blob: 905d2d08d6bca9e5621dac17f2f127cfe839ccd1 (
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
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
|
#!/bin/bash -x
# Exit on error
set -e
# Test code goes here
rm -rf corpus corpus*.log
mkdir -p corpus
test -e vocab.de.yml
test -e vocab.en.yml
extra_opts="--seed 1111 --maxi-batch 1 --maxi-batch-sort none --mini-batch 32 --optimizer sgd --dim-emb 128 --dim-rnn 256 --disp-freq 4"
# Added because default options has changes
extra_opts="$extra_opts --cost-type ce-mean --disp-label-counts false"
# Step 1: Train a model in one go, up to the update no. 70, and save training logs
$MRT_MARIAN/marian \
-m corpus/model_full.npz -t $MRT_DATA/train.max50.{en,de} -v vocab.en.yml vocab.de.yml \
--after-batches 70 $extra_opts \
--log corpus.log
test -e corpus/model_full.npz
test -e corpus.log
cat corpus.log | $MRT_TOOLS/strip-timestamps.sh | grep "Ep\. " | sed 's/ : Time.*//' > corpus.expected
# Step 2: Train a new model from scratch, but only to the update no. 40, and save the model
$MRT_MARIAN/marian \
-m corpus/model.npz -t $MRT_DATA/train.max50.{en,de} -v vocab.en.yml vocab.de.yml \
--after-batches 40 $extra_opts \
--log corpus_1.log
test -e corpus/model.npz
test -e corpus_1.log
cat corpus_1.log | $MRT_TOOLS/strip-timestamps.sh | grep "Ep\. " | sed 's/ : Time.*//' > corpus_1.out
cp corpus/model.npz.yml corpus/model.npz.1.yml
# Step 3: Restart the training from step 2 and continue up to the update no. 70, and save training logs
$MRT_MARIAN/marian \
-m corpus/model.npz -t $MRT_DATA/train.max50.{en,de} -v vocab.en.yml vocab.de.yml \
--after-batches 70 $extra_opts \
--log corpus_2.log
test -e corpus/model.npz
test -e corpus_2.log
# Step 4: Combine training logs from steps 2 and 3 and compare them with logs from step 1
cat corpus_2.log | $MRT_TOOLS/strip-timestamps.sh | grep "Ep\. " | sed 's/ : Time.*//' > corpus_2.out
cat corpus_1.out corpus_2.out > corpus.out
$MRT_TOOLS/diff-nums.py corpus.out corpus.expected -p 0.1 -o corpus.diff
# Exit with success code
exit 0
|