this PR drastically reduces the number of explicitily specified dependencies to the required minimum, removes some obsolete dependencies and updates a few of them (tensorflow, pytorch).
The reduction of explicitly defined dependencies keeps the environment working for a lot longer, as packages are often dropped quite frequently from PyPi etc. leading to unresolvable dependencies.
The cpuonly package added to the conda .yml environment file ensures that pytorch is installed without CUDA/GPU support. This is to align it with the requirements.txt based installation which already enforces cpu only mode.
Added pywin32 package (with a sys_platform check to ensure it's only installed on Windows systems). This should not be necessary anymore starting next Thursday when Jupyter isn't dependend upon it anymore
Hi,
this PR drastically reduces the number of explicitily specified dependencies to the required minimum, removes some obsolete dependencies and updates a few of them (tensorflow, pytorch).
cpuonlypackage added to the conda.ymlenvironment file ensures that pytorch is installed without CUDA/GPU support. This is to align it with therequirements.txtbased installation which already enforces cpu only mode.pywin32package (with asys_platformcheck to ensure it's only installed on Windows systems). This should not be necessary anymore starting next Thursday when Jupyter isn't dependend upon it anymoreI tested this on Linux and Windows.