How to handle Python dependencies

Processing data

The main part of your Python script is the logic for calculating the output from the input. This is done by your Python code, which can use configuration files and persistent ML models in a well-defined Python environment. To define these resources, the PythonComponent class of the simaticai.deployment module is used to add dependencies or additional files. These files are included in the configuration package and extracted in the AI Inference Server as follows:

Dependencies are installed on the server via pip.
Additional files are copied into the component directory.

Python dependencies

The AI Inference Server executes every component of a pipeline in an isolated Python virtual environment. For each component, you must specify which Python packages are required by the Python scripts in that component, including the Python packages required to load persistent Python objects.

Adding Python dependencies

The Python dependencies of a component can be added in several ways:

As a standard wheel file or a zip/tar.gz file that contains standard wheel files.In both cases, the packages, which can be precompiled wheel files or pure Python source distributions, are added to the component.dependencies dictionary and binaries are zipped into the configuration package. AI Inference server supports only installing source distributions that contain only Python source code.

component.add_python_packages('../packages/my_module-0.0.1-py3-anyany.whl')
component.add_python_packages('../packages/MyPackages.zip')
component.add_python_packages('../packages/my_source_module-0.0.2.tar.gz')

Recommendations and limitations for using source distributionsPrefer using wheel files instead of source distributions where possible.Transitive dependencies of a source distribution must be added to the requirements.txt specifying explicitly the version that matches the targeted Python runtime.Should there be any unidentified dependencies left, AI SDK will raise an error, which will guide you through resolution.

Using a standard requirements.txtThis way, the component.dependencies dictionary contains only the dependencies defined in the file. The required wheel files are downloaded while converting the edge configuration package to an edge configuration package using pip.

NOTE: The conversion step can be skipped by using the simaticai.deployment.Pipeline.export() method which is available from AI SDK v2.1.0.

component.set_requirements('./requirements.txt')

By name using a list that contains the names of the Python modules In this case, the method searches for the module in the current Python environment and adds the package with its version and all of its transitive dependencies.

component.add_dependencies(['numpy', 'scikit-learn'])

Dependencies can be added by name and version using a list that contains corresponding tuples. When the component is saved, it will perform a check if all specified dependencies can be installed together. Transitive dependencies will also be downloaded.

component.add_dependencies([('pandas', '1.3.0'), ('pyarrow', '3.0.0')])

Dependencies added to a component will be installed on the AI Inference Server with the defined version and can be imported into your Python code during execution.

# entrypoint.py or data_processor.py
import numpy as np
import pandas as pd
[...]

Download from non-public repository

Dependencies from non-public repositories can be downloaded by specifying an extra index URL at the beginning of the requirements.txt file.

--extra-index-url https://<API_KEY>@your/private/repository
simaticai

Please be aware that if a package is also available in a public repository, pip may download it from there and not look for it in the private repository, which may pose a cybersecurity risk. It is recommended to only use trusted private repositories, pin the version of the package, and check if a package with the same properties already exists on https://pypi.org/simple.

If you want to change the default https://pypi.org/simple package index, you can do it by using an index URL at the beginning of the requirements.txt file. This allows you to download dependencies exclusively from a private repository.

--index-url https://<API_KEY>@your/private/repository
simaticai

Dependency limitations

Due to the limitations of the AI Inference Server, the AI SDK will issue a warning when specific dependencies are requested. TensorFlow is subject to such limitations, as it is recommended to use TFLite instead.

The following code for example:

component.add_dependencies([('tensorflow', '2.18.0')])

Will result in the warning message:

WARNING: TensorFlow is imported. For better performance with an already trained model, consider using TensorFlow Lite (tflite) instead.

Furthermore, the AI SDK will generate an error if a source-only package containing non-Python source code is encountered. Such an example is the following code, where pysqlite3 is added to the dependencies, which contains code segments written in C language.

component.add_dependencies([("pysqlite3", "0.5.4")])

pipeline = deployment.Pipeline.from_components([component], name="pipeline_with_pysqlite3")
package_path = pipeline.export(tmpdir)

The export call on the pipeline will fail with the following error message:

ERROR: One or more source dependencies are not pure Python sources. You need to convert them to wheel files for the target platform manually.