PAI-Designer Overview: Visual No-Code ML Workflow Design - Platform For AI

Service architecture

Create and manage workflows: Create workflows from templates or build from scratch. After running, deploy the model. For more information, see Create a workflow.
Rich library of components and data sources: Designer provides hundreds of AI development components and supports data sources, such as MaxCompute and Object Storage Service (OSS), letting you build models with algorithms that incorporate Alibaba's best practices.
Visualize and analyze the training process: During training, use the visualization dashboard to analyze data, models, and evaluation metrics to find the optimal model.
Deploy and manage models: Register models from Designer workflows to model management, then deploy as online services or package as composite models. For more information, see Online prediction.
Collaborate and share: Designer supports workflow collaboration and sharing within a workspace. Deploy successful workflows to DataWorks for periodic scheduling or publish as custom templates.
Accounts and permissions: Log in to Designer with an Alibaba Cloud account or RAM user. RAM users require necessary permissions from their parent Alibaba Cloud account. For more information, see Cloud product dependencies and permissions: Designer.

Designer provides hundreds of components for various use cases. For more information about the components, see Designer component reference.

Components are categorized into three types by use case:

Traditional machine learning components: Algorithm Components for data pre-processing, feature engineering, statistical analysis, anomaly detection, recommendation algorithms, time series, and network analysis.
Deep learning framework components: Vision algorithms and Natural Language Processing algorithms based on the PAI-Easy series, as well as Deep Learning frameworks like TensorFlow and PyTorch.
Custom algorithm components: Custom Algorithm Components such as SQL script, Python script, Notebook Script, and PyAlink script for your custom requirements.

Create a workflow

A workflow is a visual canvas containing all component nodes and data flows. It is the starting point for model development.
Build and debug a model

Use pre-built algorithm components to build a model in a drag-and-drop interface. Select computing resources, such as MaxCompute, PAI-DLC, or Flink, to run the Workflow and complete model debugging and training.
Visualize and analyze

After training, use the visualization dashboard to view analysis reports and evaluate if the model meets expectations.
Deploy and predict models
- Online prediction: After training, deploy models as online services with PAI-EAS to make predictions on new data.
- Offline batch prediction: For scheduled, offline batch predictions, submit Designer workflows to DataWorks for periodic scheduling.

Designer billing is based on resources components consume at runtime. Details are as follows:

Billable item

Billing entity

Billing method

Stop billing

Billing rules

Compute Unit-hour (CU-hour) usage

Component runtime

Pay-as-you-go

Stop the component

Resources consumed by a component are converted into Compute Unit (CU) hours.

Billing formula: Billable amount = Number of CU-hours × Unit price

Where, Number of CU-hours = max(Number of CPU cores × Duration (h), Memory (GB) × Duration (h) / 4)

For unit prices of different components and more billing details, see Billing for Designer.

To get started, see the Quick start for Designer.

Artificial Intelligence Recommendation: Implement recommendation recall with FM-Embedding | FM recommendation based on the Alink framework.
Intelligent risk control: Financial risk control using graph algorithms | User churn risk warning.
General use cases: News Classification with Text Analysis | Smog Prediction | Power Plant Output Prediction.

If you encounter issues, such as workflow errors, problems with algorithm components, or data read failures, see the Designer FAQ.

PAIFlow is the workflow scheduling engine for Designer. Submit workflow tasks from Designer to PAIFlow for execution.
The PAIFlow task management page contains all Pipeline tasks submitted through manual execution via Designer and periodic scheduling of Designer workflows in DataWorks. For more information, see Manage workflow tasks.