Solving Unity ML-Agents Training Freeze and StrictVersion Errors
Recently, while setting up a training environment using Unity ML-Agents 3.0.0 and Python ML-Agents 1.1.0, I encountered a series of frustrating issues. At first, training would start normally after pressing the Play button in Unity, but as soon as I switched to another window (like the terminal or a browser), Unity would freeze, and the Python console would throw an error:
AttributeError: 'StrictVersion' object has no attribute 'version'
Unity would become unresponsive, and training could not proceed.
Problem Analysis
After some careful investigation, I found that the root cause was Unity's default behavior of pausing the game when it loses window focus.
By default, Unity Editor suspends rendering and physics calculations when it is not the active window. However, ML-Agents requires continuous, uninterrupted communication between Unity and the Python API during training. If Unity freezes (even briefly), the Python side interprets this as a communication breakdown and crashes with errors like 'StrictVersion' object has no attribute 'version'.
Solution
The solution is straightforward:
✅ Enable "Run In Background" in Unity's Player Settings so that Unity continues executing even when it is not the active window.
Steps:
- Open Unity Editor.
- Go to
Edit → Project Settings → Player. - Expand
Resolution and Presentation. - Find and enable the checkbox for
Run In Background. - This ensures that Unity keeps running smoothly in the background, maintaining a live connection with the Python API during training.