docs/lol-paper/README.md

In neurodata/fselect: Linear Optimal Low-Rank Projection

Figure Reproduction

Note: Code for Figures 1, 2, 4, and Supplementary Figure 7 can be found in the corresponding MATLAB package.

Figure 3

Figure 3 can be reproduced (from scratch) using the simulations driver, with dependencies suitably installed, from terminal:

Rscript sims_driver.R

This script, in full, took us about 500 core-hours.

this will produce a file in the directory, data/sims/lol_sims_lda.rds, or you can use our version of the outputs, found here. This can then be used in the Rmarkdown Figure 3 to produce Figure 3.

Figure 5

To reproduce Figure 5, the simplest approach, in our opinion, is to leverage the FlashLol docker container. This docker container contains the necessary dependencies to use both FlashX, a package for rapid numerical computation (optionally) using semi-external memory, and the LOL FlashX implementation, which also implements PCA, CCA, RRLDA, and Random Projections.

To build the docker container, navigate to the appropriate directory, and build like normal:

cd flashR/
docker build neurodata/flashlol:0.0.3 .

Optionally, pull the existing docker container:

docker pull neurodata/flashlol:0.0.3

Next, make a new directory somewhere on your machine; let's assume it's called <data>. Create a sub-directory, <data>/dwi.

one can navigate to the neurodata.io/mri cloud, and download the "Diffusion MRI" >> "Aligned Images" for all of the individuals with Diffusion MRI connectomes. Put in a directory, called <data>/<Dataset>/dwi, for each of the datasets employed.

Next, download the coresponding phenotypic .csv file for each dataset, also from neurodata.io/mri. Place this at <data>/<Dataset>/<Dataset>.csv.

Finally, for each dataset, one can run the file using the docker container:

docker run -ti --entrypoint /bin/bash -v <data>/<Dataset>:/brains -v <path>/<to>/<lol>/<repository>/:/lol neurodata/flashlol:0.0.3
cd /lol/docs/lol-paper/flashlol-figure/
Rscript flashlol_corr_driver.R

which will create a file at <data>/<Dataset>/Dataset-<Dataset>_flashlol.rds. A procedure to set this up on Amazon AMIs can be found here.

neurodata/fselect documentation built on March 6, 2021, 12:54 p.m.