Hi Team
\nI had been browsing the discussion / notebooks, but found very limited instances of PyTorch and fastai. So, wanted to check if I'm missing something.
\nDoes TF/keras have something over PyTorch/fastai for these kind of problems. I'm taking up this kind of challenge (DICOM images) for the first time… so may be a bit uninformed here.
Thanks!
", "messages": [ { "id": 1013775, "postDate": "2020-09-17T01:10:28.783Z", "content": "Hello @imnishantg. I think it's more of a preference and to choose which library you are more comfortable. You can do the same with any of them but for this magnitude of data, +-1TB probably tensorflow with its native TPU support will be fundamental in this competition to deal with this volume of data.
", "rawMarkdown": "Hello @imnishantg. I think it's more of a preference and to choose which library you are more comfortable. You can do the same with any of them but for this magnitude of data, +-1TB probably tensorflow with its native TPU support will be fundamental in this competition to deal with this volume of data. ", "replies": [ { "id": 1014235, "postDate": "2020-09-17T09:29:11.643Z", "content": "Thanks Ronaldo. This helps.
\nI actually started looking at it from that perspective. PyTorch has TPU support as well, but perhaps not as good as that of TF2 (assumption).
\nFastAI v2 is still a question mark, but will post on discussion forums if I find anything.
\nPartial to PyTorch/fastai since then it would be one less thing to learn :)
I am on the same page. I'm familiar with fastai and Pytorch. Check out the Pytorch Lightning library, which serves as a wrapper to Pytorch with TPU support, but like you said it doesn't always work as expected.
", "rawMarkdown": "I am on the same page. I'm familiar with fastai and Pytorch. Check out the Pytorch Lightning library, which serves as a wrapper to Pytorch with TPU support, but like you said it doesn't always work as expected." } ] }, { "id": 1012534, "postDate": "2020-09-16T06:28:27.743Z", "content": "Hi Team
\nI had been browsing the discussion / notebooks, but found very limited instances of PyTorch and fastai. So, wanted to check if I'm missing something.
\nDoes TF/keras have something over PyTorch/fastai for these kind of problems. I'm taking up this kind of challenge (DICOM images) for the first time… so may be a bit uninformed here.
Thanks!
", "rawMarkdown": "Hi Team\n\nI had been browsing the discussion / notebooks, but found very limited instances of PyTorch and fastai. So, wanted to check if I'm missing something.\nDoes TF/keras have something over PyTorch/fastai for these kind of problems. I'm taking up this kind of challenge (DICOM images) for the first time... so may be a bit uninformed here.\n\nThanks!" } ], "comments": [ { "id": 1013775, "author_name": "Ronaldo S.A. Batista", "author_url": "", "post_date": "2020-09-17T01:10:28.783000", "content": "Hello @imnishantg. I think it's more of a preference and to choose which library you are more comfortable. You can do the same with any of them but for this magnitude of data, +-1TB probably tensorflow with its native TPU support will be fundamental in this competition to deal with this volume of data.
", "votes": 0, "replies": [ { "id": 1014235, "author_name": "imnishantg", "author_url": "", "post_date": "2020-09-17T09:29:11.643000", "content": "Thanks Ronaldo. This helps.
\nI actually started looking at it from that perspective. PyTorch has TPU support as well, but perhaps not as good as that of TF2 (assumption).
\nFastAI v2 is still a question mark, but will post on discussion forums if I find anything.
\nPartial to PyTorch/fastai since then it would be one less thing to learn :)
I am on the same page. I'm familiar with fastai and Pytorch. Check out the Pytorch Lightning library, which serves as a wrapper to Pytorch with TPU support, but like you said it doesn't always work as expected.
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "1013775": "Hello @imnishantg. I think it's more of a preference and to choose which library you are more comfortable. You can do the same with any of them but for this magnitude of data, +-1TB probably tensorflow with its native TPU support will be fundamental in this competition to deal with this volume of data. ", "1012534": "Hi Team\n\nI had been browsing the discussion / notebooks, but found very limited instances of PyTorch and fastai. So, wanted to check if I'm missing something.\nDoes TF/keras have something over PyTorch/fastai for these kind of problems. I'm taking up this kind of challenge (DICOM images) for the first time... so may be a bit uninformed here.\n\nThanks!" } }