This guide we'll walk you through a series of steps of installing and setting up the necessary software for running the notebooks in the AI SDK Tutorials. For Windows we recommend using the WSL2 feature, which creates a virtual Linux environment. Having this, the rest of the steps are identical to setting up the work environment on a native Linux machine.
Prerequisites
As AI SDK (simaticai) is a Python library, and Python is an interpreted language, it ensures OS independence, so you can install it into any Python environment following the installation instructions described below.
For the installation, you will need:
Python environment managerEnvironment managers help create, manage, and switch between different Python environments, ensuring project dependencies are isolated and conflicts are minimized.
Separate, virtual Python environmentAt the time of writing, supported Python versions are 3.10 and 3.11. The virtual environment enables you to separate the new dependencies from previously installed Python libraries. The virtual environment prevents incompatibility issues between different requirements.
Python dependency managerDependency managers help install, update, and manage the libraries and packages your Python project depends on. Recommended dependency manager is pip.
Internet connectionDuring its installation, AI SDK will require installing its Python dependencies, which are available on the internet. In case you have a private Package Registry with the required Python packages, an internet connection is not necessary.
Known issues
Path length issue
Sometimes the path of project files can be too long. On Windows by default paths have a 256 character length limitation, which we recommend to change. It can be changed via the Group Policy in Computer Configuration > Administrative Templates > System > Filesystem > Enable NTFS long paths. Or when you install Python for Windows, the installer has an extra step at the end to disable the path length limitation. You will need administrator privileges in both cases.
TensorFlow Lite issue
Some of our notebooks suggests to use TFLite instead of the full TensorFlow package. This is a smaller and faster runtime for TensorFlow models. It is possible to use either on AI Inference Server, but less powerful Edge Devices can only run TFLite models. However, if you want to test your model locally, currently there is no official TFLite runtime for Windows. You can either use the full TensorFlow package, or you can build your TFLite models in a virtual Linux environment.