All Products
Search
Document Center

MaxCompute:Built-in images

Last Updated:Mar 26, 2026

MaxCompute provides built-in images with pre-installed Python packages for data analytics, scientific computing, and machine learning—including NumPy, pandas, scikit-learn, and XGBoost. Reference these images directly in your jobs instead of managing dependencies manually.

Available images

All built-in images include the base packages from the common image: NumPy, pandas, cloudpickle 2.2.1, pickle 5.0.12, requests 2.31.0, and setuptools 68.0.0.

Image nameCategoryCore libraryPython versionUse for
commonBaseNumPy 1.21.6, pandas 1.3.5Python 3.7General data processing
commonBaseNumPy 1.26.4, pandas 2.2.2Python 3.11General data processing
ubuntuBase OSUbuntu 20.04Python 3.7, Python 3.11UDF development and MaxFrame integration to improve cross-platform portability and execution performance
scipyScientific computingSciPy 1.7.3Python 3.7Statistical analysis and linear algebra
scipyScientific computingSciPy 1.13.0Python 3.11Statistical analysis and linear algebra
statsmodelsStatistical modelingstatsmodels 0.13.5Python 3.7Statistical modeling and econometrics
statsmodelsStatistical modelingstatsmodels 0.14.1Python 3.11Statistical modeling and econometrics
sklearnMachine learningscikit-learn 1.0.2Python 3.7Classification, regression, and clustering
sklearnMachine learningscikit-learn 1.4.2Python 3.11Classification, regression, and clustering
xgboostMachine learningXGBoost 1.6.2Python 3.7Distributed gradient boosting
xgboostMachine learningXGBoost 2.0.3Python 3.11Distributed gradient boosting
pytorchMachine learningPyTorch 1.13.1Python 3.7Natural language processing (NLP)
pytorchMachine learningPyTorch 2.3.0Python 3.11Natural language processing (NLP)
tensorflowMachine learningTensorFlow 2.11.0Python 3.7Programming various machine learning algorithms
tensorflowMachine learningTensorFlow 2.16.1Python 3.11Programming various machine learning algorithms

Python version flags: Use cp37 for Python 3.7 and cp311 for Python 3.11 when setting session flags.

Use built-in images

Built-in images work in MaxCompute SQL user-defined function (UDF), PyODPS, and MaxFrame development.

Important

Specify only one image per job to prevent dependency conflicts.

In SQL UDF jobs

Set the image and Python version at the SQL session level before calling the UDF:

-- For Python 3.11, replace cp37 with cp311
set odps.sql.python.version=cp37;
set odps.session.image = <image_name>;

Replace <image_name> with the image name from the table above (for example, sklearn).

In PyODPS

Note

Upgrade PyODPS to V0.11.5 or later before using images.

Pass the image name to the image parameter of the execute or persist method:

image='<image_name>'

For a complete example, see Use images in PyODPS development.

In MaxFrame

Set the image in config.options.sql.settings before submitting the job:

config.options.sql.settings = {
    "odps.session.image": "<image_name>"
}

View built-in images in the console

  1. Log on to the MaxCompute console and select a region in the upper-left corner.MaxCompute console

  2. In the navigation pane on the left, choose Manage Configurations > Images.

  3. On the Images page, click the Built-in Images tab. The tab lists all available built-in images and their version information.