All we do is ML and image processing. We use huge libraries like PyTorch, ones with insane packaging like OpenCV, and things are so stable I haven’t felt the ne

All we do is ML and image processing. We use huge libraries like PyTorch, ones with insane packaging like OpenCV, and things are so stable I haven’t felt the need to introduce version pinning yet. Maybe you’re using recent less-stable projects? I found that the main ML packages rarely break things without months of deprecation warnings.