data/download-data.sh

#!/bin/bash

# If you want to add new data files to our Azure storage, open an issue at
# https://github.com/marian-nmt/marian-regression-tests

URL=https://romang.blob.core.windows.net/mariandev/regression-tests/data
TOKEN="${AZURE_STORAGE_SAS_TOKEN:-}"

# If the SAS token is not provided, switch to the mirror server
if [ -z $TOKEN ]; then
    echo "No SAS token provided. Using statmt.org mirror"
    URL=http://data.statmt.org/romang/marian-regression-tests/data
else
    echo "SAS token provided. It has ${#TOKEN} characters"
fi
echo "Downloading from: $URL"

# Each tarball is a .tar.gz file that contains a single directory of the same
# name as the tarball
DATA_TARBALLS=(
  europarl.de-en
  exdb_mnist
)

AZCOPY=true
if ! grep -q "blob\.core\.windows\.net" <<< "$URL"; then
    echo "Warning: URL does not look like Azure blob storage URI. Using wget"
    AZCOPY=false
elif ! command -v azcopy &> /dev/null; then
    echo "Warning: azcopy is not installed in your system. Using wget"
    AZCOPY=false
fi

for name in ${DATA_TARBALLS[@]}; do
    file=$name.tar.gz

    echo Downloading checksum for $file ...
    if $AZCOPY; then
        azcopy copy "$URL/$file.md5?$TOKEN" $name.md5.newest
    else
        wget -nv -O- $URL/$file.md5 > $name.md5.newest
    fi

    # Do not download if the checksum files are identical, i.e. the archive has
    # not been updated since it was downloaded last time
    if test -s $name.md5 && $(cmp --silent $name.md5 $name.md5.newest); then
        echo File $file does not need to be updated
    else
        echo Downloading $file ...
        if $AZCOPY; then
            azcopy copy "$URL/$file?$TOKEN" .
        else
            wget -nv $URL/$file
        fi
        # Extract the archive
        #tar zxf $file
        # Remove archive to save disk space
        #rm -f $file
    fi
    mv $name.md5.newest $name.md5
done

DATA_FILES=(
  europarl.de-en/corpus.bpe.de.gz
  europarl.de-en/corpus.bpe.en.gz
)

for file in ${DATA_FILES[@]}; do
    echo Extracting $file ...
    test -s $file || exit 1
    # Uncompress if needed
    target="${file%.*}"
    test -s $target || gzip -dc $file > $target
done

# Get de-BPEed small training data
test -s europarl.de-en/corpus.small.de.gz || head -n 100000 europarl.de-en/corpus.bpe.de | sed 's/@@ //g' | gzip > europarl.de-en/corpus.small.de.gz
test -s europarl.de-en/corpus.small.en.gz || head -n 100000 europarl.de-en/corpus.bpe.en | sed 's/@@ //g' | gzip > europarl.de-en/corpus.small.en.gz