Installation Update

[mese79]

I changed the setup process by utilizing an installation script (install.sh and install.bat).

#! /bin/sh
# This script will install FeatureForest and its dependencies in a new conda environment.
echo "\nCreate a new conda environment named featureforest with Python 3.10"
conda init
conda create -n featureforest -y python=3.10
echo "\nActivate the featureforest environment"
source activate base
conda activate featureforest
echo "\nInstall PyTorch and torchvision using light-the-torch"
python -m pip install light-the-torch
ltt install 'torch>=2.5.1' 'torchvision>=0.20.1'
echo "\nInstall additional dependencies"
python -m pip install -r ./requirements.txt
echo "\nInstall FeatureForest from GitHub"
python -m pip install git+https://github.com/juglab/featureforest.git

To install PyTorch, this script is using light-the-torch package that can automatically detect local installation of CUDA and its version. Also, segmenta-anything and sam-2 will be installed from their git repos.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 24.85%. Comparing base (ba22200) to head (8303bbf).

Additional details and impacted files

@@           Coverage Diff           @@
##             main      #43   +/-   ##
=======================================
  Coverage   24.85%   24.85%           
=======================================
  Files          40       40           
  Lines        2394     2394           
=======================================
  Hits          595      595           
  Misses       1799     1799           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[jdeschamps]

Oh that's an interesting one! Does it work for metal/macOS or just nvidia systems?

[jdeschamps]

why not the pip package?

[mese79]

You mean why not pip install -r ./requirements.txt ?
Unfortunately, CI test on Windows fails on:
pip install light-the-torch
Because current light-the-torch supports an older version of pip and trying to down(up)grad the pip will throw an error. But with python -m pip install light-the-torch everything's fine.
So I decided tobe safe and used python -m pip everywhere!

[jdeschamps]

No I meant why install git+https://github.com/juglab/featureforest.git rather than the Pip package!

[mese79]

I see. I guess because the git repo is usually more up-to-date.

[mese79]

Oh that's an interesting one! Does it work for metal/macOS or just nvidia systems?

I tried on Mac and it installs the CPU version. Although I'm not sure how can I check if it has MPS accelerator!

[jdeschamps]

Oh that's an interesting one! Does it work for metal/macOS or just nvidia systems?

I tried on Mac and it installs the CPU version. Although I'm not sure how can I check if it has MPS accelerator!

Like this: https://github.com/CAREamics/careamics/blob/d863507c08a5fc72966caa9800407b0160d4cd80/src/careamics/transforms/n2v_manipulate_torch.py#L83

Annoying but these are the two conditions validating that MPS is usable.

[mese79]

Using light-the-torch to install PyTorch on Mac, the torch.backends.mps.is_available() returns True.