syft | CLI tool and library for generating a Software Bill of Materials from container images and filesyste | Continuous Deployment library

PySyft currently does not support torch 1.8. See pysyft issue torch 1.8

The version of syft you are using is v0.3.0+. It does not contain syft.TorchHook as I believe I read that it is not needed there. Use pip install syft==0.2.9 to use syft.TorchHook(). Especially if you are following that Secure and Private Course from Andrew Trask, Openmined.

Note that if you are indeed following that course, a new updated course has been introduced by the Openmined team. See https://courses.openmined.org/courses/foundations-of-private-computation for more details. It is based on an updated version of syft.

You have two problem shere:

1. You're setting the return value of get_running_loop, but an exception isn't a return value. If you want your mocked code to raise an exception, you need to configure a side_effect.

2. Your code catches the RuntimeError and doesn't re-raise is: you simply set self.loop = None and log an error. This means that even when you successfully raise a RuntimeError from get_event_loop, that exception will never be visible to your tests because it is consumed by your code.

If you were to mock your logger object, you can check that logger.error was called with the exception. E.g.:

It looks the module you are trying to use is either deprecated or has a new version. See here the file is not active.

Try searching if a newer version of this protocol exists or any similar protocol that can cater to your needs.

Edit:

It may also be that you are using an older version Syft and so I recommend you to upgrade your pip and Syft. Follow the instructions here

Not all PyTorch versions are available on Python's package registry PyPI. For instance, the CPU only version or any Windows version is only available on PyTorch's custom registry. Selecting one of these versions on PyTorch - Get Started Locally will give you an installation command including the custom registry. Installing PySyft also installs PyTorch and the specific version you're getting, requires adding the custom registry:

I came across this problem as well and pushed a fix in https://github.com/OpenMined/PySyft/pull/2948