def __init__(self, en_dir, img_path, jp_dir, img_original_dir=None, img_correspondence_path=None, jp_original_dir=None, compress_word_dim=100, compress_img_dim=100, line_flag=False):
# log setting
program = os.path.basename(__name__)
self.logger = logging.getLogger(program)
logging.basicConfig(format='%(asctime)s : %(name)s : %(levelname)s : %(message)s')
self.english_feature = txt.TextFeatures(en_dir, compress_dim=compress_word_dim, feature_name='eng')
self.japanese_feature = txt.TextFeatures(jp_dir, jp_original_dir, compress_dim=compress_word_dim)
self.image_feature = img.ImageFeatures(img_path, img_original_dir, img_correspondence_path, compress_img_dim)
self.gcca = gcca.GCCA()
self.cca = gcca.CCA()
self.line_flag = line_flag
self.logger.info("===== initializing =====")
self.logger.info("line_flag:%s", self.line_flag)
self.__prep_dir()
def get_trigram_features(self):
with open('/data/corpora/cspan/top_trigrams.pkl', 'rb') as f:
trigrams = pickle.load(f)
phrases = [t.lower() for t in self.s.alignment.get_phrase_text()]
feats = []
keys = [u[0] for u in trigrams]
for phrase in phrases:
h = {}
#keys = [u[0] for u in trigrams]
for k in keys:
h[k] = 0
grams = text_features.TextFeatures(phrase).get_n_grams(3)
for gram in grams:
if gram in keys:
h[gram] = 1
feats.append(hash_to_list(h))
return np.array(feats)
sys.path.insert(0, '../core')
import speech, text_features
with open('/data/jrgillick/speeches.pkl') as f:
bigram_vocab = {}
bigram_counts = {}
trigram_vocab = {}
trigram_counts = {}
speeches = pickle.load(f)
bigram_index = 0
trigram_index = 0
for s in tqdm(speeches):
phrases = s.alignment.get_phrase_text()
for phrase in phrases:
tf = text_features.TextFeatures(phrase)
bigrams = tf.get_n_grams(2)
trigrams = tf.get_n_grams(3)
for bigram in bigrams:
if bigram not in bigram_vocab:
bigram_vocab[bigram] = bigram_index
bigram_counts[bigram] = 1
bigram_index += 1
else:
bigram_counts[bigram] += 1
for trigram in trigrams:
if trigram not in trigram_vocab:
trigram_vocab[trigram] = trigram_index
trigram_counts[trigram] = 1
def get_euphony_features(self):
text = self.s.alignment.get_phrase_text()
tfs = [text_features.TextFeatures(t) for t in text]
feats = [hash_to_list(tf.get_euphony_features()) for tf in tfs]
return feats
def get_name_feature_names(self):
text = self.s.alignment.get_phrase_text()
tfs = text_features.TextFeatures(text[0]).get_name_features()
keys = tfs.keys()
keys.sort()
return keys
def get_mean_vector_features(self):
text = self.s.alignment.get_phrase_text()
return [
text_features.TextFeatures(unicode(text)).get_mean_vector(t)
for t in text
]
def get_name_features(self):
text = [t.lower() for t in self.s.alignment.get_phrase_text()]
tfs = [text_features.TextFeatures(t) for t in text]
feats = [hash_to_list(tf.get_name_features()) for tf in tfs]
return feats