Initial attempts to do NER with seq2seq, but didn't work
I'm using two approaches
- Conditional Random Fields (using CRFSuite)
- RNNs (using TensorFlow seq2seq)
- Python 2.7 (Anaconda recommended)
- TensorFlow 1.0
- (for training) a GPU e.g. NVIDIA 1080
- Choose a data directory and set as an environment variable RNN_HOME
- mkdir $RNN_HOME/raw and put the three files from the task in here
- mkdir $RNN_HOME/preprocessed
- mkdir $RNN_HOME/data
- mkdir $RNN_HOME/train
- In folder rnn, execute Preprocess.ipynb. This will prepare the training data for a format that it can be used to train the RNN.
- In folder rnn, execute train_model.sh
- To execute the trained model, execute run_model.sh