def test_model_joblib():
sklearn_model, data = _train_sample_model()
model_file = os.path.join(JOBLIB_FILE[0], JOBLIB_FILE[1])
joblib.dump(value=sklearn_model, filename=model_file)
model = SKLearnModel("model", JOBLIB_FILE[0])
model.load()
request = data[0:1].tolist()
response = model.predict({"instances": request})
assert response["predictions"] == [0]
def _run_pickle_model(model_dir, model_name):
sklearn_model, data = _train_sample_model()
model_file = os.path.join(model_dir, model_name)
pickle.dump(sklearn_model, open(model_file, 'wb'))
model = SKLearnModel("model", model_dir)
model.load()
request = data[0:1].tolist()
response = model.predict({"instances": request})
assert response["predictions"] == [0]
def test_model():
iris = datasets.load_iris()
X, y = iris.data, iris.target
sklearn_model = svm.SVC(gamma='scale')
sklearn_model.fit(X, y)
model_file = os.path.join((model_dir), JOBLIB_FILE)
joblib.dump(value=sklearn_model, filename=model_file)
server = SKLearnModel("sklearnmodel", model_dir)
server.load()
request = X[0:1].tolist()
response = server.predict(request)
assert response == [0]
def test_mixedtype_model_joblib():
model = SKLearnModel("model", MIXEDTYPE_DIR)
model.load()
request = [{
'MSZoning': 'RL',
'LotArea': 8450,
'LotShape': 'Reg',
'Utilities': 'AllPub',
'YrSold': 2008,
'Neighborhood': 'CollgCr',
'OverallQual': 7,
'YearBuilt': 2003,
'SaleType': 'WD',
'GarageArea': 548
}]
response = model.predict({"instances": request})
assert response["predictions"] == [12.202832815138274]
async def load(self, name: str) -> bool:
model = SKLearnModel(name, os.path.join(self.models_dir, name))
if model.load():
self.update(model)
return model.ready
# limitations under the License.
import argparse
import logging
import sys
import kfserving
from .kfserver import KFServer
from sklearnserver import SKLearnModel, SKLearnModelRepository
DEFAULT_MODEL_NAME = "model"
DEFAULT_LOCAL_MODEL_DIR = "/tmp/model"
parser = argparse.ArgumentParser(parents=[kfserving.kfserver.parser])
parser.add_argument('--model_dir', required=True,
help='A URI pointer to the model binary')
parser.add_argument('--model_name', default=DEFAULT_MODEL_NAME,
help='The name that the model is served under.')
args, _ = parser.parse_known_args()
if __name__ == "__main__":
model = SKLearnModel(args.model_name, args.model_dir)
try:
model.load()
except Exception:
ex_type, ex_value, _ = sys.exc_info()
logging.error(f"fail to load model {args.model_name} from dir {args.model_dir}. "
f"exception type {ex_type}, exception msg: {ex_value}")
model.ready = False
KFServer(registered_models=SKLearnModelRepository(args.model_dir)).start([model] if model.ready else [])
def test_dir_with_two_models():
model = SKLearnModel("model", MULTI_DIR)
with pytest.raises(RuntimeError) as e:
model.load()
assert 'More than one model file is detected' in str(e.value)
def test_dir_with_incompatible_model():
model = SKLearnModel("model", _MODEL_DIR + "/pkl")
with pytest.raises(ModuleNotFoundError) as e:
model.load()
assert 'No module named' in str(e.value)
def test_dir_with_no_model():
model = SKLearnModel("model", _MODEL_DIR)
with pytest.raises(ModelMissingError):
model.load()
def test_dir_with_no_model():
model = SKLearnModel("model", _MODEL_DIR)
with pytest.raises(RuntimeError) as e:
model.load()
assert 'Missing Model File' in str(e.value)