run python ARNN.py -h for more detailed hyperparameter settings.
For example,
ARNN.py [options] <WVFolder> <JSONOutputFile>
python3 ARNN.py --dataset ner --dropRate 0.2 random ./json_output.txt
python3 ARNN.py --dataset ner --dropRate 0.2 ../WV/201308_p/structured_linear-2/cbow/ ./json_output.txt
python3 ARNN.py --dataset ner --dropRate 0.2 --emb_dimension 500 random ./json_output.txt
python3 ARNN.py --dataset ner --dropRate 0.2 --emb_dimension 500 ../WV/201308_p/structured_linear-2/cbow/ ./json_output.txt
set the THEANO Environment Variable as
THEANO_FLAGS=cuda.root=/usr/local/cuda,device=gpu,floatX=float32
For multiple input word embeddings, use some thing like this "[file1,file2,file3]", not white space. For example,
python3 ARNN.py --lvrg 0 --h_win 0 --dataset chunk --attention general --emb_dimension 500 [../WV/201308_p/word_linear-2/skip/,../WV/201308_p/structured_linear-2/skip/] ./json_output_concat.txt
toy pre-trained word embeddings can be found:
embedding: http://202.112.113.8/d/ARNN-SL/WV/201308_p_word_linear-2_skip/sgns.words25.npy
vocab: http://202.112.113.8/d/ARNN-SL/WV/201308_p_word_linear-2_skip/sgns.words25.vocab
Following script will install the theano-related packages that needed for running the code.
sudo apt-get install python3-numpy python3-scipy python3-dev python3-pip python3-nose g++ libopenblas-dev git
sudo pip install Theano
##calculate the average similarity of each word pairs with the same label.
label indicate named entity or pos tag.
Run following:
python3 DistanceInLabels.py ../WV/201308_p_word_linear-2_skip/sgns.words100.npy ../WV/201308_p_word_linear-2_skip/sgns.words100.vocab ./json_output.txt