progress bar and version bump

CHANGELOG.md

@@ -1,6 +1,12 @@
 CHANGELOG
 ---------
 
+v0.3.8:
+  - multiprocessing support (get_vocab and apply_bpe)
+  - progress bar for learn_bpe
+  - seed parameter for deterministic BPE dropout
+  - ignore some unicode line separators which would crash subword-nmt
+
 v0.3.7:
   - BPE dropout (Provilkov et al., 2019)
   - more efficient glossaries (https://github.com/rsennrich/subword-nmt/pull/69)

setup.py

@@ -11,7 +11,7 @@ def test_suite():
 
 setup(
     name='subword_nmt',
-    version='0.3.7',
+    version='0.3.8',
     description='Unsupervised Word Segmentation for Neural Machine Translation and Text Generation',
     long_description=(codecs.open("README.md", encoding='utf-8').read() +
                       "\n\n" + codecs.open("CHANGELOG.md", encoding='utf-8').read()),
@@ -28,6 +28,8 @@ def test_suite():
     'Programming Language :: Python :: 2',
     'Programming Language :: Python :: 3',
     ],
+    install_requires=['mock',
+                      'tqdm'],
     packages=find_packages(),
     entry_points={
         'console_scripts': ['subword-nmt=subword_nmt.subword_nmt:main'],

subword_nmt/learn_bpe.py

@@ -25,6 +25,12 @@
 from multiprocessing import Pool, cpu_count
 from collections import defaultdict, Counter
 
+try:
+    from tqdm import tqdm
+except ImportError:
+    def tqdm(iterator, *args, **kwargs):
+        return iterator
+
 # hack for python2/3 compatibility
 from io import open
 argparse.open = open
@@ -294,7 +300,7 @@ def learn_bpe(infile, outfile, num_symbols, min_frequency=2, verbose=False, is_d
 
     # threshold is inspired by Zipfian assumption, but should only affect speed
     threshold = max(stats.values()) / 10
-    for i in range(num_symbols):
+    for i in tqdm(range(num_symbols)):
         if stats:
             most_frequent = max(stats, key=lambda x: (stats[x], x))