I attempted to create a learning rate with the following code
def get_data(sz):
tfms = tfms_from_model(f_model, sz, aug_tfms=transforms_top_down, max_zoom=1.05)
return ImageClassifierData.from_csv(PATH, 'boneage-training-dataset', label_csv ,
bs = 64, tfms= tfms, val_idxs=val_idxs, suffix='.png',
test_name=None, continuous=False, skip_header=True, num_workers=4)
data = get_data(256)
learn = ConvLearner.pretrained(arch,data,precompute=True)
100%|██████████| 158/158 [02:46<00:00, 1.05s/it]
100%|██████████| 40/40 [00:42<00:00, 1.07s/it]
Why does this code run 2 runs — 158 and 40?
learn.sched.plot_lr()
Is realistic – But why does my loss curve have such a weird shape – Would you choose a learning rate of 0.1 in this case?