Hey all! I am new to Kaggle and have a genuine doubt. What are the rules regarding use of pre-trained models? Can we use any pre-trained model including ones which we have trained outside of Kaggle on other datasets? What's stopping someone from training the classifier for a longer duration outside Kaggle and then making submissions using that model? Thank you!
", "messages": [ { "id": "1294249", "postDate": "05/05/2021 13:59:52", "content": "Hey all! I am new to Kaggle and have a genuine doubt. What are the rules regarding use of pre-trained models? Can we use any pre-trained model including ones which we have trained outside of Kaggle on other datasets? What's stopping someone from training the classifier for a longer duration outside Kaggle and then making submissions using that model? Thank you!
", "rawMarkdown": "Hey all! I am new to Kaggle and have a genuine doubt. What are the rules regarding use of pre-trained models? Can we use any pre-trained model including ones which we have trained outside of Kaggle on other datasets? What's stopping someone from training the classifier for a longer duration outside Kaggle and then making submissions using that model? Thank you!", "votes": null }, { "id": "1303445", "postDate": "05/12/2021 04:31:18", "content": "Yes you can use pre trained model , but I would suggest , have a primary pre-trained backbone and then make your own customisations to the model , in order to get your desired result
\nThank you
Thank you for your reply! However, if I train my own model outside of Kaggle for more than 2 hours and then submit predictions using that model, would it not be violating the competiton rule which states gpu-time should be less than 2 hours? Thanks again.
", "rawMarkdown": "Thank you for your reply! However, if I train my own model outside of Kaggle for more than 2 hours and then submit predictions using that model, would it not be violating the competiton rule which states gpu-time should be less than 2 hours? Thanks again.", "votes": null }, { "id": "1303463", "postDate": "05/12/2021 04:44:27", "content": "Like I think training the model for 2 or more hours might happen if you train the model with very high number of epochs and yes you can always try the Kaggle notebook editor, and try submitting the results using the same. I don't think you will have a problem . Like I used ResNet 50 v2 with some customisation it didn't take more than 40 mins so , it won't cause a problem.
", "rawMarkdown": "Like I think training the model for 2 or more hours might happen if you train the model with very high number of epochs and yes you can always try the Kaggle notebook editor, and try submitting the results using the same. I don't think you will have a problem . Like I used ResNet 50 v2 with some customisation it didn't take more than 40 mins so , it won't cause a problem.", "votes": null }, { "id": "1303474", "postDate": "05/12/2021 04:54:17", "content": "Oh I see. Sure I will try training on Kaggle itself.
", "rawMarkdown": "Oh I see. Sure I will try training on Kaggle itself.", "votes": null }, { "id": "1303479", "postDate": "05/12/2021 04:57:20", "content": "All the best !
", "rawMarkdown": "All the best !", "votes": null } ], "comments": [ { "id": 1303445, "author_name": "amartyabhattacharya", "author_url": "", "post_date": "05/12/2021 04:31:18", "content": "Yes you can use pre trained model , but I would suggest , have a primary pre-trained backbone and then make your own customisations to the model , in order to get your desired result
\nThank you
Thank you for your reply! However, if I train my own model outside of Kaggle for more than 2 hours and then submit predictions using that model, would it not be violating the competiton rule which states gpu-time should be less than 2 hours? Thanks again.
", "votes": null, "replies": [] }, { "id": 1303463, "author_name": "amartyabhattacharya", "author_url": "", "post_date": "05/12/2021 04:44:27", "content": "Like I think training the model for 2 or more hours might happen if you train the model with very high number of epochs and yes you can always try the Kaggle notebook editor, and try submitting the results using the same. I don't think you will have a problem . Like I used ResNet 50 v2 with some customisation it didn't take more than 40 mins so , it won't cause a problem.
", "votes": null, "replies": [] }, { "id": 1303474, "author_name": "rohitramaprasad", "author_url": "", "post_date": "05/12/2021 04:54:17", "content": "Oh I see. Sure I will try training on Kaggle itself.
", "votes": null, "replies": [] }, { "id": 1303479, "author_name": "amartyabhattacharya", "author_url": "", "post_date": "05/12/2021 04:57:20", "content": "All the best !
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1294249": "Hey all! I am new to Kaggle and have a genuine doubt. What are the rules regarding use of pre-trained models? Can we use any pre-trained model including ones which we have trained outside of Kaggle on other datasets? What's stopping someone from training the classifier for a longer duration outside Kaggle and then making submissions using that model? Thank you!", "1303445": "Yes you can use pre trained model , but I would suggest , have a primary pre-trained backbone and then make your own customisations to the model , in order to get your desired result\nThank you", "1303459": "Thank you for your reply! However, if I train my own model outside of Kaggle for more than 2 hours and then submit predictions using that model, would it not be violating the competiton rule which states gpu-time should be less than 2 hours? Thanks again.", "1303463": "Like I think training the model for 2 or more hours might happen if you train the model with very high number of epochs and yes you can always try the Kaggle notebook editor, and try submitting the results using the same. I don't think you will have a problem . Like I used ResNet 50 v2 with some customisation it didn't take more than 40 mins so , it won't cause a problem.", "1303474": "Oh I see. Sure I will try training on Kaggle itself.", "1303479": "All the best !" }, "source": "meta" }