def test_anchor_images():
os.environ.clear()
alibi_model = os.path.join(
kfserving.Storage.download(IMAGENET_EXPLAINER_URI), EXPLAINER_FILENAME)
with open(alibi_model, "rb") as f:
model = InceptionV3(weights="imagenet")
predictor = lambda x: model.predict(x) # pylint:disable=unnecessary-lambda
alibi_model = dill.load(f)
anchor_images = AnchorImages(predictor,
alibi_model,
batch_size=25,
stop_on_first=True)
category = "Persian cat"
image_shape = (299, 299, 3)
data, _ = fetch_imagenet(category,
nb_images=10,
target_size=image_shape[:2],
seed=2,
return_X_y=True)
images = preprocess_input(data)
print(images.shape)
np.random.seed(0)
explanation = anchor_images.explain(images[0:1])
exp_json = json.loads(explanation.to_json())
assert exp_json["data"]["precision"] > 0.9
def test_cifar10_images(): # pylint: disable-msg=too-many-locals
alibi_model = make_anchor_image()
anchor_images = AnchorImages(alibi_model)
_, test = tf.keras.datasets.cifar10.load_data()
X_test, _ = test
X_test = X_test.astype("float32") / 255
idx = 12
test_example = X_test[idx : idx + 1]
np.random.seed(0)
explanation = anchor_images.explain(test_example)
exp_json = json.loads(explanation.to_json())
assert exp_json["data"]["precision"] > 0.9
def test_anchor_images():
os.environ.clear()
alibi_model = os.path.join(
kfserving.Storage.download(IMAGENET_EXPLAINER_URI), EXPLAINER_FILENAME)
with open(alibi_model, "rb") as f:
model = InceptionV3(weights="imagenet")
predictor = lambda x: model.predict(x) # pylint:disable=unnecessary-lambda
alibi_model = dill.load(f)
anchor_images = AnchorImages(predictor,
alibi_model,
batch_size=25,
stop_on_first=True)
category = "Persian cat"
image_shape = (299, 299, 3)
# the image downloader comes from seldonio/alibi
# https://github.com/SeldonIO/alibi/blob/76e6192b6d78848dd47c11ba6f6348ca94c424c6/alibi/datasets.py#L104-L125
img_urls = json.load(open('alibiexplainer/tests/persian_cat.json'))
seed = 2
random.seed(seed)
random.shuffle(img_urls)
data = []
nb = 0
nb_images = 10
target_size = image_shape[:2]
min_std = 10.
for img_url in img_urls:
try:
resp = requests.get(img_url, timeout=2)
resp.raise_for_status()
except RequestException:
continue
try:
image = PIL.Image.open(BytesIO(resp.content)).convert('RGB')
except OSError:
continue
image = np.expand_dims(image.resize(target_size), axis=0)
if np.std(image) < min_std: # do not include empty images
continue
data.append(image)
nb += 1
if nb == nb_images:
break
data = np.concatenate(data, axis=0)
images = preprocess_input(data)
print(images.shape)
np.random.seed(0)
explanation = anchor_images.explain(images[0:1])
exp_json = json.loads(explanation.to_json())
assert exp_json["data"]["precision"] > 0.9
def test_cifar10_images(): # pylint: disable-msg=too-many-locals
alibi_model = os.path.join(
kfserving.Storage.download(CIFAR10_EXPLAINER_URI), EXPLAINER_FILENAME
)
with open(alibi_model, "rb") as f:
alibi_model = dill.load(f)
url = "https://storage.googleapis.com/seldon-models/alibi-detect/classifier/"
path_model = os.path.join(url, "cifar10", "resnet32", "model.h5")
save_path = tf.keras.utils.get_file("resnet32", path_model)
model = tf.keras.models.load_model(save_path)
_, test = tf.keras.datasets.cifar10.load_data()
X_test, _ = test
X_test = X_test.astype("float32") / 255
idx = 12
test_example = X_test[idx: idx + 1]
anchor_images = AnchorImages(
lambda x: model.predict(x), alibi_model) # pylint: disable-msg=unnecessary-lambda
np.random.seed(0)
explanation = anchor_images.explain(test_example)
exp_json = json.loads(explanation.to_json())
assert exp_json["data"]["precision"] > 0.9