Designer is a visual, low-code modeling tool from PAI. It allows you to build a workflow by dragging and dropping algorithm components, and supports online model deployment and scheduled offline tasks.
Product architecture
Features
-
Create and manage workflows: You can create a workflow from a template or from scratch. Workflows created from a template can be quickly run, and the resulting model can be deployed. For more information, see Create a workflow.
-
Rich components and data sources: Designer provides hundreds of AI development components and supports multiple data sources, such as MaxCompute and Object Storage Service (OSS). You can build models by using algorithms that incorporate Alibaba's best practices.
-
Visualize and analyze the training process: During model training, Designer provides a visualization dashboard to help you analyze data, models, and evaluation metrics to find the optimal model.
-
Deploy and manage models: After you develop and test a model in a Designer workflow, you can register it with the model management service. Then, you can deploy it as an online service with a single click or package it as a composite model. For more information, see Model prediction and deployment.
-
Collaborate and share: Designer supports workflow collaboration and sharing within a workspace. You can also deploy a successfully run workflow to DataWorks for periodic scheduling or publish it as a custom template.
-
Accounts and permissions: Designer supports signing in with an Alibaba Cloud account or a RAM user. If you use a RAM user, the Alibaba Cloud account must grant the required permissions to the RAM user. For more information, see Cloud product dependencies and permissions for Designer.
Workflow components
Designer provides hundreds of components for a variety of use cases. For more information about the components, see Designer component overview.
Components are classified into the following three types based on their use cases:
-
Traditional machine learning components: Includes algorithm components for data preprocessing, feature engineering, statistical analysis, anomaly detection, recommendation algorithms, time series, and network analysis.
-
Deep learning framework components: Includes components based on the PAI-Easy series for vision algorithms and natural language processing algorithms, and on deep learning frameworks such as TensorFlow and PyTorch.
-
Custom algorithm components: Includes custom algorithm components such as SQL Script, Python Script, Notebook Script, and PyAlink Script to meet your specific requirements.
How it works
-
A workflow is a visual canvas that contains all component nodes and data flows. It is the starting point for model development.
-
Use the drag-and-drop interface to build a model with pre-built algorithm components. Select the required computing resources, such as MaxCompute, PAI-DLC, or Flink, to run the workflow and complete model debugging and training.
-
After you build the model, you can use the visualization dashboard to quickly view analysis reports and evaluate whether the model meets your expectations.
-
Model deployment and prediction
-
Online prediction: After you build the model, you can use PAI-EAS to deploy the model as an online service to make predictions on new data.
-
Scheduled offline tasks: You can also submit a workflow to DataWorks for periodic scheduling to perform offline batch predictions.
-
Billing
Designer billing is based on the resources that components consume at runtime. The following table describes the billing details.
|
Billable item |
Billing entity |
Billing method |
Billing end |
Billing rules |
|
CU-hour usage |
Component runtime |
pay-as-you-go |
When the component stops running |
The resources consumed by a running component are converted into CU-hours and billed on a pay-as-you-go basis. Billing formula: where |
For information about the unit prices of different components and more billing details, see Billing of Designer.
Quick start
If you are new to Designer, we recommend that you read Quick start for Designer to get started.
Use cases
Intelligent recommendation: Use FM-Embedding for recommendation and recall | FM-based recommendation by using the Alink framework.
Intelligent risk control: Implement financial risk control by using graph algorithms | User churn prediction and warning.
General cases: Text analysis for news classification | Smog prediction | Power plant output prediction.
Get help
If you encounter issues such as workflow errors, problems with algorithm components, or data read failures, see Designer FAQ.
PAIFlow
-
PAIFlow is the underlying workflow scheduling engine for Designer. You can submit workflow tasks from Designer to PAIFlow for execution.
-
The PAIFlow task management page lists all workflow tasks that are submitted by using the manual execution in Designer, or periodic scheduling of Designer workflows in DataWorks. For more information, see Manage workflow tasks.