def test_dataset_registration(config):
ap = ArangoPipe(config=config)
ds_info = {"name" : "wine dataset",\
"description": "Wine quality ratings","source": "UCI ML Repository" }
ds_reg = ap.register_dataset(ds_info)
return
def setUp(self):
self.config = ArangoPipeConfig()
self.config.set_dbconnection(hostname = "localhost", port = 8529,\
root_user = "******",\
root_user_password = "******")
self.admin = ArangoPipeAdmin(config = self.config)
self.ap = ArangoPipe(config = self.config)
self.provision_project()
def test_model_registration(config):
ap = ArangoPipe(config=config)
model_info = {
"name": "elastic_net_wine_model",
"type": "elastic net regression"
}
model_reg = ap.register_model(model_info)
return
def test_log_servingperf(config):
to_date = datetime.datetime.now()
from_date = to_date - datetime.timedelta(days=30)
ex_servingperf = {"rmse": 0.822242, "r2": 0.12678, "mae": 0.62787,\
"from_date": str(from_date), "to_date": str(to_date)}
dep_tag = "Wine_Elastic_Net_Regression"
user_id = "prvileged user"
ap = ArangoPipe(config=config)
ret = ap.log_serving_perf(ex_servingperf, dep_tag, user_id)
return ret
def test_featureset_registration(config):
ap = ArangoPipe(config=config)
fs_info = {"fixed acidity": "float64",\
"volatile acidity": "float64",\
"citric acid": "float64",\
"residual sugar": "float64",\
"chlorides": "float64",\
"free sulfur dioxide": "float64",\
"total sulfur dioxide": "float64",\
"density": "float64",\
"pH": "float64",\
"sulphates": "float64",\
"alcohol": "float64",\
"quality": "int64",\
"name": "wine_no_transformations"
}
ds_reg = ap.lookup_dataset("wine dataset")
fs_reg = ap.register_featureset(fs_info, ds_reg["_key"])
return
def test_log_run(config):
ap = ArangoPipe(config=config)
ds_reg = ap.lookup_dataset("wine dataset")
fs_reg = ap.lookup_featureset("wine_no_transformations")
model_reg = ap.lookup_model("elastic_net_wine_model")
model_params = { "l1_ratio": 0.1, "alpha": 0.2,\
"run_id": "0ef73d9edf08487793c77a1742f4033e"}
model_perf = { "rmse": 0.7836984021909766, "r2": 0.20673590971167466,\
"mae": 0.6142020452688988, "run_id": "0ef73d9edf08487793c77a1742f4033e",\
"timestamp": "2019-06-06 12:52:11.190048"}
run_info = {"dataset" : ds_reg["_key"],\
"featureset": fs_reg["_key"],\
"run_id": "0ef73d9edf08487793c77a1742f4033e",\
"model": model_reg["_key"],\
"model-params": model_params,\
"model-perf": model_perf,\
"pipeline" : "Wine-Regression-Pipeline",\
"project": "Wine-Quality-Assessment",\
"deployment_tag": "Wine_Elastic_Net_Regression",\
"tag": "wine regression model test 1"}
ap.log_run(run_info)
return
class TestArangopipe(unittest.TestCase):
def setUp(self):
self.config = ArangoPipeConfig()
self.config.set_dbconnection(hostname = "localhost", port = 8529,\
root_user = "******",\
root_user_password = "******")
self.admin = ArangoPipeAdmin(config = self.config)
self.ap = ArangoPipe(config = self.config)
self.provision_project()
def provision_project(self):
err_raised = False
try:
proj_info = {"name": "Wine-Quality-Regression-Modelling"}
proj_reg = self.admin.register_project(proj_info)
except:
err_raised = True
print ('-'*60)
traceback.print_exc(file=sys.stdout)
print ('-'*60)
self.assertTrue(err_raised,\
'Exception raised while provisioning project')
self.assertFalse(err_raised)
return
def register_dataset(self):
ds_info = {"name" : "wine_dataset",\
"description": "Wine quality ratings",\
"source": "UCI ML Repository" }
ds_reg = self.ap.register_dataset(ds_info)
return
def lookup_dataset(self):
ds_reg = self.ap.lookup_dataset("wine_dataset")
return
def lookup_featureset(self):
fs_reg = self.ap.lookup_featureset("wine_no_transformations")
return
def register_model(self):
model_info = {"name": "elastic_net_wine_model", \
"type": "elastic net regression"}
model_reg = self.ap.register_model(model_info)
return
def lookup_model(self):
model_reg = self.ap.lookup_model("elastic_net_wine_model")
return
def log_run(self):
ds_reg = self.ap.lookup_dataset("wine_dataset")
fs_reg = self.ap.lookup_featureset("wine_no_transformations")
model_reg = self.ap.lookup_model("elastic_net_wine_model")
model_params = { "l1_ratio": 0.1, "alpha": 0.2,\
"run_id": "0ef73d9edf08487793c77a1742f4033e"}
model_perf = { "rmse": 0.7836984021909766, "r2": 0.20673590971167466,\
"mae": 0.6142020452688988, "run_id": "0ef73d9edf08487793c77a1742f4033e",\
"timestamp": "2019-06-06 12:52:11.190048"}
run_info = {"dataset" : ds_reg["_key"],\
"featureset": fs_reg["_key"],\
"run_id": "0ef73d9edf08487793c77a1742f4033e",\
"model": model_reg["_key"],\
"model-params": model_params,\
"model-perf": model_perf,\
"pipeline" : "Wine-Regression-Pipeline",\
"project": "Wine-Quality-Assessment",\
"deployment_tag": "Wine_Elastic_Net_Regression",\
"tag": "wine regression model test 1"}
self.ap.log_run(run_info)
return
def provision_deployment(self):
ret = self.admin.register_deployment("Wine_Elastic_Net_Regression")
return
def register_featureset(self):
fs_info = {"fixed acidity": "float64",\
"volatile acidity": "float64",\
"citric acid": "float64",\
"residual sugar": "float64",\
"chlorides": "float64",\
"free sulfur dioxide": "float64",\
"total sulfur dioxide": "float64",\
"density": "float64",\
"pH": "float64",\
"sulphates": "float64",\
"alcohol": "float64",\
"quality": "int64",\
"name": "wine_no_transformations"
}
ds_reg = self.ap.lookup_dataset("wine_dataset")
fs_reg = self.ap.register_featureset(fs_info, ds_reg["_key"])
return
def log_servingperf(self):
to_date = datetime.datetime.now()
from_date = to_date - datetime.timedelta(days = 30)
ex_servingperf = {"rmse": 0.822242, "r2": 0.12678, "mae": 0.62787,\
"from_date": str(from_date), "to_date": str(to_date)}
dep_tag = "Wine_Elastic_Net_Regression"
user_id = "prvileged user"
ret = self.ap.log_serving_perf(ex_servingperf, dep_tag, user_id)
return
def dataset_shift_positive(self):
ds_path = os.path.join(os.path.dirname(os.path.abspath(__file__)), "covariate_shift/cal_housing.csv")
df = pd.read_csv(ds_path)
req_cols = df.columns.tolist()
df = df[req_cols]
df1 = df.query("lat <= -119")
df2 = df.query("lat > -119")
rfd = RF_DatasetShiftDetector()
score = rfd.detect_dataset_shift(df1, df2)
print ("Detaset shift score : ", score)
return score
def dataset_shift_negative(self):
ds_path = os.path.join(os.path.dirname(os.path.abspath(__file__)), "covariate_shift/cal_housing.csv")
df = pd.read_csv(ds_path)
req_cols = df.columns.tolist()
df = df[req_cols]
df1 = df.query("lat <= -119")
df2 = df1.copy()
rfd = RF_DatasetShiftDetector()
score = rfd.detect_dataset_shift(df1, df2)
print ("Detaset shift score : ", score)
return score
def test_register_dataset(self):
err_raised = False
try:
self.register_dataset()
except:
err_raised = True
print ('-'*60)
traceback.print_exc(file=sys.stdout)
print ('-'*60)
self.assertTrue(err_raised,\
'Exception raised while registering dataset')
self.assertFalse(err_raised)
return
def test_lookup_dataset(self):
err_raised = False
try:
self.register_dataset()
self.lookup_dataset()
except:
err_raised = True
print ('-'*60)
traceback.print_exc(file=sys.stdout)
print ('-'*60)
self.assertTrue(err_raised,\
'Exception raised while looking up dataset')
self.assertFalse(err_raised)
return
def test_register_featureset(self):
err_raised = False
try:
self.register_dataset()
self.register_featureset()
except:
err_raised = True
print ('-'*60)
traceback.print_exc(file=sys.stdout)
print ('-'*60)
self.assertTrue(err_raised,\
'Exception raised while registering featureset')
self.assertFalse(err_raised)
return
def test_lookup_featureset(self):
err_raised = False
try:
self.register_dataset()
self.register_featureset()
self.lookup_featureset()
except:
err_raised = True
print ('-'*60)
traceback.print_exc(file=sys.stdout)
print ('-'*60)
self.assertTrue(err_raised,\
'Exception raised while registering featureset')
self.assertFalse(err_raised)
return
def test_register_model(self):
err_raised = False
try:
self.register_model()
except:
err_raised = True
print ('-'*60)
traceback.print_exc(file=sys.stdout)
print ('-'*60)
self.assertTrue(err_raised,\
'Exception raised while registering model')
self.assertFalse(err_raised)
return
def test_lookup_model(self):
err_raised = False
try:
self.register_model()
self.lookup_model()
except:
err_raised = True
print ('-'*60)
traceback.print_exc(file=sys.stdout)
print ('-'*60)
self.assertTrue(err_raised,\
'Exception raised while looking up model')
self.assertFalse(err_raised)
return
def test_log_run(self):
err_raised = False
try:
self.register_dataset()
self.register_featureset()
self.register_model()
self.log_run()
except:
err_raised = True
print ('-'*60)
traceback.print_exc(file=sys.stdout)
print ('-'*60)
self.assertTrue(err_raised,\
'Exception raised while logging performance')
self.assertFalse(err_raised)
return
def test_provision_deployment(self):
err_raised = False
try:
self.register_dataset()
self.register_featureset()
self.register_model()
self.log_run()
self.provision_deployment()
except:
err_raised = True
print ('-'*60)
traceback.print_exc(file=sys.stdout)
print ('-'*60)
self.assertTrue(err_raised,\
'Exception raised while provisioning deployment')
self.assertFalse(err_raised)
return
def test_log_serving_performance(self):
err_raised = False
try:
self.register_dataset()
self.register_featureset()
self.register_model()
self.log_run()
self.provision_deployment()
self.log_servingperf()
except:
err_raised = True
print ('-'*60)
traceback.print_exc(file=sys.stdout)
print ('-'*60)
self.assertTrue(err_raised,\
'Exception raised while logging serving performance')
self.assertFalse(err_raised)
return
def test_dataset_shift_positive(self):
score = self.dataset_shift_positive()
self.assertTrue(score > 0.8)
return
def test_dataset_shift_negative(self):
score = self.dataset_shift_negative()
self.assertTrue(score < 0.6)
return
def tearDown(self):
#pass
self.admin.delete_arangomldb()
self.ap = None
self.admin = None
return
def eval_metrics(actual, pred):
rmse = np.sqrt(mean_squared_error(actual, pred))
mae = mean_absolute_error(actual, pred)
r2 = r2_score(actual, pred)
return rmse, mae, r2
if __name__ == "__main__":
warnings.filterwarnings("ignore")
np.random.seed(40)
conn_config = ArangoPipeConfig()
conn_config.set_dbconnection(hostname = "localhost", port = 8529,\
root_user = "******", root_user_password = "******")
ap = ArangoPipe(config=conn_config)
# Read the wine-quality csv file (make sure you're running this from the root of MLflow!)
wine_path = os.path.join(os.path.dirname(os.path.abspath(__file__)),
"wine-quality.csv")
data = pd.read_csv(wine_path)
ds_reg = ap.lookup_dataset("wine dataset")
fs_reg = ap.lookup_featureset("wine_no_transformations")
# Split the data into training and test sets. (0.75, 0.25) split.
train, test = train_test_split(data)
# The predicted column is "quality" which is a scalar from [3, 9]
train_x = train.drop(["quality"], axis=1)
test_x = test.drop(["quality"], axis=1)
train_y = train[["quality"]]
def test_model_lookup(config):
ap = ArangoPipe(config=config)
model_reg = ap.lookup_model("elastic_net_wine_model")
return
def test_featureset_lookup(config):
ap = ArangoPipe(config=config)
fs_reg = ap.lookup_featureset("wine_no_transformations")
return
def test_dataset_lookup(config):
ap = ArangoPipe(config=config)
ds_reg = ap.lookup_dataset("wine dataset")
return
r2 = r2_score(actual, pred)
return rmse, mae, r2
if __name__ == "__main__":
proj_info = {"name": "Wine-Quality-Regression-Modelling"}
conn_config = ArangoPipeConfig()
conn_config.set_dbconnection(hostname = "localhost", port = 8529,\
root_user = "******", root_user_password = "******")
admin = ArangoPipeAdmin(config=conn_config)
proj_reg = admin.register_project(proj_info)
warnings.filterwarnings("ignore")
np.random.seed(40)
ap = ArangoPipe(conn_config)
# Read the wine-quality csv file (make sure you're running this from the root of MLflow!)
wine_path = os.path.join(os.path.dirname(os.path.abspath(__file__)),
"wine-quality.csv")
data = pd.read_csv(wine_path)
ds_info = {"name" : "wine dataset",\
"description": "Wine quality ratings","source": "UCI ML Repository" }
ds_reg = ap.register_dataset(ds_info)
featureset = data.dtypes.to_dict()
featureset = {k: str(featureset[k]) for k in featureset}
featureset["name"] = "wine_no_transformations"
fs_reg = ap.register_featureset(featureset, ds_reg["_key"])
model_info = {
"name": "elastic_net_wine_model",
"type": "elastic net regression"
def generate_runs(clean=False):
conn_config = ArangoPipeConfig()
conn_config.set_dbconnection(hostname = "localhost", port = 8529,\
root_user = "******", root_user_password = "******")
admin = ArangoPipeAdmin(config=conn_config)
ap = ArangoPipe(config=conn_config)
if clean:
admin.delete_arangomldb()
admin.create_db()
admin.create_enterprise_ml_graph()
proj_info = {"name": "Home_Value_Assessor"}
proj_reg = admin.register_project(proj_info)
period = period_string_generator()
ds_info = {"description": "Housing Price Data"}
featureset = data.dtypes.to_dict()
featureset = {k: str(featureset[k]) for k in featureset}
count = 1
for data_tuple in dataset_generator():
logger.info("Processing Dataset:" + str(count))
count = count + 1
aperiod = next(period)
X_train = data_tuple[0]
X_test = data_tuple[1]
y_train = data_tuple[2]
y_test = data_tuple[3]
X_val = data_tuple[4]
y_val = data_tuple[5]
alpha_random = np.random.uniform(0.0005, 0.001)
lrm = linear_model.Lasso(alpha=alpha_random)
lrm.fit(X_train, y_train)
predicted_val = lrm.predict(X_val)
(rmse, mae, r2) = eval_metrics(y_val, predicted_val)
ruuid = uuid.uuid4()
model_perf = {"rmse": rmse, "r2": r2, "mae": mae, "run_id": str(ruuid), \
"timestamp": str(dt.datetime.now())}
serving_pred = lrm.predict(X_test)
(rmse, mae, r2) = eval_metrics(y_test, serving_pred)
ex_servingperf = {"rmse": rmse, "r2": r2, "mae": mae,\
"period" : aperiod}
deployment_tag = "Deployment_HPE_" + aperiod
dataset_tag = "Housing_Dataset_" + aperiod
pipeline_tag = "Pipeline_HPE" + aperiod
feature_pipeline_tag = "Feature Pipeline HPE" + aperiod
ds_info["name"] = dataset_tag
ds_info["tag"] = dataset_tag
ds_info["source"] = "Housing Price Operational Data Store"
featureset["generated_by"] = feature_pipeline_tag
featureset["name"] = "log_transformed_house_value"
featureset["tag"] = dataset_tag
ds_reg = ap.register_dataset(ds_info)
fs_reg = ap.register_featureset(featureset, ds_reg["_key"])
model_tag = "model_period:" + aperiod
model_info = {"name": "Housing Regression Model",\
"type": "LASSO regression", "tag": model_tag}
model_reg = ap.register_model(model_info,
project="Home_Value_Assessor")
model_params = {"alpha": alpha_random, "run_id": str(ruuid)}
run_info = {"dataset" : ds_reg["_key"],\
"featureset": fs_reg["_key"],\
"run_id": str(ruuid),\
"model": model_reg["_key"],\
"model-params": model_params,\
"model-perf": model_perf,\
"pipeline" : pipeline_tag,\
"project": "Housing Price Assessor",
"tag_for_deployment": True,\
"deployment_tag": deployment_tag}
ap.log_run(run_info)
admin.register_deployment(deployment_tag)
user_id = "Arangopipe Test Data Generator"
ap.log_serving_perf(ex_servingperf, deployment_tag, user_id)
return
def run_driver():
params = {'batch_size': 128, 'shuffle': True, 'num_workers': 6}
trng_dataset = CH_Dataset()
test_dataset = CH_Dataset(train=False)
training_generator = data.DataLoader(trng_dataset, **params)
test_generator = data.DataLoader(test_dataset, **params)
input_size = trng_dataset.input_size
output_size = trng_dataset.output_size
m = CH_LinearRegression(inputSize=input_size, outputSize=output_size)
cost_func = nn.MSELoss()
learning_rate = 0.1
optimizer = torch.optim.Adam(m.parameters(), lr=learning_rate)
all_losses = []
test_pred_list = []
test_acts_list = []
num_epochs = 100
loss_sched = {}
for e in range(num_epochs):
batch_losses = []
for ix, (Xb, yb) in enumerate(training_generator):
_X = Variable(Xb).float()
_y = Variable(yb).float()
#==========Forward pass===============
preds = m(_X)
preds = torch.flatten(preds)
loss = cost_func(preds, _y)
#==========backward pass==============
optimizer.zero_grad()
loss.backward()
optimizer.step()
batch_losses.append(loss.item())
all_losses.append(loss.item())
mbl = sqrt(np.mean(batch_losses))
if e % 5 == 0:
print("training loss: " + str(mbl))
loss_sched[e] = mbl
# prepares model for inference when trained with a dropout layer
# print(m.training)
# m.eval()
# print(m.training)
test_batch_losses = []
test_pred_list = []
test_acts_list = []
for _X, _y in test_generator:
_X = Variable(_X).float()
_y = Variable(_y).float()
#apply model
test_preds = m(_X)
test_preds = torch.flatten(test_preds)
test_loss = cost_func(test_preds, _y)
test_pred_list.extend(test_preds.detach().numpy().ravel())
test_acts_list.extend(_y.numpy().ravel())
test_batch_losses.append(test_loss.item())
# print("Batch loss: {}".format(test_loss.item()))
tmbl = sqrt(np.mean(test_batch_losses))
print("test loss: " + str(tmbl))
# Store experiment results in Arangopipe
conn_config = ArangoPipeConfig()
conn_config.set_dbconnection(hostname = "localhost", port = 8529,\
root_user = "******", root_user_password = "******")
proj_info = {"name": "Housing_Price_Estimation_Project"}
admin = ArangoPipeAdmin(config=conn_config)
proj_reg = admin.register_project(proj_info)
ap = ArangoPipe(conn_config)
ruuid = str(uuid.uuid4().int)
model_name = "pytorch-linear-reg" + "_dev_run_" + ruuid
model_info = {"name": model_name, "type": "model-development"}
model_reg = ap.register_model(model_info,
project="Housing_Price_Estimation_Project")
ds_info = trng_dataset.get_dataset()
ds_reg = ap.register_dataset(ds_info)
fs = trng_dataset.get_featureset()
fs_reg = ap.register_featureset(fs, ds_reg["_key"])
model_params = {"optimizer": "Adam", "training_epochs": 100,\
"batch_size": 128, "learning_rate": learning_rate,\
"run_id": ruuid}
model_perf = {"training_loss_schedule": jsonpickle.encode(loss_sched),\
"run_id": ruuid, "timestamp": str(datetime.datetime.now())}
run_tag = "Housing-Price-Pytorch-Experiment" + "_dev_run_" + ruuid
run_info = {"dataset" : ds_reg["_key"],\
"featureset": fs_reg["_key"],\
"run_id": ruuid,\
"model": model_reg["_key"],\
"model-params": model_params,\
"model-perf": model_perf,\
"tag": run_tag,\
"project": "Housing Price Estimation Project"}
ap.log_run(run_info)
mp = ap.lookup_modelperf(run_tag)
print(
"A look up of the loss schedule for this experiment in Arangopipe yields:"
)
print(str(mp["training_loss_schedule"]))
return