Testing/Documenting Windows/Mac/Linux Install Procedure

[ian-coccimiglio]

Testing on Python=3.10 first. Tests are being performed on Linux, Windows 10, and Mac M1 architectures. I'm using the env name napari_EABDL, though the name probably should change to something easier to type.

Any changes/deviations from existing instructions are written in comments.

General notes:

• The install process invokes installation of numpy 2.1.x.
• But then, in a later step, numpy gets downgraded back to 1.26.4. Pip's dependency resolver is fickle.
• pip install git+https://github.com/bnorthan/napari-easy-augment-batch-dl.git - fails currently

[ian-coccimiglio]

Linux installation:

• Everything appears to work with the workflow below
• Torch recognizes CUDA
• Tensorflow also recognizes CUDA
• Main libraries worked on import. (Cellpose, Stardist, segment_anything, segment_everything)

conda create -n napari_EABDL python=3.10
conda activate napari_EABDL
pip install numpy==1.26 # 
pip install "napari[all]" # requires quotation marks to parse the square brackets on Linux, not sure if generalizes
pip install albumentations
pip install matplotlib
pip install "tensorflow[and-cuda]" # as above, requires quotation marks
pip install stardist 
pip install gputools==0.2.15 # I think numpy v2.1.2 gets installed here for me
pip install edt # pip throws: numba v0.60.0, tensorflow v2.18.0 require lower versions of numpy. Should still work but iffy.

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
pip install pytorch-lightning # pip throws: reikna 0.9.0 incompatibility with numpy version. Doesn't seem to impact functionality
pip install monai
pip install scipy
pip install tifffile
pip install cellpose
pip install git+https://github.com/True-North-Intelligent-Algorithms/segment-everything.git
### pip install git+https://github.com/bnorthan/napari-easy-augment-batch-dl.git # this one fails, private repo perhaps?
pip install git+https://github.com/True-North-Intelligent-Algorithms/tnia-python.git 
pip install git+https://github.com/True-North-Intelligent-Algorithms/napari-segment-everything.git

[ian-coccimiglio]

Mac M1 installation:

• Torch works (as well as metal)
• Make sure to remove the index-url for torch installation to work
• Tensorflow works (Doesn't seem like there's Metal support, probably fixable but not necessary for upcoming talks)
• Cellpose + Stardist work
• Install process throws a bunch of dependency warnings, but everything seems to pan out mostly.

conda create -n napari_EABDL python=3.10
conda activate napari_EABDL
pip install numpy==1.26 # 
pip install "napari[all]" # also requires quotes on Mac
pip install "tensorflow[and-cuda]" # as above, requires quotation marks. Pip throws a bunch of conflicts, but whatever.
pip install stardist
pip install gputools==0.2.15 # pip dependency incompatibilities
pip install edt 
pip install torch torchvision torchaudio # remove the index flag url
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
pip install pytorch-lightning
pip install monai
pip install scipy
pip install tifffile
pip install cellpose
pip install git+https://github.com/True-North-Intelligent-Algorithms/segment-everything.git
### pip install git+https://github.com/bnorthan/napari-easy-augment-batch-dl.git # this one doesn't work - private repo?
pip install git+https://github.com/True-North-Intelligent-Algorithms/tnia-python.git 
pip install git+https://github.com/True-North-Intelligent-Algorithms/napari-segment-everything.git