niiMLr

The goal of niiMLr is to wrap functions from other neuroimaging and deep learning packages to facilitate 3D neural network modeling. Many preprocessing functions depend on an AFNI installation. For Windows users, this will also require installation of the Windows Subsystem for Linux (WSL). AFNI installation instructions can be found at: https://afni.nimh.nih.gov/pub/dist/doc/htmldoc/background_install/install_instructs/index.html.

Installation

niiMLr is not yet published to CRAN, but you can install the development version from GitHub with:

# install.packages("devtools")
devtools::install_github("willi3by/niiMLr")

Python

niiMLr requires Python in order to run TensorFlow. Before using niiMLr, you may need to tell reticulate which version of Python to use.

Installation on M1 Macs (click to view)

Getting TensorFlow to work on an M1 Mac currently requires some extra work. (The default Ananconda installation, for example, use the wrong installation of TensorFlow.) Follow the instructions at https://developer.apple.com/metal/tensorflow-plugin/ to install the correct one using miniforge, a community-driven distribution that supports ARM.

Then, use this code to tell reticulate which Python installation to use:

Sys.setenv(RETICULATE_PYTHON = paste0(Sys.getenv("HOME"), "/miniforge3/bin/python"))

Alternatively, you can use homebrew to install miniforge, use the installation script from the tensorflow_macos github page, and install dependencies:

brew install miniforge
conda create --name tensorflow_macos python=3.8 numpy
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/apple/tensorflow_macos/master/scripts/download_and_install.sh)"

During the script installation, you will have to specify the path to the newly created conda env, which should be something like: /opt/homebrew/Caskroom/miniforge/base/envs/tensorflow_macos

After installation of tensorflow, activate the environment conda activate tensorflow_macos and run the following to install dependencies that must be installed with conda instead of pip:

conda install numba
conda install scikit-learn
conda install scipy

Finally, in R, set the RETICULATE_PYTHON variable to the python in the conda env:

Sys.setenv(RETICULATE_PYTHON = "/opt/homebrew/Caskroom/miniforge/base/envs/tensorflow_macos/bin/python")