Unit 3. Training, Registering and Deploying Model¶

After designing and orchestrating the main pipeline, you can design and orchestrate the pipeline, train, register, and deploy the model in the sub-canvas of the ParallelFor operator to predict the wind power of each site.

Designing Pipelines¶

The processing logic for training, registering and deploying the models is given as follows:

Model training: Pass the site ID (the value of item, such as abcde0001) to the Python script for model training, and the model will be generated after the training is completed.
Model creation: Use the item name to automatically create a model, and output the model name; if the model already exists, you don’t need to create it, and the model name will be outputted directly.
Model version staging: Stage the model version corresponding to the mode generated in Step 2.
Model testing: Before the model version is officially deployed, test the staged model version. The model version can be officially deployed only after it is qualified in the test.
Create a model deployment instance.
Model deployment: Deploy the model version that has qualified in the test online.

Double-click the ParallelFor operator and drag it to the sub-canvas of the ParallelFor operator, and the pipeline after orchestration is shown in the figure below:

The configuration instructions for each operator orchestrated in the pipeline are given as follows:

Git Directory Operator¶

Name: Git directory for transform2

Description: pull the Python script for model training from the Git directory

Input parameters

Parameter Name	Data type	Operation Type	Value
data_source_name	String	Declaration	Name of the registered Git data source
branch	String	Declaration	master
project	String	Declaration	workspace1
paths	List	Declaration	[“workspace1/kmmlds”]

Output parameters

Parameter Name	Value
workspace	directory
paths	list

An sample of operator configuration is given as follows:

Python Operator¶

Name: Transform2

Description: format the input file and take it as the input of Notebook operator.

Input parameters

Parameter Name	Data type	Operation Type	Value
workspace	Directory	Reference	Git directory for transform2.workspace
entrypoint	String	Declaration	workspace1/kmmlds/transform2.py
requirements_file_path	String	Declaration
string_data	variable	Reference	item

Output parameters

Parameter Name	Value
output_list	list

An sample of operator configuration is given as follows:

Notebook Operator¶

Name: Model Traning

Description: train the model

Input parameters

Parameter Name	Data type	Operation Type	Value
workspace	Directory	Reference	Git directory for transform2.workspace
entrypoint	String	Declaration	workspace1/kmmlds/train2.ipynb
requirements_file_path	String	Declaration	workspace1/kmmlds/requirements.txt
env	List	Reference	Transform2.output_list

Output parameters

Parameter Name	Value
mlflow_model_file_paths	list

An sample of operator configuration is given as follows:

Model Operator¶

Name: Model

Description: register the model

Input parameters

Parameter Name	Data type	Operation Type	Value
category	String	Declaration	Predictor
model_name	String	Reference	item
input_data_type	String	Declaration	Text
scope	String	Declaration	Private
technique	String	Declaration	Regression
usecase	String	Declaration	Wind
publisher	String	Declaration	User_name (enter the username)
input_format	String	Declaration	Input parameters of the model feature in JSON format. See the sample.
output_format	String	Declaration	Model target output in JSON format, See the sample.
interface	String	Declaration	REST
error_on_exist	String	Declaration	false

Output parameters

Parameter Name	Data type
model_name_output	string

An sample of operator configuration is given as follows:

input_format sample

[{
    "name": "X-basic.hour",
    "dtype": "int",
    "ftype": "continuous",
    "range": [0, 23],
    "annotations": "",
    "repeat": null,
    "defaultValue": 10
}, {
    "name": "X-basic.horizon",
    "dtype": "int",
    "ftype": "continuous",
    "range": [0, 49],
    "annotations": "",
    "repeat": null,
    "defaultValue": 8
}, {
    "name": "i-set",
    "dtype": "int",
    "ftype": "continuous",
    "range": [0, 440],
    "annotations": "",
    "repeat": null,
    "defaultValue": 300
}, {
    "name": "EC-ws",
    "dtype": "float",
    "ftype": "continuous",
    "range": [1, 2],
    "annotations": "",
    "repeat": null,
    "defaultValue": "1.5"
}, {
    "name": "EC-wd",
    "dtype": "float",
    "ftype": "continuous",
    "range": [240, 300],
    "annotations": "",
    "repeat": null,
    "defaultValue": 250
}, {
    "name": "EC-tmp",
    "dtype": "float",
    "ftype": "continuous",
    "range": [18, 30],
    "annotations": "",
    "repeat": null,
    "defaultValue": 20
}, {
    "name": "EC-pres",
    "dtype": "float",
    "ftype": "continuous",
    "range": [820, 900],
    "annotations": "",
    "repeat": null,
    "defaultValue": 850
}, {
    "name": "EC-rho",
    "dtype": "float",
    "ftype": "continuous",
    "range": [1, 2],
    "annotations": "",
    "repeat": null,
    "defaultValue": 1
}, {
    "name": "EC-dist",
    "dtype": "float",
    "ftype": "continuous",
    "range": [12, 100],
    "annotations": "",
    "repeat": null,
    "defaultValue": 14
}, {
    "name": "GFS-ws",
    "dtype": "float",
    "ftype": "continuous",
    "range": [1, 2],
    "annotations": "",
    "repeat": null,
    "defaultValue": 1
}, {
    "name": "GFS-wd",
    "dtype": "float",
    "ftype": "continuous",
    "range": [40, 300],
    "annotations": "",
    "repeat": null,
    "defaultValue": 50
}, {
    "name": "GFS-tmp",
    "dtype": "float",
    "ftype": "continuous",
    "range": [18, 20],
    "annotations": "",
    "repeat": null,
    "defaultValue": 19
}, {
    "name": "GFS-pres",
    "dtype": "float",
    "ftype": "continuous",
    "range": [840, 900],
    "annotations": "",
    "repeat": null,
    "defaultValue": 850
}, {
    "name": "GFS-rho",
    "dtype": "float",
    "ftype": "continuous",
    "range": [1, 2],
    "annotations": "",
    "repeat": null,
    "defaultValue": 1
}, {
    "name": "GFS-dist",
    "dtype": "int",
    "ftype": "continuous",
    "range": [12, 100],
    "annotations": "",
    "repeat": null,
    "defaultValue": 20
}, {
    "name": "sequence",
    "dtype": "int",
    "ftype": "continuous",
    "range": [1, 26901],
    "annotations": "",
    "repeat": null,
    "defaultValue": 20
}]

output_format sample

[{
    "name": "power",
    "dtype": "float",
    "ftype": "continuous",
    "range": [],
    "annotations": "",
    "repeat": null,
    "defaultValue": 0
}]

Mlflow Model Version Register Operator¶

Name: Model Version Register

Description: stage the model version

Input parameters

Parameter Name	Data type	Operation Type	Value
input_data	String	Declaration	Model version parameter input. See the sample.
version_rule	String	Declaration	time
annotation	String	Declaration	test
architecture	String	Declaration	x86
coprocessor	String	Declaration	None
env_param	List	Declaration	[]
framework	String	Declaration	sklearn
language	String	Declaration	python3
model_reference	String	Reference	Model.model_name_output
publisher	String	Declaration	User_name (name of the model version creator)
minio_paths	List	Reference	Model Traning.mlflow_model_file_paths

Output parameters

Parameter Name	Parameter type
create_model_revision	String
model_revision_name	String
model_builder_name	String

An sample of operator configuration is given as follows:

Input_data sample

{
    "data": {
        "names": ["sequence", "X-basic.hour", "X-basic.horizon", "i-set", "EC-ws", "EC-wd", "EC-tmp", "EC-pres", "EC-rho", "EC-dist", "GFS-ws", "GFS-wd", "GFS-tmp", "GFS-pres", "GFS-rho", "GFS-dist"],
        "ndarray": [
            [20000, 11, 37, 1, 2, 257, 18, 85, 0, 15, 1, 6, 20, 879, 1, 59],
            [200500, 1, 3, 1, 2, 57, 18, 85, 0, 15, 1, 1, 20, 879, 1, 59]
        ]
    }
}

Model Test Operator¶

Name: Model Test

Description: test the model version

Input parameters

Parameter Name	Data type	Operation Type	Value
input_data	String	Declaration	Enter the model testing data in JSON format. See the sample.
model_builder	String	Reference	Model Version Register.model_builder_name

Output parameters

Parameter Name	Parameter type
create_model_test	String
model_test_output	String

An sample of operator configuration is given as follows:

Input_data sample

{
    "data": {
        "names": ["sequence", "X-basic.hour", "X-basic.horizon", "i-set", "EC-ws", "EC-wd", "EC-tmp", "EC-pres", "EC-rho", "EC-dist", "GFS-ws", "GFS-wd", "GFS-tmp", "GFS-pres", "GFS-rho", "GFS-dist"],
        "ndarray": [
            [20000, 11, 37, 1, 2, 257, 18, 85, 0, 15, 1, 6, 20, 879, 1, 59],
            [200500, 1, 3, 1, 2, 57, 18, 85, 0, 15, 1, 1, 20, 879, 1, 59]
        ]
    }
}

Single Instance Operator¶

Name: Model Instance

Description: model deployment instance

Input parameters

Parameter Name	Data type	Operation Type	Value
name	String	Declaration	Enter the name of the model deployment instance (e.g. abctest)
resource_pool	String	Declaration	Select the deployment model resource pool
model_name	String	Reference	Model.model_name_output
labels	List	Declaration	(Optional) enter the tag of the model deployment instance
description	String	Declaration	(Optional) enter the description of the model deployment instance
deploy_mode	String	Declaration	ONLINE
error_on_exist	String	Declaration	false

Output parameters

Parameter Name	Parameter type
instance_name_output	String

An sample of operator configuration is given as follows:

Single Model Deployment Operator¶

Name: Single Model Deployment

Description: model version deployment

Input parameters

Parameter Name	Data type	Operation Type	Value
model_revision	String	Reference	Model Version Register.model_revision_name
instance_name	String	Declaration	Model Instance.instance_name_output
request_cpu	Number	Declaration	0.5
request_memory	Number	Declaration	0.5
limit_cpu	Number	Declaration	1.0
limit_memory	Number	Declaration	1.0
timeout	Number	Declaration	360

Output parameters

Parameter Name	Parameter type
create_model_deployment	String

An sample of operator configuration is given as follows:

Next Unit¶

Designing Mechanism for Monitoring Data Arrival