blob: efa17933f275acbbc365c50c2bea0ccb04e52ece (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
|
#!/bin/bash -x
#####################################################################
# SUMMARY: Train a model on TSV data with guided alignment and data weighting
# TAGS: sentencepiece tsv train align dataweights
#####################################################################
# Exit on error
set -e
# Remove old artifacts and create working directory
rm -rf train_align_weights train_align_weights.{log,out,diff}
mkdir -p train_align_weights
# Run marian command
$MRT_MARIAN/marian \
--cost-type ce-mean --no-shuffle --clip-norm 1 --seed 7777 --dim-emb 32 --dim-rnn 64 --maxi-batch 1 --maxi-batch-sort none --optimizer sgd --learn-rate 0.1 \
-m train_align_weights/model.npz --tsv -t train2.de-w-aln-en.tsv -v $MRT_MODELS/rnn-spm/vocab.deen.{spm,spm} \
--after-batches 60 --disp-freq 4 \
--guided-alignment 2 --guided-alignment-weight 1.0 --data-weighting 1 \
--log train_align_weights.log
# Check if files exist
test -e train_align_weights/model.npz
test -e train_align_weights.log
grep -qi "word alignments from" train_align_weights.log
grep -qi "weights from" train_align_weights.log
# Compare the current output with the expected output
cat train_align_weights.log | $MRT_TOOLS/extract-costs.sh > train_align_weights.out
$MRT_TOOLS/diff-nums.py train_align_weights.out train_align_weights.expected -p 0.01 -o train_align_weights.diff
# Exit with success code
exit 0
|
