pytorch implementation of semantically-conditioned LSTM on multiwoz data
semantically-conditioned LSTM: https://arxiv.org/pdf/1508.01745.pdf
dataSplitFile=resource/woz3/data_split/Boo_ResDataSplitRand0925.json
l=1
lr=0.005
model_path=./sclstm.pt
log=./sclstm.log
res=./sclstm.res
TRAIN
python3 run_woz3.py --data_split=$dataSplitFile --mode=train --model_path=$model_path --n_layer=$l --lr=$lr > $log
TEST
python3 run_woz3.py --data_split=$dataSplitFile --mode=test --model_path=$model_path --n_layer=$l --beam_size=10 > $res
Calculate BLEU
python3 bleu.py --res_file=$res