Add tests for creating *.spm vocabs

4 files changed, 88 insertions, 0 deletions

diff --git a/data/download-data.sh b/data/download-data.sh
index a1e123a..15eed62 100755
--- a/data/download-data.sh
+++ b/data/download-data.sh
@@ -19,3 +19,7 @@ for file in ${MODEL_FILES[@]}; do
         fi
     fi
 done
+
+# Get de-BPEed training data
+test -s europarl.de-en/corpus.de.gz || pigz -dc europarl.de-en/corpus.bpe.de.gz | sed 's/@@ //g' | pigz > europarl.de-en/corpus.de.gz
+test -s europarl.de-en/corpus.en.gz || pigz -dc europarl.de-en/corpus.bpe.en.gz | sed 's/@@ //g' | pigz > europarl.de-en/corpus.en.gz
diff --git a/tests/sentencepiece/setup.sh b/tests/sentencepiece/setup.sh
new file mode 100644
index 0000000..eff1524
--- /dev/null
+++ b/tests/sentencepiece/setup.sh
@@ -0,0 +1,7 @@
+# Skip if compiled without SentencePiece
+if [ ! $MRT_MARIAN_USE_SENTENCEPIECE ]; then
+    exit 100
+fi
+
+test -f $MRT_DATA/europarl.de-en/corpus.en || exit 1
+test -f $MRT_DATA/europarl.de-en/corpus.de || exit 1
diff --git a/tests/sentencepiece/test_create_joint_spm_vocab.sh b/tests/sentencepiece/test_create_joint_spm_vocab.sh
new file mode 100644
index 0000000..eeac67b
--- /dev/null
+++ b/tests/sentencepiece/test_create_joint_spm_vocab.sh
@@ -0,0 +1,36 @@
+#!/bin/bash -x
+
+#####################################################################
+# SUMMARY: Create a joint SentencePiece vocabulary
+# AUTHOR: snukky
+#####################################################################
+
+# Exit on error
+set -e
+
+# Remove old artifacts and create working directory
+rm -rf vocab.joint vocab.joint.*{log,out,diff}
+mkdir -p vocab.joint
+
+# Run marian command
+$MRT_MARIAN/marian \
+    --no-shuffle --seed 1111 --dim-emb 32 --dim-rnn 64 --maxi-batch 1 --maxi-batch-sort none --after-batches 1 \
+    -m vocab.joint/model.npz -t $MRT_DATA/europarl.de-en/corpus.{en,de} \
+    --dim-vocabs 8000 -v vocab.joint/vocab.ende.spm vocab.joint/vocab.ende.spm \
+    --log vocab.joint.log
+
+# Check if files exist
+test -e vocab.joint/model.npz
+test -e vocab.joint/vocab.ende.spm
+test -e vocab.joint.log
+
+# Check logging messages
+grep -q "Creating SentencePiece vocabulary.* vocab.ende.spm" vocabs.joint.log
+grep -q "Sampling from.* corpus.en.* corpus.de" vocabs.joint.log
+
+# Extract a textual vocabulary and compare with the expected output
+$MRT_MRT/spm_export_vocab --model vocab.joint/vocab.ende.spm > vocab.joint.out
+$MRT_TOOLS/diff-nums.py vocab.joint.out vocab.joint.expected -o vocab.joint.diff
+
+# Exit with success code
+exit 0
diff --git a/tests/sentencepiece/test_create_spm_vocabs.sh b/tests/sentencepiece/test_create_spm_vocabs.sh
new file mode 100644
index 0000000..700bdc1
--- /dev/null
+++ b/tests/sentencepiece/test_create_spm_vocabs.sh
@@ -0,0 +1,41 @@
+#!/bin/bash -x
+
+#####################################################################
+# SUMMARY: Create SentencePiece vocabularies
+# AUTHOR: snukky
+#####################################################################
+
+# Exit on error
+set -e
+
+# Remove old artifacts and create working directory
+rm -rf vocabs vocabs.*{log,out,diff}
+mkdir -p vocabs
+
+# Run marian command
+$MRT_MARIAN/marian \
+    --no-shuffle --seed 1111 --dim-emb 32 --dim-rnn 64 --maxi-batch 1 --maxi-batch-sort none \
+    -m vocabs/model.npz -t $MRT_DATA/europarl.de-en/corpus.{en,de} \
+    --dim-vocabs 4000 4000 -v vocabs/vocab.en.spm vocabs/vocab.de.spm \
+    --after-batches 1 \
+    --log vocabs.log
+
+# Check if files exist
+test -e vocabs/model.npz
+test -e vocabs/vocab.en.spm
+test -e vocabs/vocab.de.spm
+test -e vocabs.log
+
+# Check logging messages
+grep -q "Creating SentencePiece vocabulary.* vocabs.en.spm" vocabs.log
+grep -q "Creating SentencePiece vocabulary.* vocabs.de.spm" vocabs.log
+
+# Extract a textual vocabulary and compare with the expected output
+$MRT_MRT/spm_export_vocab --model vocabs/vocab.en.spm > vocabs.en.out
+$MRT_TOOLS/diff-nums.py vocabs.en.out vocabs.en.expected -o vocabs.en.diff
+
+$MRT_MRT/spm_export_vocab --model vocabs/vocab.de.spm > vocabs.de.out
+$MRT_TOOLS/diff-nums.py vocabs.de.out vocabs.de.expected -o vocabs.de.diff
+
+# Exit with success code
+exit 0