num_workers=args.workers,
pin_memory=True,
sampler=RangeSampler(args.range))
#Load training data
hardModel = ResNet18(num_classes=200)
hardModel = HardNBDT(pretrained=True,
dataset='TinyImagenet200',
arch='ResNet18',
model=hardModel)
hardModel = hardModel.cuda()
softModel = ResNet18(num_classes=200)
softModel = SoftNBDT(pretrained=True,
dataset='TinyImagenet200',
arch='ResNet18',
model=softModel)
softModel = softModel.cuda()
explainer = Explainer(hardModel, (64, 64), 500)
explainer.generate_masks(1000, 8, 0.1, 'temp.npy')
klen = 11
ksig = 5
kern = gkern(klen, ksig)
blur = lambda x: nn.functional.conv2d(x, kern, padding=klen // 2)
insertion = CausalMetric(softModel,
'ins',
64 * 8,
batch_size=batch_size,
shuffle=False,
num_workers=1,
pin_memory=True,
sampler=RangeSampler(range_sample))
hardModel = wrn28_10_cifar10()
hardModel = HardNBDT(pretrained=True,
dataset='CIFAR10',
arch='wrn28_10_cifar10',
model=hardModel)
hardModel = hardModel.cuda()
softModel = wrn28_10_cifar10()
softModel = SoftNBDT(pretrained=True,
dataset='CIFAR10',
arch='wrn28_10_cifar10',
model=softModel)
softModel = softModel.cuda()
explainer = Explainer(hardModel, (32, 32), 500)
explainer.generate_masks(1000, 8, 0.1, 'temp.npy')
klen = 11
ksig = 5
kern = gkern(klen, ksig)
blur = lambda x: nn.functional.conv2d(x, kern, padding=klen // 2)
insertion = CausalMetric(softModel,
'ins',
32 * 8,
import streamlit as st
import numpy as np
import os
from nbdt.model import SoftNBDT
from nbdt.models import ResNet18, wrn28_10_cifar10, wrn28_10_cifar100, wrn28_10 # use wrn28_10 for TinyImagenet200
from torchvision import transforms
from nbdt.utils import DATASET_TO_CLASSES, load_image_from_path, maybe_install_wordnet
from IPython.display import display
model = wrn28_10_cifar10()
model = SoftNBDT(pretrained=True,
dataset='CIFAR10',
arch='wrn28_10_cifar10',
model=model,
hierarchy='wordnet')
transforms = transforms.Compose([
transforms.Resize(32),
transforms.CenterCrop(32),
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010)),
])
############### streamlit stuff ###############
def file_selector(folder_path='.'):
filenames = os.listdir(folder_path)
selected_filename = st.selectbox('Select a file', filenames)
return os.path.join(folder_path, selected_filename)
testloader = torch.utils.data.DataLoader(testset, batch_size=100, shuffle=False, num_workers=2)
Colors.cyan(f'Training with dataset {args.dataset} and {len(trainset.classes)} classes')
# Model
print('==> Building model..')
model = getattr(models, args.arch)
model_kwargs = {'num_classes': len(trainset.classes) }
if args.pretrained:
print('==> Loading pretrained model..')
net = model(pretrained=True, dataset=args.dataset, **model_kwargs)
else:
print('==> Loading NBDT model..')
net = ResNet18()
net = SoftNBDT(pretrained=True, dataset='CIFAR10', arch='ResNet18', model=net)
#Analyzer
class_analysis = getattr(analysis, args.analysis or 'Noop')
analyzer_kwargs = generate_kwargs(args, class_analysis,
name=f'Analyzer {args.analysis}',
keys=analysis.keys,
globals=globals())
analyzer = class_analysis(**analyzer_kwargs)
#Criterion
criterion = nn.CrossEntropyLoss()
class_criterion = getattr(loss, args.loss)
loss_kwargs = generate_kwargs(args, class_criterion,
name=f'Loss {args.loss}',
keys=loss.keys,
import streamlit as st
import numpy as np
import os
from nbdt.model import SoftNBDT
from nbdt.models import ResNet18, wrn28_10_cifar10, wrn28_10_cifar100, wrn28_10 # use wrn28_10 for TinyImagenet200
from torchvision import transforms
from nbdt.utils import DATASET_TO_CLASSES, load_image_from_path, maybe_install_wordnet
from IPython.display import display
import base64
model = wrn28_10_cifar10()
model = SoftNBDT(
pretrained=True,
dataset='CIFAR10',
arch='wrn28_10_cifar10',
model=model,
hierarchy='wordnet')
def set_bg_hack():
# set bg name
main_bg = "nbdt_bg.png"
main_bg_ext = "png"
# we can add a side bg if necessary
# side_bg = "sample.jpg"
# side_bg_ext = "jpg"
st.markdown(
f"""
def test_nbdt_soft_cifar10(input_cifar10, resnet18_cifar10):
model_soft = SoftNBDT(dataset="CIFAR10",
model=resnet18_cifar10,
hierarchy="induced")
model_soft(input_cifar10)
def test_nbdt_soft_tinyimagenet200(input_tinyimagenet200,
resnet18_tinyimagenet200):
model_soft = SoftNBDT(dataset="TinyImagenet200",
model=resnet18_tinyimagenet200,
hierarchy="induced")
model_soft(input_tinyimagenet200)