Hi, \nI trained 3 models on local machine, without reading the code requirements. I know I can upload my model and submit it via kernel.
\n\nBut is it within competition rules? Or I need to train my model on kernels.
\n\nAs it is written in competition rules.
\n\nhttps://www.kaggle.com/c/deepfake-detection-challenge/overview/code-requirements
\n\nI think they only mean pre-trained models and not locally trained models specifically for this task. If someone able to clarify it, it be very helpful.
", "messages": [ { "id": 722251, "postDate": "2020-01-18T10:06:32.503Z", "content": "Hi, \nI trained 3 models on local machine, without reading the code requirements. I know I can upload my model and submit it via kernel.
\n\nBut is it within competition rules? Or I need to train my model on kernels.
\n\nAs it is written in competition rules.
\n\nhttps://www.kaggle.com/c/deepfake-detection-challenge/overview/code-requirements
\n\nI think they only mean pre-trained models and not locally trained models specifically for this task. If someone able to clarify it, it be very helpful.
", "rawMarkdown": "Hi, \nI trained 3 models on local machine, without reading the code requirements. I know I can upload my model and submit it via kernel. \n\nBut is it within competition rules? Or I need to train my model on kernels. \n\nAs it is written in competition rules. \n\n###External data is allowed up to 1 GB in size. External data must be freely & publicly available, including pre-trained models. \n\nhttps://www.kaggle.com/c/deepfake-detection-challenge/overview/code-requirements\n\nI think they only mean pre-trained models and not locally trained models specifically for this task. If someone able to clarify it, it be very helpful. ", "votes": 3 }, { "id": 722427, "postDate": "2020-01-18T14:58:49.057Z", "content": "I got the answer from here, so its alright, if you using any other data you need to disclose that
\n\n\n\n\nWhat @humananalog shared is all accurate. I’ll also reiterate that you are free to train offline and upload that trained model as external data into your Kaggle notebook submission. And any datasets you might use in that training or building your model should be able to be used by all participants and publicly declared on the external data thread. You do not need to “declare” your trained model, just the input original datasets used in your model.
\n
In addition, the “AutoML” rules allow you to use such tools in your model (like Google’s AutoML), as long as you can share the associated parameters.
\n\nhttps://www.kaggle.com/c/deepfake-detection-challenge/discussion/126334
", "rawMarkdown": "I got the answer from here, so its alright, if you using any other data you need to disclose that\n\n\n> What @humananalog shared is all accurate. I’ll also reiterate that you are free to train offline and upload that trained model as external data into your Kaggle notebook submission. And any datasets you might use in that training or building your model should be able to be used by all participants and publicly declared on the external data thread. You do not need to “declare” your trained model, just the input original datasets used in your model.\n\nIn addition, the “AutoML” rules allow you to use such tools in your model (like Google’s AutoML), as long as you can share the associated parameters.\n\nhttps://www.kaggle.com/c/deepfake-detection-challenge/discussion/126334", "votes": 2, "replies": [ { "id": 723395, "postDate": "2020-01-20T01:18:29.923Z", "content": "I think this makes sense given the amount of training data and that there will be 10x more data when the models are evaluated again. Only having 9 hours for training and inference on this type of data would be very restrictive.
", "rawMarkdown": "I think this makes sense given the amount of training data and that there will be 10x more data when the models are evaluated again. Only having 9 hours for training and inference on this type of data would be very restrictive. " } ] }, { "id": 722381, "postDate": "2020-01-18T13:51:16.373Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 722414, "postDate": "2020-01-18T14:53:19.700Z", "content": "I think you are not getting.
\n\nI am only asking can I use a model which I trained on local machine to be uploaded in notebook, and using it in kernel to submit predictions on test dataset.
\n\nIs it within competition rules? Or I need to train model in Kernel?
", "rawMarkdown": "I think you are not getting. \n\nI am only asking can I use a model which I trained on local machine to be uploaded in notebook, and using it in kernel to submit predictions on test dataset. \n\nIs it within competition rules? Or I need to train model in Kernel? " } ] } ], "comments": [ { "id": 722427, "author_name": "Arpit Kanodia", "author_url": "", "post_date": "2020-01-18T14:58:49.057000", "content": "I got the answer from here, so its alright, if you using any other data you need to disclose that
\n\n\n\n\nWhat @humananalog shared is all accurate. I’ll also reiterate that you are free to train offline and upload that trained model as external data into your Kaggle notebook submission. And any datasets you might use in that training or building your model should be able to be used by all participants and publicly declared on the external data thread. You do not need to “declare” your trained model, just the input original datasets used in your model.
\n
In addition, the “AutoML” rules allow you to use such tools in your model (like Google’s AutoML), as long as you can share the associated parameters.
\n\nhttps://www.kaggle.com/c/deepfake-detection-challenge/discussion/126334
", "votes": 2, "replies": [ { "id": 723395, "author_name": "Jack Vial", "author_url": "", "post_date": "2020-01-20T01:18:29.923000", "content": "I think this makes sense given the amount of training data and that there will be 10x more data when the models are evaluated again. Only having 9 hours for training and inference on this type of data would be very restrictive.
", "votes": 0, "replies": [] } ] }, { "id": 722381, "author_name": "", "author_url": "", "post_date": "2020-01-18T13:51:16.373000", "content": "", "votes": 0, "replies": [ { "id": 722414, "author_name": "Arpit Kanodia", "author_url": "", "post_date": "2020-01-18T14:53:19.700000", "content": "I think you are not getting.
\n\nI am only asking can I use a model which I trained on local machine to be uploaded in notebook, and using it in kernel to submit predictions on test dataset.
\n\nIs it within competition rules? Or I need to train model in Kernel?
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "722251": "Hi, \nI trained 3 models on local machine, without reading the code requirements. I know I can upload my model and submit it via kernel. \n\nBut is it within competition rules? Or I need to train my model on kernels. \n\nAs it is written in competition rules. \n\n###External data is allowed up to 1 GB in size. External data must be freely & publicly available, including pre-trained models. \n\nhttps://www.kaggle.com/c/deepfake-detection-challenge/overview/code-requirements\n\nI think they only mean pre-trained models and not locally trained models specifically for this task. If someone able to clarify it, it be very helpful. ", "722427": "I got the answer from here, so its alright, if you using any other data you need to disclose that\n\n\n> What @humananalog shared is all accurate. I’ll also reiterate that you are free to train offline and upload that trained model as external data into your Kaggle notebook submission. And any datasets you might use in that training or building your model should be able to be used by all participants and publicly declared on the external data thread. You do not need to “declare” your trained model, just the input original datasets used in your model.\n\nIn addition, the “AutoML” rules allow you to use such tools in your model (like Google’s AutoML), as long as you can share the associated parameters.\n\nhttps://www.kaggle.com/c/deepfake-detection-challenge/discussion/126334", "722381": "" } }