While doing the dog breeds classification task, after running the initial training with
precompute=True and then setting it to
false I tried rerunning the LR finder but get results like this:
which make sense I guess since the model has been saturated. So how do you choose the LR now since from the graph it is not obvious which to pick? Just continue with the same one from before?
Edit: Similarly when finding LR after unfreezing the model this is the result (hard to know which one to pick):