I want to share my baseline notebooks with you. This is a bare-bones PyTorch solution with no much hyperparameter tuning. I believe you can easily drop the score down a bit with some hyperparameters magic.
\nTried to comment the solution as much as possible to make it useful for beginner ML practitioners.
\nHope you'll find it useful!
", "messages": [ { "id": "1917244", "postDate": "08/28/2022 15:15:44", "content": "I want to share my baseline notebooks with you. This is a bare-bones PyTorch solution with no much hyperparameter tuning. I believe you can easily drop the score down a bit with some hyperparameters magic.
\nTried to comment the solution as much as possible to make it useful for beginner ML practitioners.
\nHope you'll find it useful!
", "rawMarkdown": "I want to share my baseline notebooks with you. This is a bare-bones PyTorch solution with no much hyperparameter tuning. I believe you can easily drop the score down a bit with some hyperparameters magic. \n\n* Train: [PyTorch-EffNetV2 baseline CV:0.49](https://www.kaggle.com/vslaykovsky/train-pytorch-effnetv2-baseline-cv-0-49)\n* Inference: [PyTorch-EffNetV2 single-model PL:0.49; 5-fold ensemble PL:0.47](https://www.kaggle.com/code/vslaykovsky/infer-pytorch-effnetv2-single-model-pl-0-49) \n\nTried to comment the solution as much as possible to make it useful for beginner ML practitioners. \n\nHope you'll find it useful!", "votes": null }, { "id": "1917303", "postDate": "08/28/2022 16:02:31", "content": "great work! thanks for sharing
", "rawMarkdown": "great work! thanks for sharing", "votes": null }, { "id": "1917361", "postDate": "08/28/2022 16:57:31", "content": "Does your \"single model\" means 1 fold or 5-folds combined?
", "rawMarkdown": "Does your \"single model\" means 1 fold or 5-folds combined?", "votes": null }, { "id": "1917369", "postDate": "08/28/2022 17:05:03", "content": "Nice & Clean baseline. Thanks for sharing to the community.
", "rawMarkdown": "Nice & Clean baseline. Thanks for sharing to the community.", "votes": null }, { "id": "1917375", "postDate": "08/28/2022 17:11:15", "content": "Single model is a model that is trained on all train data folds. It is stored as 'effnetv2' in the training notebook. 5 models for each of 5 folds are also trained mainly for CV score.
", "rawMarkdown": "Single model is a model that is trained on all train data folds. It is stored as 'effnetv2' in the training notebook. 5 models for each of 5 folds are also trained mainly for CV score.", "votes": null }, { "id": "1917693", "postDate": "08/29/2022 01:39:29", "content": "my forked submission timeout.
", "rawMarkdown": "my forked submission timeout.", "votes": null }, { "id": "1917896", "postDate": "08/29/2022 06:37:05", "content": "Hi. This is because MODEL_NAMES was set to all models. Try to use the single model in MODEL_NAMES
", "rawMarkdown": "Hi. This is because MODEL_NAMES was set to all models. Try to use the single model in MODEL_NAMES", "votes": null }, { "id": "1917924", "postDate": "08/29/2022 06:54:14", "content": "Thanks for sharing. may I ask how long it took you for tranning and submission?
", "rawMarkdown": "Thanks for sharing. may I ask how long it took you for tranning and submission?", "votes": null }, { "id": "1918302", "postDate": "08/29/2022 13:30:04", "content": "Training locally on a laptop approx 4 hours. But a single model training is just around 1hour on a 3080 laptop. Submission is approx 3-7 hours, depends on the number of models to use
", "rawMarkdown": "Training locally on a laptop approx 4 hours. But a single model training is just around 1hour on a 3080 laptop. Submission is approx 3-7 hours, depends on the number of models to use", "votes": null }, { "id": "1918319", "postDate": "08/29/2022 13:46:33", "content": "Very helpful, thanks again.
", "rawMarkdown": "Very helpful, thanks again.", "votes": null }, { "id": "1939085", "postDate": "09/14/2022 14:18:39", "content": "How much time did it take for a single epoch ?
", "rawMarkdown": "How much time did it take for a single epoch ?", "votes": null }, { "id": "1985137", "postDate": "10/13/2022 06:02:49", "content": "Thanks for sharing. these codes are useful for me, but i got some trouble.
\n When i run the code of train using GPU model, i got this error: AttributeError: module 'torchvision.models.efficientnet' has no attribute 'EfficientNet_V2_S_Weights'. Could you tell me how can i solve it? Thanks in advance.
Hi. Make sure that the right version of Pytorch is install via pip install ... .whl command
Are you using transfer learning from the pretrained model? It seems that the layers of the effnet model are not frozen.
", "rawMarkdown": "Are you using transfer learning from the pretrained model? It seems that the layers of the effnet model are not frozen.", "votes": null } ], "comments": [ { "id": 1917303, "author_name": "dragonzhang", "author_url": "", "post_date": "08/28/2022 16:02:31", "content": "great work! thanks for sharing
", "votes": null, "replies": [] }, { "id": 1917361, "author_name": "harshitsheoran", "author_url": "", "post_date": "08/28/2022 16:57:31", "content": "Does your \"single model\" means 1 fold or 5-folds combined?
", "votes": null, "replies": [ { "id": 1917375, "author_name": "vslaykovsky", "author_url": "", "post_date": "08/28/2022 17:11:15", "content": "Single model is a model that is trained on all train data folds. It is stored as 'effnetv2' in the training notebook. 5 models for each of 5 folds are also trained mainly for CV score.
", "votes": null, "replies": [] }, { "id": 1917693, "author_name": "dragonzhang", "author_url": "", "post_date": "08/29/2022 01:39:29", "content": "my forked submission timeout.
", "votes": null, "replies": [] }, { "id": 1917896, "author_name": "vslaykovsky", "author_url": "", "post_date": "08/29/2022 06:37:05", "content": "Hi. This is because MODEL_NAMES was set to all models. Try to use the single model in MODEL_NAMES
", "votes": null, "replies": [] } ] }, { "id": 1917369, "author_name": "nischaydnk", "author_url": "", "post_date": "08/28/2022 17:05:03", "content": "Nice & Clean baseline. Thanks for sharing to the community.
", "votes": null, "replies": [] }, { "id": 1917924, "author_name": "meli19", "author_url": "", "post_date": "08/29/2022 06:54:14", "content": "Thanks for sharing. may I ask how long it took you for tranning and submission?
", "votes": null, "replies": [ { "id": 1918302, "author_name": "vslaykovsky", "author_url": "", "post_date": "08/29/2022 13:30:04", "content": "Training locally on a laptop approx 4 hours. But a single model training is just around 1hour on a 3080 laptop. Submission is approx 3-7 hours, depends on the number of models to use
", "votes": null, "replies": [] }, { "id": 1918319, "author_name": "meli19", "author_url": "", "post_date": "08/29/2022 13:46:33", "content": "Very helpful, thanks again.
", "votes": null, "replies": [] }, { "id": 1939085, "author_name": "mithilsalunkhe", "author_url": "", "post_date": "09/14/2022 14:18:39", "content": "How much time did it take for a single epoch ?
", "votes": null, "replies": [] } ] }, { "id": 1985137, "author_name": "zzl847245720", "author_url": "", "post_date": "10/13/2022 06:02:49", "content": "Thanks for sharing. these codes are useful for me, but i got some trouble.
\n When i run the code of train using GPU model, i got this error: AttributeError: module 'torchvision.models.efficientnet' has no attribute 'EfficientNet_V2_S_Weights'. Could you tell me how can i solve it? Thanks in advance.
Hi. Make sure that the right version of Pytorch is install via pip install ... .whl command
Are you using transfer learning from the pretrained model? It seems that the layers of the effnet model are not frozen.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1917244": "I want to share my baseline notebooks with you. This is a bare-bones PyTorch solution with no much hyperparameter tuning. I believe you can easily drop the score down a bit with some hyperparameters magic. \n\n* Train: [PyTorch-EffNetV2 baseline CV:0.49](https://www.kaggle.com/vslaykovsky/train-pytorch-effnetv2-baseline-cv-0-49)\n* Inference: [PyTorch-EffNetV2 single-model PL:0.49; 5-fold ensemble PL:0.47](https://www.kaggle.com/code/vslaykovsky/infer-pytorch-effnetv2-single-model-pl-0-49) \n\nTried to comment the solution as much as possible to make it useful for beginner ML practitioners. \n\nHope you'll find it useful!", "1917303": "great work! thanks for sharing", "1917361": "Does your \"single model\" means 1 fold or 5-folds combined?", "1917369": "Nice & Clean baseline. Thanks for sharing to the community.", "1917375": "Single model is a model that is trained on all train data folds. It is stored as 'effnetv2' in the training notebook. 5 models for each of 5 folds are also trained mainly for CV score.", "1917693": "my forked submission timeout.", "1917896": "Hi. This is because MODEL_NAMES was set to all models. Try to use the single model in MODEL_NAMES", "1917924": "Thanks for sharing. may I ask how long it took you for tranning and submission?", "1918302": "Training locally on a laptop approx 4 hours. But a single model training is just around 1hour on a 3080 laptop. Submission is approx 3-7 hours, depends on the number of models to use", "1918319": "Very helpful, thanks again.", "1939085": "How much time did it take for a single epoch ?", "1985137": "Thanks for sharing. these codes are useful for me, but i got some trouble. \n When i run the code of train using GPU model, i got this error: AttributeError: module 'torchvision.models.efficientnet' has no attribute 'EfficientNet_V2_S_Weights'. Could you tell me how can i solve it? Thanks in advance.", "1988652": "Hi. Make sure that the right version of Pytorch is install via `pip install ... .whl` command", "1989460": "Are you using transfer learning from the pretrained model? It seems that the layers of the effnet model are not frozen." }, "source": "meta" }