Congratulations to all those who moved up by miles to get into Medal positions, rest who got affected by shake up its always a learning at end of day.
\nJust for those who wish to find out the solution to get to 90.1 score on private lb. Still to know how near is 90.1 was to gold place.
\nhttps://www.kaggle.com/jaideepvalani/with-b4-pytorch-efficientnet-baseline?scriptVersionId=54134057
\nAbove is note book.
\nBelow is our solution outline
\n1)Ensemble models
\n a) B4 Fastai- Trained using Label Smoothing cross Entropy EPS 0.35
\nb) Resnext IBN/Resnext101 /ViT- Trained using Tempered Softmax loss(BiT)
\nThanks to my team mate @cswwp347724 for these models
and also @rohitsingh9990 for putting very hard efforts to help us reach 90.7 at lastday
\nIts sad that we couldnt select our best submission even though there was second thought about selecting a conservative Public LB score. But never mind we are fortunate that we are still in silver.
\nOur all models were trained using similar methodology.
I was hoping for @mobassir and @cdeotte team to have stayed at winning position or atleast in gold, but its always a good learning at the end of every competition.
\nWell done you guys !
", "messages": [ { "id": 1209928, "postDate": "2021-02-19T05:39:58.453Z", "content": "Congratulations to all those who moved up by miles to get into Medal positions, rest who got affected by shake up its always a learning at end of day.
\nJust for those who wish to find out the solution to get to 90.1 score on private lb. Still to know how near is 90.1 was to gold place.
\nhttps://www.kaggle.com/jaideepvalani/with-b4-pytorch-efficientnet-baseline?scriptVersionId=54134057
\nAbove is note book.
\nBelow is our solution outline
\n1)Ensemble models
\n a) B4 Fastai- Trained using Label Smoothing cross Entropy EPS 0.35
\nb) Resnext IBN/Resnext101 /ViT- Trained using Tempered Softmax loss(BiT)
\nThanks to my team mate @cswwp347724 for these models
and also @rohitsingh9990 for putting very hard efforts to help us reach 90.7 at lastday
\nIts sad that we couldnt select our best submission even though there was second thought about selecting a conservative Public LB score. But never mind we are fortunate that we are still in silver.
\nOur all models were trained using similar methodology.
I was hoping for @mobassir and @cdeotte team to have stayed at winning position or atleast in gold, but its always a good learning at the end of every competition.
\nWell done you guys !
", "rawMarkdown": "Congratulations to all those who moved up by miles to get into Medal positions, rest who got affected by shake up its always a learning at end of day.\n\nJust for those who wish to find out the solution to get to 90.1 score on private lb. Still to know how near is 90.1 was to gold place.\n\nhttps://www.kaggle.com/jaideepvalani/with-b4-pytorch-efficientnet-baseline?scriptVersionId=54134057\n\nAbove is note book. \n\nBelow is our solution outline\n1)Ensemble models \n a) B4 Fastai- Trained using Label Smoothing cross Entropy EPS 0.35\nb) Resnext IBN/Resnext101 /ViT- Trained using Tempered Softmax loss(BiT) \nThanks to my team mate @cswwp347724 for these models\n\n and also @rohitsingh9990 for putting very hard efforts to help us reach 90.7 at lastday \n\n\nIts sad that we couldnt select our best submission even though there was second thought about selecting a conservative Public LB score. But never mind we are fortunate that we are still in silver. \nOur all models were trained using similar methodology.\n\nI was hoping for @mobassir and @cdeotte team to have stayed at winning position or atleast in gold, but its always a good learning at the end of every competition.\n\nWell done you guys !\n\n", "votes": 10 }, { "id": 1215000, "postDate": "2021-02-23T08:56:52.153Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 1215000, "author_name": "", "author_url": "", "post_date": "2021-02-23T08:56:52.153000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1209928": "Congratulations to all those who moved up by miles to get into Medal positions, rest who got affected by shake up its always a learning at end of day.\n\nJust for those who wish to find out the solution to get to 90.1 score on private lb. Still to know how near is 90.1 was to gold place.\n\nhttps://www.kaggle.com/jaideepvalani/with-b4-pytorch-efficientnet-baseline?scriptVersionId=54134057\n\nAbove is note book. \n\nBelow is our solution outline\n1)Ensemble models \n a) B4 Fastai- Trained using Label Smoothing cross Entropy EPS 0.35\nb) Resnext IBN/Resnext101 /ViT- Trained using Tempered Softmax loss(BiT) \nThanks to my team mate @cswwp347724 for these models\n\n and also @rohitsingh9990 for putting very hard efforts to help us reach 90.7 at lastday \n\n\nIts sad that we couldnt select our best submission even though there was second thought about selecting a conservative Public LB score. But never mind we are fortunate that we are still in silver. \nOur all models were trained using similar methodology.\n\nI was hoping for @mobassir and @cdeotte team to have stayed at winning position or atleast in gold, but its always a good learning at the end of every competition.\n\nWell done you guys !\n\n", "1215000": "" } }