Model and data FAQ

更新时间:
复制 MD 格式

Use this page to find answers to common questions about model deployment, pipeline configuration, data upload, and incremental training in PAI Machine Learning Designer and Studio.

IssueJump to
No models available when deploying to EASNo models available for EAS deployment
How to auto-update a model service on each pipeline runAutomated model service update per pipeline run
How to upload on-premises data for trainingUpload on-premises data for training
Text displays as blob in Designer or StudioText displayed as blob in Machine Learning Designer or Studio
Whether Designer supports incremental trainingIncremental training support in Machine Learning Designer

No models available for EAS deployment

Symptom: When you choose Model Deployment > EAS (Online Model Service), a message indicates that no models are available for deployment.

Cause: The pipeline is not configured to output a deployable model.

Solution:

  1. Open the pipeline details page in Machine Learning Designer and add the component you want to deploy.

  2. On the Fields Setting tab of the component, select Whether To Generate PMML.

  3. On the Pipeline Attributes tab of the pipeline, configure the Data Storage parameter.

  4. Rerun the component.

Automated model service update per pipeline run

Add the Update EAS Service component to your pipeline. On the Parameters Settings tab, enter the name of the Elastic Algorithm Service (EAS) service to update. Each time the component runs, it retrieves the Object Storage Service (OSS) model path from the upstream component and updates the EAS service automatically.

Upload on-premises data for training

Choose a method based on your file size:

MethodFile sizeSteps
Method 1: Read CSV File component (direct upload)CSV file ≤ 1 GBUse the Read CSV File component to upload the file directly.
Method 2: Upload to OSS, then readCSV file > 1 GBUpload the CSV file to an OSS bucket first, then use the Read CSV File component to read it.
Method 3: DataWorks table uploadAny size, table-basedLog on to the DataWorks console, find your workspace, and click DataStudio in the Actions column. In DataStudio, create a table and upload the data. For details, see Create tables and upload data.

Text displayed as blob in Machine Learning Designer or Studio

Symptom: When you right-click a component on the canvas and select View data, some text appears as blob.

Cause: Characters that cannot be transcoded are displayed as blob.

Impact: This does not affect data processing or retrieval by downstream components. No action is required.

Incremental training support in Machine Learning Designer

Machine Learning Designer supports incremental training for select algorithms, including:

  • Deep learning algorithms

  • EasyRec algorithms

To find out whether the specific algorithm you use supports incremental training, submit a ticket.

Can PAI dataset management connect to data from other S3-compatible object storage services?

No. Currently, only OSS and NAS are supported.

How to create a MaxCompute table

You can execute a CREATE statement in the SQL component or create the table in the corresponding MaxCompute project.

Are TensorFlow 2.x versions supported?

The TensorFlow component supports versions up to 1.15. If you require a later version, you can use the custom Python script component. This component supports more versions and runtime images. For more information, see Python Script.

Is there an API to separately get information from subpages like stdOut and stdErr in the LogView of a training task?

Currently, there is no API to query stdOut or stdErr from LogView.

Data display in Designer shows blob characters

Symptom: On the canvas, when you right-click a component and click View Data from the shortcut menu, some text is displayed as blob characters.

Solution: Some characters are displayed as blob characters because they cannot be transcoded. This does not affect how downstream nodes read and process the data.

How to export a data table from a visual modeling workflow to Excel?

Currently, exporting data directly to an Excel file is not supported. However, if a downstream component stores data in a MaxCompute table, you can use MaxCompute to query the data and then generate an Excel file.