Need advice with training on a pretty big image dataset

Part 1 (2019)
1

Hello people.
I’m presently training the food-101 dataset that classifies 101 dishes. However the dataset is massive, and every epoch takes 30 mins on Colab

Is there a way (in fastai/ logically speaking) to sample a subset of the dataset or should I do it manually?
Thanks!

1 Like
2

Yes, you can sample a random subset use_partial_data or create your own filter using the data block api: https://docs.fast.ai/data_block.html

2 Likes
3

Thank you! That was very helpful!