree that this can be confusing to new users. We will consider revising this.
", "rawMarkdown": "@kishupro Thanks for your feedback. We have been using \"Notebook\" to cover both \"Notebook or Script\" types as a general term. But agree that this can be confusing to new users. We will consider revising this.", "votes": 1 }, { "id": 682158, "postDate": "2019-11-27T02:45:55.030Z", "content": "Thanks for your consideration! In my personal opinion, Kernel was a better term to cover them both technically and otherwise too. But surely the nomenclature would be better without a duplication of the term Notebook.
Do we need to use tensorflow 2.0 or we are free to use any other framework?\nAlso how we can upload the trained model?
", "rawMarkdown": "Do we need to use tensorflow 2.0 or we are free to use any other framework?\nAlso how we can upload the trained model?", "replies": [ { "id": 665128, "postDate": "2019-11-04T16:53:28.387Z", "content": "https://www.kaggle.com/c/tensorflow2-question-answering/discussion/115492#latest-665127
", "rawMarkdown": "https://www.kaggle.com/c/tensorflow2-question-answering/discussion/115492#latest-665127" } ] }, { "id": 663818, "postDate": "2019-11-02T17:48:14.563Z", "content": "Hi, \nI am looking for a team if you still someone can participate with you?
", "rawMarkdown": "Hi, \nI am looking for a team if you still someone can participate with you?" }, { "id": 662057, "postDate": "2019-10-31T03:31:04.707Z", "content": "Do we need to use tensorflow 2.0 or we are free to use any other framework?\nAlso how we can upload the trained model?
", "rawMarkdown": "Do we need to use tensorflow 2.0 or we are free to use any other framework?\nAlso how we can upload the trained model?" }, { "id": 685605, "postDate": "2019-12-02T01:54:13.943Z", "rawMarkdown": "", "isDeleted": true }, { "id": 673706, "postDate": "2019-11-15T11:13:58.117Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 682441, "author_name": "Yury Kashnitsky", "author_url": "", "post_date": "2019-11-27T12:39:52.630000", "content": "Hi @philculliton and @juliaelliott
\n\nIt'd be really great if you listen to these shout-outs about sharing the official metric implementation. It's good for Kaggle to learn from its mistakes - the community can at least verify that the metric is calculated in a right transparent way. Otherwise the greatest part of the competition is about reverse-engineering.
", "votes": 6, "replies": [ { "id": 682703, "author_name": "Dieter", "author_url": "", "post_date": "2019-11-27T19:08:08.847000", "content": "I second that. Quite some discussion posts requesting for metric clarification. 🙄 It just makes no sense to start modeling if that has not been clarified
", "votes": 2, "replies": [] }, { "id": 682797, "author_name": "Yury Kashnitsky", "author_url": "", "post_date": "2019-11-27T21:30:45.003000", "content": "Some more \"fun\" here.
", "votes": 0, "replies": [] } ] }, { "id": 709941, "author_name": "Ejmejm", "author_url": "", "post_date": "2020-01-04T05:13:56.387000", "content": "Is the 3 hour GPU limit on submissions going to be for the 88% of the data that's private, the 12% of the data that is public, or for all the test data combined?
", "votes": 1, "replies": [ { "id": 709971, "author_name": "Julia Elliott", "author_url": "", "post_date": "2020-01-04T06:29:32.113000", "content": "@ejmejm It applies to both. So you should prepare for your committed notebook to also have adequate time to re-run on the private test set (which is ~7 times the public test set).
", "votes": 1, "replies": [] } ] }, { "id": 698675, "author_name": "Hikkiiiiiiiii", "author_url": "", "post_date": "2019-12-19T14:58:49.240000", "content": "Hi @juliaelliott and @philculliton
\nI submit the same kernel twice, once successful and once prompt timeout. Is there any way to solve this problem?
@wochidadonggua You'll need to ensure your notebook can be successfully run when submitted against the private test set, which is considerably larger. Since your notebook's run is likely non-deterministic or if you've made any changes (even small) to the code, I imagine it could vary in runtime from run to run.
", "votes": 0, "replies": [] } ] }, { "id": 660340, "author_name": "rilesdg3", "author_url": "", "post_date": "2019-10-29T02:34:46.397000", "content": "How do I participate if all programming is done in C++? Do I just build and train my model in c++ then write a script in Python just to read the model and make inference? I am seeing more and more of these code competitions and they do not really allow for C++.
", "votes": 2, "replies": [] }, { "id": 719483, "author_name": "Yury Kashnitsky", "author_url": "", "post_date": "2020-01-15T14:36:44.980000", "content": "Hi, @juliaelliott !
\n\nSorry if that's already been asked but it's not clear for our team how many Kaggle GPU credits we'll have.
\n\nI see these 30h reset on Saturday. Do we share these hours or it's 30h for each one of us?
", "votes": 0, "replies": [ { "id": 719566, "author_name": "Julia Elliott", "author_url": "", "post_date": "2020-01-15T15:45:56.303000", "content": "@kashnitsky The weekly quota is per person and global (across the Kaggle platform and not competition-specific), regardless of how you team.
", "votes": 1, "replies": [] }, { "id": 719574, "author_name": "Yury Kashnitsky", "author_url": "", "post_date": "2020-01-15T15:54:47.487000", "content": "Thanks for the clarification!
", "votes": 0, "replies": [] } ] }, { "id": 680537, "author_name": "Kishore Badyakar", "author_url": "", "post_date": "2019-11-24T21:05:46.903000", "content": "It should be officially made very clear that the acceptable coding Type could be either Notebook OR Script. As this is a Code Competition, rather than only saying Submissions to this competition must be made through Kaggle Notebooks in the Code Requirements section, it could be written as Submissions to this competition must be made through Kaggle Notebooks (Notebook or Script type) or something on those lines to make it more clear. Otherwise, having a button with the New Notebook label and then again given a choice between Notebook and Script, can be a little bit unclear at first especially for this type of competition.
@kishupro Thanks for your feedback. We have been using \"Notebook\" to cover both \"Notebook or Script\" types as a general term. But agree that this can be confusing to new users. We will consider revising this.
", "votes": 1, "replies": [] }, { "id": 682158, "author_name": "Kishore Badyakar", "author_url": "", "post_date": "2019-11-27T02:45:55.030000", "content": "Thanks for your consideration! In my personal opinion, Kernel was a better term to cover them both technically and otherwise too. But surely the nomenclature would be better without a duplication of the term Notebook.
Do we need to use tensorflow 2.0 or we are free to use any other framework?\nAlso how we can upload the trained model?
", "votes": 0, "replies": [ { "id": 665128, "author_name": "Julia Elliott", "author_url": "", "post_date": "2019-11-04T16:53:28.387000", "content": "https://www.kaggle.com/c/tensorflow2-question-answering/discussion/115492#latest-665127
", "votes": 0, "replies": [] } ] }, { "id": 663818, "author_name": "Medou Neine", "author_url": "", "post_date": "2019-11-02T17:48:14.563000", "content": "Hi, \nI am looking for a team if you still someone can participate with you?
", "votes": 0, "replies": [] }, { "id": 662057, "author_name": "atulkum", "author_url": "", "post_date": "2019-10-31T03:31:04.707000", "content": "Do we need to use tensorflow 2.0 or we are free to use any other framework?\nAlso how we can upload the trained model?
", "votes": 0, "replies": [] }, { "id": 685605, "author_name": "", "author_url": "", "post_date": "2019-12-02T01:54:13.943000", "content": "", "votes": 0, "replies": [] }, { "id": 673706, "author_name": "", "author_url": "", "post_date": "2019-11-15T11:13:58.117000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "660205": "Welcome to the TensorFlow 2.0 Question Answering competition.\n\nIn this competition we're answering questions by identifying relevant excerpts from Wikipedia pages. Please make sure to check out our Evaluation page to see how our metric works, and how it differs from the standard Natural Questions metric. As always, there's pertinent information on the Data page re: data and prediction structure.\n\nThe original TF 1.x Bert baseline for this dataset has been updated to work with TF 2.0, and is included as both a benchmark and a [starter notebook](https://www.kaggle.com/philculliton/using-tensorflow-2-0-w-bert-on-nq). Please fork and improve it! Several utility scripts are included that open up some of the Bert back end if desired.\n\nNote that this is an [inference-only code competition](https://www.kaggle.com/c/tensorflow2-question-answering/overview/code-requirements) - you should be doing most of your training offline, as GPU notebooks have a limit of only 3 hours.\n\nPlease let us know if you have any questions. Best of luck!", "682441": "Hi @philculliton and @juliaelliott \n\nIt'd be really great if you listen to these shout-outs about sharing the official metric implementation. It's good for Kaggle to learn from its mistakes - the community can at least verify that the metric is calculated in a right transparent way. Otherwise the greatest part of the competition is about reverse-engineering. ", "709941": "Is the 3 hour GPU limit on submissions going to be for the 88% of the data that's private, the 12% of the data that is public, or for all the test data combined?", "698675": "Hi @juliaelliott and @philculliton \nI submit the same kernel twice, once successful and once prompt timeout. Is there any way to solve this problem?\n\n\n", "660340": "How do I participate if all programming is done in C++? Do I just build and train my model in c++ then write a script in Python just to read the model and make inference? I am seeing more and more of these code competitions and they do not really allow for C++. ", "719483": "Hi, @juliaelliott !\n\nSorry if that's already been asked but it's not clear for our team how many Kaggle GPU credits we'll have. \n\nI see these 30h reset on Saturday. Do we share these hours or it's 30h for each one of us? ", "680537": "It should be officially made very clear that the **acceptable** coding `Type` could be either `Notebook` **OR** `Script`. As this is a `Code Competition`, rather than only saying `Submissions to this competition must be made through Kaggle Notebooks` in the *Code Requirements* section, it could be written as `Submissions to this competition must be made through Kaggle Notebooks (Notebook or Script type)` or something on those lines to make it more clear. Otherwise, having a button with the `New Notebook` label and then again given a choice between `Notebook` and `Script`, can be a little bit unclear at first especially for this type of competition.", "664078": "Do we need to use tensorflow 2.0 or we are free to use any other framework?\nAlso how we can upload the trained model?", "663818": "Hi, \nI am looking for a team if you still someone can participate with you?", "662057": "Do we need to use tensorflow 2.0 or we are free to use any other framework?\nAlso how we can upload the trained model?", "685605": "", "673706": "" } }