MercyMarkus

p,t = learn.get_preds(ds_type=DatasetType.Test)

p = to_np(p); 
p.shape

ids = np.array([f.name for f in (combined_test)]);
ids.shape

sample_sub = Path('data/SampleSubmission.csv')
df_sample = pd.read_csv(sample_sub)

MercyMarkus

learn.fit_one_cycle(5, slice(1e-4))

MercyMarkus

tfms = get_transforms(flip_vert=True, max_lighting=0.1, max_zoom=1.05, max_warp=0.)
data = (src.transform(tfms, size=256)
        .databunch().normalize(imagenet_stats))

learn.data = data

data.train_ds[0][0].shape

MercyMarkus

learn.fit_one_cycle(5, 1e-6)

MercyMarkus

tfms = get_transforms(flip_vert=True, max_lighting=0.1, max_zoom=1.05, max_warp=0.)
data = (src.transform(tfms, size=200)
        .databunch().normalize(imagenet_stats))

learn.data = data

data.train_ds[0][0].shape

MercyMarkus

probs,val_labels = learn.get_preds(ds_type=DatasetType.Valid) 
print('Accuracy',accuracy(probs,val_labels)),
print('Error Rate', error_rate(probs, val_labels))
print('AUC', auc_score(probs,val_labels))

MercyMarkus

learn.fit_one_cycle(7, max_lr=slice(1e-6,1e-4))

MercyMarkus

interp = ClassificationInterpretation.from_learner(learn)
interp.plot_confusion_matrix(dpi=120)

MercyMarkus

learn.save('resnet50-stg1')

MercyMarkus

from sklearn.metrics import roc_auc_score
def auc_score(y_score,y_true):
    return torch.tensor(roc_auc_score(y_true,y_score[:,1]))
probs,val_labels = learn.get_preds(ds_type=DatasetType.Valid) 
print('Accuracy',accuracy(probs,val_labels)),
print('Error Rate', error_rate(probs, val_labels))
print('AUC', auc_score(probs,val_labels))