Manage (create and delete) a throwaway subset/ directory containing some files copied from an existing source training/ directory.

create-subset.sh

#!/bin/bash

# create-subset.sh
#
# Create a throwaway subset/ directory containing files copied from an existing source
# training/ directory. Assumes .jpg. Once you are done with the (redundant) subset/ and
# want to clean it up, it's recommended to use remove-subset.sh instead of manual
# commands, since with these scripts we are operating close to training data and we
# don't want a mistake to result in the deletion of the wrong directory.

export size=100 # number of images per category... change this

export fisher_yates_shuffle='for(@l=<>,$i=@l;--$i;){$j=int rand($i+1);next if $i==$j;@l[$i,$j]=@l[$j,$i];}print(@l);'

which perl > /dev/null || { echo "perl was not found. This tool relies on perl to randomly partition the data. Exiting without making any changes."; exit -1; }

if [[ -d "subset" ]]; then
  echo "subset/ already exists."
  echo "Exiting without making any changes."
  exit -1
fi

if [[ ! -d "training" ]]; then
  echo "training/ source directory not found."
  echo "Exiting without making any changes."
  exit -1
fi

cd training

for category in *; do
  if [[ -d "$category" ]]; then
    echo doing [$category]
    mkdir -p ../subset/training/$category
    cd $category
    for file in `find . -type f -name "*.jpg" -print | perl -e "$fisher_yates_shuffle" | head -$size`; do
      cp $file ../../subset/training/$category
    done
    cd ..
  fi
done

remove-subset.sh

#!/bin/bash

if [[ -d "subset" ]]; then
  /bin/rm -r subset
fi