tensorflow_1_2.py

def read_labels_from_file(data_root, label_path):
    data = pd.read_csv(label_path)
    data['fn'] = data.id.map(lambda x: os.path.join(
        data_root, 'train', x + '.jpg'))
    int_to_breed = data.breed.unique()
    breed_to_int = dict((v, k) for k, v in enumerate(int_to_breed))
    data['int_breed'] = data.breed.map(lambda x: breed_to_int[x])
    
    return data.fn, data.int_breed, int_to_breed, breed_to_int