- batch size, start at 64 till out of memory, then half , and half and half and so on
Increasing size
pass in a larger sized data (images)through a pretrained model
starting training on small images, then start on larger images, avoid over fitting.
under fitting validation loss is a lot less then training set loss => increase cycle_len
Reading Information from csv Lesson 3 34:27
data=ImageClassifier.from_csv
Using panads for writing txt file lesson 3 35:31
Lesson 2 notebook starts at 1:53
satellite images
aws set 1:59