Hi everyone,
My first post here, I watched the first video and did the exercise with some data I scrapped.
I just downloaded 30 images of 9 types of fruit/vegs so 270 in total. I ran resnet 50 and got the best result of 7.4%.
When doing interp.most_confused() I get 4 mistakes in total, that means 100% is 54 images.
54 is 20% of 270 so this is standard split of 80/20 then?
Is there any way to change this ratio?
Coming from scikit I wonder do you never score the train set is it always directly on test?
I guess some of this will be covered in later videos as well…