def main():
# get url parameters for HTML template
name_arg = request.args.get('name', 'mnist')
addr_arg = request.args.get('addr', 'mnist-service')
port_arg = request.args.get('port', '9000')
args = {"name": name_arg, "addr": addr_arg, "port": port_arg}
logging.info("Request args: %s", args)
output = None
connection = {"text": "", "success": False}
img_id = str(uuid.uuid4())
img_path = "static/tmp/" + img_id + ".png"
try:
# get a random test MNIST image
x, y, _ = random_mnist(img_path)
# get prediction from TensorFlow server
pred, scores, ver = get_prediction(x,
server_host=addr_arg,
server_port=int(port_arg),
server_name=name_arg,
timeout=10)
# if no exceptions thrown, server connection was a success
connection["text"] = "Connected (model version: " + str(ver) + ")"
connection["success"] = True
# parse class confidence scores from server prediction
scores_dict = []
for i in range(0, 10):
scores_dict += [{"index": str(i), "val": scores[i]}]
output = {
"truth": y,
"prediction": pred,
"img_path": img_path,
"scores": scores_dict
}
except Exception as e: # pylint: disable=broad-except
logging.info("Exception occured: %s", e)
# server connection failed
connection["text"] = "Exception making request: {0}".format(e)
# after 10 seconds, delete cached image file from server
t = Timer(10.0, remove_resource, [img_path])
t.start()
# render results using HTML template
return render_template('index.html',
output=output,
connection=connection,
args=args)
def main():
# get url parameters for HTML template
model_arg = request.args.get('model', 'kfserving-mnist-01')
server_name_arg = request.args.get('name', 'kfserving-mnist-01.kubeflow.example.com')
server_ip_arg = request.args.get('addr', '10.108.37.106')
args = {"model": model_arg, "name": server_name_arg, "addr": server_ip_arg}
logging.info("Request args: %s", args)
output = None
connection = {"text": "", "success": False}
img_id = str(uuid.uuid4())
img_path = "static/tmp/" + img_id + ".png"
logging.info("img_path " + img_path)
try:
# get a random test MNIST image
x, y, _ = random_mnist(img_path)
# get prediction from TensorFlow server
pred = get_prediction(x,
model_name=model_arg,
server_ip=server_ip_arg,
server_name=server_name_arg)
logging.info("pred " + pred)
# if no exceptions thrown, server connection was a success
connection["text"] = "Connected (model version: )"
connection["success"] = True
# parse class confidence scores from server prediction
scores_dict = []
for i in range(0, 10):
scores_dict += [{"index": str(i)}]
output = {"truth": y, "prediction": pred,
"img_path": img_path, "scores": scores_dict}
except Exception as e: # pylint: disable=broad-except
logging.info("Exception occured: %s", e)
# server connection failed
connection["text"] = "Exception making request: {0}".format(e)
# after 10 seconds, delete cached image file from server
t = Timer(10.0, remove_resource, [img_path])
t.start()
# render results using HTML template
return render_template('index.html', output=output,
connection=connection, args=args)
def main():
# get url parameters for HTML template
name_arg = request.args.get('name', 'mnist-classifier')
addr_arg = request.args.get('addr', 'localhost')
port_arg = request.args.get('port', '8080')
args = {"name": name_arg, "addr": addr_arg, "port": port_arg}
print(args)
output = None
connection = {"text": "", "success": False}
img_id = str(uuid.uuid4())
img_path = "static/tmp/" + img_id + ".png"
try:
# get a random test MNIST image
x, y, _ = random_mnist(img_path)
# get prediction from TensorFlow server
pred = get_prediction(x,
server_host=addr_arg,
server_port=int(port_arg),
deployment_name=name_arg,
timeout=10)
# if no exceptions thrown, server connection was a success
connection["text"] = "Connected to Seldon GRPC model serving service"
connection["success"] = True
# parse class confidence scores from server prediction
scores_dict = []
for i in range(0, 10):
scores_dict += [{"index": str(i), "val": pred[i]}]
output = {"truth": y, "img_path": img_path, "scores": scores_dict}
except Exception as e:
# server connection failed
connection["text"] = "Could Not Connect to Server: " + str(e)
# after 10 seconds, delete cached image file from server
t = Timer(10.0, remove_resource, [img_path])
t.start()
# render results using HTML template
return render_template('index.html',
output=output,
connection=connection,
args=args)
def main():
# get url parameters for HTML template
name_arg = request.args.get('name', 'mnist')
addr_arg = request.args.get('addr', 'tf-serving.kubeflow.svc')
port_arg = request.args.get('port', '9000')
args = {"name": name_arg, "addr": addr_arg, "port": port_arg}
print(args)
output = None
connection = {"text": "", "success": False}
img_id = str(uuid.uuid4())
img_path = "static/tmp/" + img_id + ".png"
try:
# get a random test MNIST image
x, y, _ = random_mnist(img_path)
# get prediction from TensorFlow server
pred, scores, ver = get_prediction(x,
server_host=addr_arg,
server_port=int(port_arg),
server_name=name_arg,
timeout=10)
# if no exceptions thrown, server connection was a success
connection["text"] = "Connected (model version: " + str(ver) + ")"
connection["success"] = True
# parse class confidence scores from server prediction
scores_dict = []
for i in range(0, 10):
scores_dict += [{"index": str(i), "val": scores[i]}]
output = {
"truth": y,
"prediction": pred,
"img_path": img_path,
"scores": scores_dict
}
except Exception, e:
# server connection failed
connection["text"] = "Could Not Connect to Server: " + str(e)
def test_predict(self): # pylint: disable=no-self-use
this_dir = os.path.dirname(__file__)
data_dir = os.path.join(this_dir, "..", "data")
img_path = os.path.abspath(data_dir)
x, _, _ = mnist_client.random_mnist(img_path)
server_host = "localhost"
server_port = 9000
model_name = "mnist"
# get prediction from TensorFlow server
pred, scores, _ = mnist_client.get_prediction(x,
server_host=server_host,
server_port=server_port,
server_name=model_name,
timeout=10)
if pred < 0 or pred >= 10:
raise ValueError(
"Prediction {0} is not in the range [0, 9]".format(pred))
if len(scores) != 10:
raise ValueError(
"Scores should have dimension 10. Got {0}".format(scores))