enviroment:
tensorflow 1.4
python 3.5
tqdm
Samples for three datasets are shown in data_sample/
preprocessed data can be found in data/webnlg_seq/
To start training :
First, set the experiment setting in config.json file. For example, "data_name" can be written as "NYT" or "WEBNLG", "model_name" can be written as "baseline", "para" or "layer"
Second, in file "config.py", "train_flag" can be set as "0", means that initialing the parameters randomly and start training, "train_flag" set as "1", means that initialing the parameters from pkl file.
Third, use the command as follows, and start training:
python main.py