Monte Carlo Cross-Validation helper script

mccv

#!/bin/bash
# 
# Monte Carlo Cross-Validation helper script
#
nfolds=$1
natlases=$2
ntemplates=$3
pool=(input/pool/brains/*.mnc)

function usage {
cat <<EOF
Samples from labelled images to create MAGeT input folders for cross validation
folds.
    
usage: 
  $(basename $0)  folds atlases templates

  folds      - number of validation folds to produce
  altases    - number of atlases to use in each fold
  templates  - number of templates to use in each fold

Expects that in the current folder is ./input/pool/ that contains brains/ and
/labels/ folders of images (in brains/) and corresponding labels (in labels/).
Labels should be named following the MAGeT brain conventions. E.g. 
  brains/s1.mnc  ->   labels/s1_labels.mnc
  brains/s2.mnc  ->   labels/s2_labels.mnc
  brains/s3.mnc  ->   labels/s3_labels.mnc

Each "fold" is a new folder in input/ (e.g. input/fold1) that has an atlases/
templates/ and subjects/ folder containing images randomly sampled from the
pool according to the parameters you set. For instance, if you run this script
like so: 

  $(basename $0) 10 5 19

You'll have a ten fold directories created, with each one containing a
different atlas library (of 5 images and labels), subject library (the
remaining images in the pool), and template library (19 of subjects).

NOTE: this script uses 'cp -l' to create hard links to the images in your pool
folder, rather than make full copies (run 'man ln' to learn more about hard
links).

# Using the fold directories with MAGeT 

Once you have created these fold folders, you can then run MAGeT so that it
uses one of these folds as its input. e.g. 

  mb --input_dir=input/fold1 --output_dir=output/fold1 run 

NOTE: since the folds will likely share registrations (from atlas->template,
and template->subject), if you are doing many folds to save time you may want
to run all the pairwise registrations in the entire pool. Exercise left for the
reader. 
EOF
}

if [[ -z "$nfolds" || -z "$natlases" || -z "$ntemplates" || ! -e input/pool ]]; then
    usage
    exit 1
fi

for fold in $(seq $nfolds); do
    #shuffle
    pool=($(printf "%s\n" "${pool[@]}" | sort -R))
    atlases=("${pool[@]:0:$natlases}")
    subjects=("${pool[@]:$natlases}")
    templates=("${subjects[@]:0:$ntemplates}")

    folddir=input/fold$fold
    mkdir -p $folddir/{atlases,templates,subjects}/{brains,labels}
    cp -l "${atlases[@]}" $folddir/atlases/brains/
    tmp=("${atlases[@]/.mnc/_labels.mnc}")
    cp -l "${tmp[@]/brains/labels}" $folddir/atlases/labels
    cp -l "${templates[@]}" $folddir/templates/brains/
    cp -l "${subjects[@]}" $folddir/subjects/brains/
done