Very excited when private LB release the result and some top winner share their approaches to solve this problem.
\n\nAnyone willing to share notebook that is using FastAI with EfficientNet and train with regression method?
\n\nI find it difficult to train EfficientNet in FastAI .
", "messages": [ { "id": "621429", "postDate": "09/08/2019 14:06:08", "content": "Very excited when private LB release the result and some top winner share their approaches to solve this problem.
\n\nAnyone willing to share notebook that is using FastAI with EfficientNet and train with regression method?
\n\nI find it difficult to train EfficientNet in FastAI .
", "rawMarkdown": "Very excited when private LB release the result and some top winner share their approaches to solve this problem.\n\nAnyone willing to share notebook that is using FastAI with EfficientNet and train with regression method? \n\nI find it difficult to train EfficientNet in FastAI .", "votes": null }, { "id": "621563", "postDate": "09/08/2019 17:00:12", "content": "I think @drhabib used FastAI: \nhttps://www.kaggle.com/c/aptos2019-blindness-detection/discussion/108030#latest-621520
", "rawMarkdown": "I think @drhabib used FastAI: \nhttps://www.kaggle.com/c/aptos2019-blindness-detection/discussion/108030#latest-621520", "votes": null }, { "id": "621565", "postDate": "09/08/2019 17:03:06", "content": "and here is a notebook you are looking for: https://www.kaggle.com/drhabib/starter-kernel-for-0-79\nI used this model to train further on 2019 data and get lb-0.797(Pr_lb - ~0.917) but did not select it for final 2 submissions
", "rawMarkdown": "and here is a notebook you are looking for: https://www.kaggle.com/drhabib/starter-kernel-for-0-79\nI used this model to train further on 2019 data and get lb-0.797(Pr_lb - ~0.917) but did not select it for final 2 submissions", "votes": null }, { "id": "621585", "postDate": "09/08/2019 17:20:19", "content": "Would like to hear about that as well
", "rawMarkdown": "Would like to hear about that as well", "votes": null }, { "id": "621588", "postDate": "09/08/2019 17:24:23", "content": "I used fastai =) You can read here
", "rawMarkdown": "I used fastai =) You can read [here](https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/108030#latest-621520)", "votes": null }, { "id": "621638", "postDate": "09/08/2019 19:26:52", "content": "I used fastai and got 19th position ;)\nIt was a handy and powerfull tool, however I struggled a bit when I required using neural net architectures with higher input resolution. I hope they add more pretrained models to use them easily ;)
", "rawMarkdown": "I used fastai and got 19th position ;)\nIt was a handy and powerfull tool, however I struggled a bit when I required using neural net architectures with higher input resolution. I hope they add more pretrained models to use them easily ;)", "votes": null }, { "id": "621641", "postDate": "09/08/2019 19:27:50", "content": "https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/108057#latest-621632 we used fastai as well
", "rawMarkdown": "https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/108057#latest-621632 we used fastai as well", "votes": null }, { "id": "622048", "postDate": "09/09/2019 08:05:50", "content": "I used fastai + efficientnet + regression. I trained Efficientnet B1 to B7 using fastai for my experiments. There were even some public kernels using fastai + efficientnet.
", "rawMarkdown": "I used fastai + efficientnet + regression. I trained Efficientnet B1 to B7 using fastai for my experiments. There were even some public kernels using fastai + efficientnet.", "votes": null }, { "id": "622833", "postDate": "09/10/2019 06:19:51", "content": "I definitely tried on @drhabib kernel, but when i uncomment
\n\nlearn.fit_one_cycle(10, 1e-3)
It will failed after few epochs when running in Kaggle kernel.
\nHas anyone can fully execute learn.fit_one_cycle() in DrHB kernel ?
Btw, can we directly use DrHB trained weights to continue train on other model ?
", "rawMarkdown": "Btw, can we directly use DrHB trained weights to continue train on other model ?", "votes": null }, { "id": "623050", "postDate": "09/10/2019 12:22:53", "content": "I used Fast.ai for all my models. I've shared my kernels here: https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/107947#latest-620941
\n\nI did find myself fighting with it a little bit, especially as I started to customise more and more of my pipeline.
", "rawMarkdown": "I used Fast.ai for all my models. I've shared my kernels here: https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/107947#latest-620941\n\nI did find myself fighting with it a little bit, especially as I started to customise more and more of my pipeline.", "votes": null } ], "comments": [ { "id": 621563, "author_name": "anuragtr", "author_url": "", "post_date": "09/08/2019 17:00:12", "content": "I think @drhabib used FastAI: \nhttps://www.kaggle.com/c/aptos2019-blindness-detection/discussion/108030#latest-621520
", "votes": null, "replies": [] }, { "id": 621565, "author_name": "anuragtr", "author_url": "", "post_date": "09/08/2019 17:03:06", "content": "and here is a notebook you are looking for: https://www.kaggle.com/drhabib/starter-kernel-for-0-79\nI used this model to train further on 2019 data and get lb-0.797(Pr_lb - ~0.917) but did not select it for final 2 submissions
", "votes": null, "replies": [ { "id": 622834, "author_name": "ccjoshua", "author_url": "", "post_date": "09/10/2019 06:20:45", "content": "Btw, can we directly use DrHB trained weights to continue train on other model ?
", "votes": null, "replies": [] } ] }, { "id": 621585, "author_name": "chrisfs", "author_url": "", "post_date": "09/08/2019 17:20:19", "content": "Would like to hear about that as well
", "votes": null, "replies": [ { "id": 621588, "author_name": "drhabib", "author_url": "", "post_date": "09/08/2019 17:24:23", "content": "I used fastai =) You can read here
", "votes": null, "replies": [] }, { "id": 622833, "author_name": "ccjoshua", "author_url": "", "post_date": "09/10/2019 06:19:51", "content": "I definitely tried on @drhabib kernel, but when i uncomment
\n\nlearn.fit_one_cycle(10, 1e-3)
It will failed after few epochs when running in Kaggle kernel.
\nHas anyone can fully execute learn.fit_one_cycle() in DrHB kernel ?
I used fastai and got 19th position ;)\nIt was a handy and powerfull tool, however I struggled a bit when I required using neural net architectures with higher input resolution. I hope they add more pretrained models to use them easily ;)
", "votes": null, "replies": [] }, { "id": 621641, "author_name": "insaff", "author_url": "", "post_date": "09/08/2019 19:27:50", "content": "https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/108057#latest-621632 we used fastai as well
", "votes": null, "replies": [] }, { "id": 622048, "author_name": "harshthaker", "author_url": "", "post_date": "09/09/2019 08:05:50", "content": "I used fastai + efficientnet + regression. I trained Efficientnet B1 to B7 using fastai for my experiments. There were even some public kernels using fastai + efficientnet.
", "votes": null, "replies": [] }, { "id": 623050, "author_name": "lextoumbourou", "author_url": "", "post_date": "09/10/2019 12:22:53", "content": "I used Fast.ai for all my models. I've shared my kernels here: https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/107947#latest-620941
\n\nI did find myself fighting with it a little bit, especially as I started to customise more and more of my pipeline.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "621429": "Very excited when private LB release the result and some top winner share their approaches to solve this problem.\n\nAnyone willing to share notebook that is using FastAI with EfficientNet and train with regression method? \n\nI find it difficult to train EfficientNet in FastAI .", "621563": "I think @drhabib used FastAI: \nhttps://www.kaggle.com/c/aptos2019-blindness-detection/discussion/108030#latest-621520", "621565": "and here is a notebook you are looking for: https://www.kaggle.com/drhabib/starter-kernel-for-0-79\nI used this model to train further on 2019 data and get lb-0.797(Pr_lb - ~0.917) but did not select it for final 2 submissions", "621585": "Would like to hear about that as well", "621588": "I used fastai =) You can read [here](https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/108030#latest-621520)", "621638": "I used fastai and got 19th position ;)\nIt was a handy and powerfull tool, however I struggled a bit when I required using neural net architectures with higher input resolution. I hope they add more pretrained models to use them easily ;)", "621641": "https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/108057#latest-621632 we used fastai as well", "622048": "I used fastai + efficientnet + regression. I trained Efficientnet B1 to B7 using fastai for my experiments. There were even some public kernels using fastai + efficientnet.", "622833": "I definitely tried on @drhabib kernel, but when i uncomment \n\n`learn.fit_one_cycle(10, 1e-3)`\n\nIt will failed after few epochs when running in Kaggle kernel. \nHas anyone can fully execute learn.fit_one_cycle() in DrHB kernel ?", "622834": "Btw, can we directly use DrHB trained weights to continue train on other model ?", "623050": "I used Fast.ai for all my models. I've shared my kernels here: https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/107947#latest-620941\n\nI did find myself fighting with it a little bit, especially as I started to customise more and more of my pipeline." }, "source": "meta" }