Merge pull request #222 from lvapeab/master

Fixbug build_vocabulary

Author: lvapeab

.travis.yml

@@ -35,7 +35,8 @@ install:
   # Replace dep1 dep2 ... with your dependencies
   - travis_wait 30 conda create --quiet -n test-environment python=$TRAVIS_PYTHON_VERSION
   - source activate test-environment
-  - travis_wait 30 conda install tensorflow=*=mkl* pytables --quiet
+  - travis_wait 30 conda install pytables --quiet
+  - pip install tensorflow==1.14.0
   - pip install -e git+https://github.com/MarcBS/keras#egg=keras --progress-bar off
   - pip install pyux pandas flaky Pillow theano pytest pytest-cache pytest-cov pytest-forked pytest-pep8 pytest-xdist --progress-bar off
   - pip install -e git+https://github.com/lvapeab/coco-caption.git#egg=coco-caption --progress-bar off

keras_wrapper/dataset.py

@@ -1683,14 +1683,7 @@ def preprocessText(self, annotations_list, data_id, set_name, tokenization, buil
             tokfun = None
 
         # Build vocabulary
-        if build_vocabulary:
-            self.build_vocabulary(sentences, data_id,
-                                  max_text_len != 0,
-                                  min_occ=min_occ,
-                                  n_words=max_words,
-                                  use_extra_words=(max_text_len != 0),
-                                  use_unk_class=use_unk_class)
-        elif isinstance(build_vocabulary, str):
+        if isinstance(build_vocabulary, str):
             if build_vocabulary in self.vocabulary:
                 self.vocabulary[data_id] = self.vocabulary[build_vocabulary]
                 self.vocabulary_len[data_id] = self.vocabulary_len[build_vocabulary]
@@ -1706,6 +1699,14 @@ def preprocessText(self, annotations_list, data_id, set_name, tokenization, buil
             if not self.silence:
                 logger.info('\tReusing vocabulary from dictionary for data with data_id "' + data_id + '".')
 
+        elif build_vocabulary:
+            self.build_vocabulary(sentences, data_id,
+                                  max_text_len != 0,
+                                  min_occ=min_occ,
+                                  n_words=max_words,
+                                  use_extra_words=(max_text_len != 0),
+                                  use_unk_class=use_unk_class)
+
         if data_id not in self.vocabulary:
             raise Exception('The dataset must include a vocabulary with data_id "' + data_id +
                             '" in order to process the type "text" data. Set "build_vocabulary" to True if you want to use the current data for building the vocabulary.')