As clarified here and Kernel Requirement, participants are permitted to train a model outside of Kernels. OK, I understand. But, do I need to make my model trained at home open ?
\n\nAs for Jigsow, Kaggle team clarifies that participants need to declare external data including pre-trained models, but not need to share how to transform them and transformed data, e.g. fine-tuned models. (For details, please see this discussion)\n
In this competition, as mentioned here, no external data should be used but pre-trained model from public released academic datasets are allowed, and we have to specify all the external data used in our submission in the specified discussion post.
\n\nThen, all we have to do is declare public pre-trained models we use? or, we have to share fine-tuned models and how to use them?\n
Thanks!
", "messages": [ { "id": "507373", "postDate": "04/04/2019 15:50:29", "content": "As clarified here and Kernel Requirement, participants are permitted to train a model outside of Kernels. OK, I understand. But, do I need to make my model trained at home open ?
\n\nAs for Jigsow, Kaggle team clarifies that participants need to declare external data including pre-trained models, but not need to share how to transform them and transformed data, e.g. fine-tuned models. (For details, please see this discussion)\n
In this competition, as mentioned here, no external data should be used but pre-trained model from public released academic datasets are allowed, and we have to specify all the external data used in our submission in the specified discussion post.
\n\nThen, all we have to do is declare public pre-trained models we use? or, we have to share fine-tuned models and how to use them?\n
Thanks!
", "rawMarkdown": "As clarified [here](https://www.kaggle.com/c/imet-2019-fgvc6/discussion/87544) and [Kernel Requirement](https://www.kaggle.com/c/imet-2019-fgvc6#Kernels-Requirements), participants are permitted to train a model outside of Kernels. OK, I understand. But, do I need to _make_ **my model trained at home** _open_ ?\n\nAs for [Jigsow](https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification#Kernels-Requirements), Kaggle team clarifies that participants need to declare external data including pre-trained models, **but not need to share how to transform them and transformed data, e.g. fine-tuned models**. (For details, please see [this discussion](https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/discussion/87719))\nThis post on this discussion https://www.kaggle.com/c/imet-2019-fgvc6/discussion/87544#latest-506684
\n\n\n\n\nOur own @inversion wrote a nice explanation of why we'd run Kernels-only competitions in this type of format: https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/discussion/87719 (that applies to a different competition, but the spirit is the same here).
\n
Links to this https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/discussion/87719 where it says:
\n\n\n", "rawMarkdown": "This post on this discussion https://www.kaggle.com/c/imet-2019-fgvc6/discussion/87544#latest-506684\n> Our own @inversion wrote a nice explanation of why we'd run Kernels-only competitions in this type of format: https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/discussion/87719 (that applies to a different competition, but the spirit is the same here).\n\nLinks to this https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/discussion/87719 where it says:\n> If you transform any of these sources of external data, you still need to declare that you are using them in the forums, but you do not need to share the transformed source itself, or how you transformed it. For example, if you fine-tune VGG19, you should post that source in the forum thread, but you are not required to share your uploaded fine-tuned weights (i.e., it can be a private dataset).", "votes": null }, { "id": "507584", "postDate": "04/04/2019 22:53:23", "content": "If you transform any of these sources of external data, you still need to declare that you are using them in the forums, but you do not need to share the transformed source itself, or how you transformed it. For example, if you fine-tune VGG19, you should post that source in the forum thread, but you are not required to share your uploaded fine-tuned weights (i.e., it can be a private dataset).
\n
Yes, and he wrote,\n> that applies to a different competition, but the spirit is the same here.
\n\nI think the post shows why they run \"kernels-only\" competition which permits us to train models locally, doesn't declare that all we have to do is specify pre-trained models. Thus I post this topic, thanks :).
", "rawMarkdown": "Yes, and he wrote,\n> that applies to a different competition, but the _spirit_ is the same here.\n\nI think the post shows why they run \"kernels-only\" competition which permits us to train models locally, doesn't declare that all we have to do is specify pre-trained models. Thus I post this topic, thanks :).", "votes": null } ], "comments": [ { "id": 507555, "author_name": "mnpinto", "author_url": "", "post_date": "04/04/2019 21:36:10", "content": "This post on this discussion https://www.kaggle.com/c/imet-2019-fgvc6/discussion/87544#latest-506684
\n\n\n\n\nOur own @inversion wrote a nice explanation of why we'd run Kernels-only competitions in this type of format: https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/discussion/87719 (that applies to a different competition, but the spirit is the same here).
\n
Links to this https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/discussion/87719 where it says:
\n\n\n", "votes": null, "replies": [ { "id": 507584, "author_name": "ttahara", "author_url": "", "post_date": "04/04/2019 22:53:23", "content": "If you transform any of these sources of external data, you still need to declare that you are using them in the forums, but you do not need to share the transformed source itself, or how you transformed it. For example, if you fine-tune VGG19, you should post that source in the forum thread, but you are not required to share your uploaded fine-tuned weights (i.e., it can be a private dataset).
\n
Yes, and he wrote,\n> that applies to a different competition, but the spirit is the same here.
\n\nI think the post shows why they run \"kernels-only\" competition which permits us to train models locally, doesn't declare that all we have to do is specify pre-trained models. Thus I post this topic, thanks :).
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "507373": "As clarified [here](https://www.kaggle.com/c/imet-2019-fgvc6/discussion/87544) and [Kernel Requirement](https://www.kaggle.com/c/imet-2019-fgvc6#Kernels-Requirements), participants are permitted to train a model outside of Kernels. OK, I understand. But, do I need to _make_ **my model trained at home** _open_ ?\n\nAs for [Jigsow](https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification#Kernels-Requirements), Kaggle team clarifies that participants need to declare external data including pre-trained models, **but not need to share how to transform them and transformed data, e.g. fine-tuned models**. (For details, please see [this discussion](https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/discussion/87719))\n