kafkasl / contextuallstm Goto Github PK

http://ftp.acc.umu.se/mirror/wikimedia.org/dumps/enwiki/20180120/enwiki-20180120-pages-articles.xml.bz2

The preprocessing for the normal dataset and the one with the context information is a bit different.
They are both intertwined and some cleanup would be nice to remove duplicate code and unify the pipelines.

Traceback (most recent call last):
File "preprocess.py", line 4, in
from preprocess.cleaner import clean_data
File "../src/preprocess/cleaner.py", line 2, in
from pattern.en import tokenize
ImportError: No module named pattern.en
Embeddings path: ../models/idWordVec_500.pklz
Traceback (most recent call last):
File "../src/lstm/lstm.py", line 423, in
tf.app.run()
File "/home/ankit/anaconda2/lib/python2.7/site-packages/tensorflow/python/platform/app.py", line 126, in run
_sys.exit(main(argv))
File "../src/lstm/lstm.py", line 360, in main
embeddings = VectorManager.read_vector(FLAGS.embeddings)
File "../src/utils/vector_manager.py", line 74, in read_vector
with open(filename, 'rb') as f:
IOError: [Errno 2] No such file or directory: '../models/idWordVec_500.pklz'

The run_pipeline.sh only works if the trained embeddings use all the words present in the dataset (i.e. there is no minimum number of ocurrences per word to create its embedding).
If the threshold is higher than 1 it crashes because the script that substitutes removed words with tags has not been called.