I just finished lesson 2 and attempted the dog breed competition with results that were in-line with Jeremy’s from the video; I’m now attempting to do the same for the Humpback Whale comp (https://www.kaggle.com/c/whale-categorization-playground) but my results are terrible (.11) accuracy
Does anyone have any tips on how I should go about structuring my workflow? As of now I’m doing this:
(1) Basic data exploration (size of images, what does it look like)
(2) Test on simple model (i.e. no lr_finder, no precompute or training earlier layers)
(3) Precompute = False and training earlier layers (as found in Jeremys’ instructions)
Any tips would be rad!