Sequence-labeling

Sequence Labeling implemented by RNN/GRU/LSTM using Tensorflow.

Install requirements

• tensorflow 1.6
• numpy
• python 3.6

Model hyperparameters

• init_scale: the initial value of parameters
• learning_rate: the initial value of learning rate
• max_grad_norm: the maximum norm value of gradient
• num_layers: the layer number of model
• hidden_size: the node number of hidden layer
• epoch: the the total number of epochs for training
• keep_prob: the dropout probability
• lr_decay: the decay of the learning rate
• batch_size: the number of inputs
• model: model type (rnn/gru/lstm)
• save_path: the folder to save parameter after training
• data_file: data of model
• kfold: the k-fold value of cross-validation test method

Evaluation

• kfold cross-validation method (warning: because of extremely long running time, in deep learning area, we don’t use kfold cross-validation to evaluate model. This is just a funny experiment.)
• f1 score metric

Training and evaluate model

python rnn_model.py --config_file=config.cfg