def test_experiment_yaml_serde(Skil):
file_name = "./mock.yml"
serialize_config(MOCK_CONFIG, file_name, 'yaml')
config = deserialize_config(file_name)
assert config == MOCK_CONFIG
exp = Experiment.load(file_name, Skil())
assert exp.id == 'foo'
os.remove(file_name)
def test_e2e(Skil):
model_path = './dummy.pb'
skil_server = Skil()
skil_server.upload_model(model_path)
ws = WorkSpace(skil_server, 'jupyter_ws')
experiment = Experiment(ws, 'test_exp')
model = Model(model_path, experiment)
model.add_evaluation(accuracy=0.93)
deployment = Deployment(skil_server, 'test_deployment')
model.deploy(deployment, start_server=False)
def test_experiment(Skil):
skil_server = Skil()
ws = WorkSpace(skil_server)
experiment = Experiment(ws)
from skil import Skil, WorkSpace, Experiment, Model, Deployment
from keras.datasets import mnist
skil_server = Skil()
work_space = WorkSpace(skil_server)
experiment = Experiment(work_space)
model = Model('model.pb', model_id='tf_model', experiment=experiment)
deployment = Deployment(skil_server)
service = model.deploy(deployment,
input_names=['input'],
output_names=['output'])
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_test = x_test.reshape(10000, 784)
x_test = x_test.astype('float32')
x_test /= 255
print(service.predict_single(x_test[0]))
print(service.predict(x_test[:10]))
from skil import Skil, WorkSpace, Experiment, Model, Deployment
import json
# Initialize a new experiment in a workspace
skil_server = Skil()
ws = WorkSpace(skil_server)
experiment = Experiment(ws)
deployment = Deployment(skil_server, "keras_models")
skil_models = []
for epoch in range(2):
model_name = 'model_{epoch:02d}.hdf5'.format(epoch=epoch + 1)
model = Model(model_name, id=epoch, experiment=experiment)
model.deploy(start_server=False, deployment=deployment)
skil_models.append(model)
with open('history.json', 'r') as f:
history = json.loads(f.read())
acc = history.get('val_acc')
best_model = skil_models[acc.index(max(acc))]
best_model.serve()
from skil import Skil, WorkSpace, Experiment, Model, Deployment
import json
# Initialize a new experiment in a workspace
skil_server = Skil()
work_space = WorkSpace(skil_server, name='mnist-workspace')
experiment = Experiment(work_space, name='mnist-experiment')
deployment = Deployment(skil_server, name='mnist-deployment')
# Deploy all models with SKIL
skil_services = []
for epoch in range(2):
model_name = 'model_{epoch:02d}.hdf5'.format(epoch=epoch + 1)
model = Model(model_name, model_id=epoch, experiment=experiment)
service = model.deploy(start_server=False, deployment=deployment)
skil_services.append(service)
# Serve the best one
with open('history.json', 'r') as f:
history = json.loads(f.read())
acc = history.get('val_acc')
best_service = skil_services[acc.index(max(acc))]
best_service.start()
#import skil_client
#
#configuration = skil_client.Configuration()
#configuration.host = 'http://52.15.103.124:9008'
#configuration.api_key['authorization'] = 'eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.eyJhdWQiOiJTa2lsVXNlciIsInN1YiI6IntcInVzZXJJZFwiOlwiYWRtaW5cIixcInVzZXJOYW1lXCI6XCJhZG1pblwiLFwicm9sZVwiOlwiYWRtaW5cIixcInNjb3BlXCI6XCJhZG1pblwifSIsImlzcyI6IlNraWxBdXRoTWFuYWdlciIsImV4cCI6MTU4NjgyODcxMCwiaWF0IjoxNTU1MjkyNzEwfQ.vXXpr0kk9H8nMbkcM8sPWPJwuws1GGvI8SWc76qKvUo'
#api_instance = skil_client.DefaultApi(skil_client.ApiClient(configuration))
#api_response = api_instance.predictimage("yolo", "default", "yolomodel", image='input.jpg')
#
from skil import Skil
skil_server = Skil(host='52.15.103.124', password='******')
from skil import WorkSpace, Experiment, Model
work_space = WorkSpace(skil_server, name="Object detection project")
experiment = Experiment(work_space, name="YOLO")
model = Model('yolo_v2.pb', name='yolo_model', experiment=experiment)
#
from skil import Deployment
deployment = Deployment(skil_server)
service = model.deploy(deployment,
scale=1,
input_names=['input'],
output_names=['output'])
#
#from skil.utils.yolo import annotate_image
#import cv2
#
#image = cv2.imread('input.jpg')
#detection = service.detect_objects(image)
skil_server = Skil(host=skil_conf['host'],
port=skil_conf['port'],
user_id=skil_conf['username'],
password=skil_conf['password'])
work_space = None
if skil_conf['workspace_id'] == '':
work_space = WorkSpace(skil_server, name="Mnist Sample")
skil_conf['workspace_id'] = work_space.id
else:
work_space = skil.get_workspace_by_id(None, skil_server,
skil_conf['workspace_id'])
experiment = None
if skil_conf['experiment_id'] == '':
experiment = Experiment(work_space, name='Simple MLP')
skil_conf['experiment_id'] = experiment.id
else:
experiment = skil.get_experiment_by_id(work_space,
skil_conf['experiment_id'])
with open('.skil', 'w') as f:
json.dump(skil_conf, f)
# Train the model.
batch_size = 128
num_classes = 10
epochs = 10
(x_train, y_train), (x_test, y_test) = mnist.load_data()
model.compile(loss='categorical_crossentropy',
optimizer='sgd',
metrics=['accuracy'])
history = model.fit(x_train,
y_train,
batch_size=128,
epochs=10,
verbose=1,
validation_data=(x_test, y_test))
_, acc = model.evaluate(x_test, y_test)
# Load the experiment from file (alternatively create one with Experiment())
# Make sure to run: "pyskil init-experiment -f exp.json" before
experiment = Experiment.load('exp.json')
# Save the model to SKIL
skil_model = SkilModel(model, experiment=experiment)
# Add accuracy as evaluation metric to SKIL
skil_model.add_evaluation(accuracy=acc, name="MNIST test set")
# Load the deployment from file (or create one with Deployment()) and deploy model as service
# Make sure to run: "pyskil init-deployment -f dep.json" before
deployment = Deployment.load('dep.json')
service = skil_model.deploy(deployment=deployment)
# Evaluate from scratch, using SKIL
num_correct = 0
for i in range(10000):