Add tests with shuffled TSV data

3 files changed, 68 insertions, 6 deletions

diff --git a/tests/interface/input-tsv/.gitignore b/tests/interface/input-tsv/.gitignore
index 191de36..52173ab 100644
--- a/tests/interface/input-tsv/.gitignore
+++ b/tests/interface/input-tsv/.gitignore
@@ -13,12 +13,8 @@ train_vocabs_nopaths
 train_lm
 train_empty_lines
 train_extra_tabs
-train_align
-train_align0
-train_weights
-train_weights0
-train_align_weights
-train_align_stdin
+train_align*
+train_weights*
 
 train.de
 train.en
diff --git a/tests/interface/input-tsv/test_tsv_train_with_align_shuffle.sh b/tests/interface/input-tsv/test_tsv_train_with_align_shuffle.sh
new file mode 100644
index 0000000..dd919a9
--- /dev/null
+++ b/tests/interface/input-tsv/test_tsv_train_with_align_shuffle.sh
@@ -0,0 +1,33 @@
+#!/bin/bash -x
+
+#####################################################################
+# SUMMARY: Train a model on shuffled TSV data with guided alignment
+# TAGS: sentencepiece tsv train align
+#####################################################################
+
+# Exit on error
+set -e
+
+# Remove old artifacts and create working directory
+rm -rf train_align_shuffle train_align_shuffle.{log,out,diff}
+mkdir -p train_align_shuffle
+
+# Run marian command
+$MRT_MARIAN/marian \
+    --seed 4444 --dim-emb 32 --dim-rnn 64 --maxi-batch 1 --maxi-batch-sort none --optimizer sgd --learn-rate 0.1 \
+    -m train_align_shuffle/model.npz --tsv -t train2.aln-de-en.tsv -v $MRT_MODELS/rnn-spm/vocab.deen.{spm,spm} \
+    --after-batches 100 --disp-freq 4 \
+    --guided-alignment 0 --guided-alignment-weight 1.0 \
+    --log train_align_shuffle.log
+
+# Check if files exist
+test -e train_align_shuffle/model.npz
+test -e train_align_shuffle.log
+grep -qi "word alignments from" train_align_shuffle.log
+
+# Compare the current output with the expected output
+cat train_align_shuffle.log | $MRT_TOOLS/extract-costs.sh > train_align_shuffle.out
+$MRT_TOOLS/diff-nums.py train_align_shuffle.out train_align_shuffle.expected -p 0.01 -o train_align_shuffle.diff
+
+# Exit with success code
+exit 0
diff --git a/tests/interface/input-tsv/test_tsv_train_with_align_shuffle_in_ram.sh b/tests/interface/input-tsv/test_tsv_train_with_align_shuffle_in_ram.sh
new file mode 100644
index 0000000..a65bfcf
--- /dev/null
+++ b/tests/interface/input-tsv/test_tsv_train_with_align_shuffle_in_ram.sh
@@ -0,0 +1,33 @@
+#!/bin/bash -x
+
+#####################################################################
+# SUMMARY: Train a model on shuffled (in RAM) TSV data with guided alignment
+# TAGS: sentencepiece tsv train align
+#####################################################################
+
+# Exit on error
+set -e
+
+# Remove old artifacts and create working directory
+rm -rf train_align_shuffle_ram train_align_shuffle_ram.{log,out,diff}
+mkdir -p train_align_shuffle_ram
+
+# Run marian command
+$MRT_MARIAN/marian \
+    --shuffle-in-ram --seed 4444 --dim-emb 32 --dim-rnn 64 --maxi-batch 1 --maxi-batch-sort none --optimizer sgd --learn-rate 0.1 \
+    -m train_align_shuffle_ram/model.npz --tsv -t train2.aln-de-en.tsv -v $MRT_MODELS/rnn-spm/vocab.deen.{spm,spm} \
+    --after-batches 100 --disp-freq 4 \
+    --guided-alignment 0 --guided-alignment-weight 1.0 \
+    --log train_align_shuffle_ram.log
+
+# Check if files exist
+test -e train_align_shuffle_ram/model.npz
+test -e train_align_shuffle_ram.log
+grep -qi "word alignments from" train_align_shuffle_ram.log
+
+# Compare the current output with the expected output
+cat train_align_shuffle_ram.log | $MRT_TOOLS/extract-costs.sh > train_align_shuffle_ram.out
+$MRT_TOOLS/diff-nums.py train_align_shuffle_ram.out train_align_shuffle.expected -p 0.01 -o train_align_shuffle_ram.diff
+
+# Exit with success code
+exit 0