In the paper Customization Scenarios for De-identification of Clinical Notes by Hartman et al. from Google LLC, we used a text de-identification system that's very similar to this code. This code is based on NeuroNER, an open source program that performs named-entity recognition (NER). Website: neuroner.com.
NOTICE: This is not an officially supported Google product.
- Embedding: glove.840B.300d.zip
- Types:
AGE
,CITY
,DATE
,EMAIL
,ID
,MEDICALRECORD
,NAME
,PHONE
,STREET
,ZIPCODE
- We have released a labeling of Physionet Gold Corpus following the I2B2-2014 guidelines in order to facilitate fair comparison deid-annotations.
- We comprised three versions for each note in our train set to robustify our model to case changes: lower, upper and original.
-
Model Training:
run.py --train --dataset_text_folder=<dataset> --token_embedding_dimension=300 --output_folder=<something> --threads_tf=128 --threads_prediction=10
-
Evaluate saved training models @epoch:
-
Save model @epoch to
trained_model
directory:share_model.py
-
Run evaluation of a model on a dataset:
util.py --eval --pretrained_model_folder=<trained_model> --dataset_text_folder=<dataset> --rbias=0 --edim=300
-
Evaluate binary/typed results in
results.json
:eval.py --metrics=<binary|token> --datasets=<eval-on-dataset-folder>
-
Next is the rest of the original NeuroNER README.md
.
NeuroNER is a program that performs named-entity recognition (NER). Website: neuroner.com.
This page gives step-by-step instructions to install and use NeuroNER.
NeuroNER relies on Python 3, TensorFlow 1.0+, and optionally on BRAT:
- Python 3: NeuroNER does not work with Python 2.x. On Windows, it has to be Python 3.6 64-bit or later.
- TensorFlow is a library for machine learning. NeuroNER uses it for its NER engine, which is based on neural networks. Official website: https://www.tensorflow.org
- BRAT (optional) is a web-based annotation tool. It only needs to be installed if you wish to conveniently create annotations or view the predictions made by NeuroNER. Official website: http://brat.nlplab.org
For GPU support, GPU requirements for Tensorflow must be satisfied. If your system does not meet these requirements, you should use the CPU version. To install neuroner:
# For CPU support (no GPU support):
pip3 install pyneuroner[cpu]
# For GPU support:
pip3 install pyneuroner[gpu]
You will also need to download some support packages.
- The English language module for Spacy:
# Download the SpaCy English module
python -m spacy download en
- Download word embeddings from http://neuroner.com/data/word_vectors/glove.6B.100d.zip, unzip them to the folder
./data/word_vectors
# Get word embeddings
wget -P data/word_vectors http://neuroner.com/data/word_vectors/glove.6B.100d.zip
unzip data/word_vectors/glove.6B.100d.zip -d data/word_vectors/
- Load sample datasets. These can be loaded by calling the
neuromodel.fetch_data()
function from a Python interpreter or with the--fetch_data
argument at the command line.
# Load a dataset from the command line
neuroner --fetch_data=conll2003
neuroner --fetch_data=example_unannotated_texts
neuroner --fetch_data=i2b2_2014_deid
# Load a dataset from a Python interpreter
from neuroner import neuromodel
neuromodel.fetch_data('conll2003')
neuromodel.fetch_data('example_unannotated_texts')
neuromodel.fetch_data('i2b2_2014_deid')
- Load a pretrained model. The models can be loaded by calling the
neuromodel.fetch_model()
function from a Python interpreter or with the--fetch_trained_models
argument at the command line.
# Load a pre-trained model from the command line
neuroner --fetch_trained_model=conll_2003_en
neuroner --fetch_trained_model=i2b2_2014_glove_spacy_bioes
neuroner --fetch_trained_model=i2b2_2014_glove_stanford_bioes
neuroner --fetch_trained_model=mimic_glove_spacy_bioes
neuroner --fetch_trained_model=mimic_glove_stanford_bioes
# Load a pre-trained model from a Python interpreter
from neuroner import neuromodel
neuromodel.fetch_model('conll_2003_en')
neuromodel.fetch_model('i2b2_2014_glove_spacy_bioes')
neuromodel.fetch_model('i2b2_2014_glove_stanford_bioes')
neuromodel.fetch_model('mimic_glove_spacy_bioes')
neuromodel.fetch_model('mimic_glove_stanford_bioes')
BRAT is a tool that can be used to create, change or view the BRAT-style annotations. For installation and usage instructions, see the BRAT website.
Perl is required because the official CoNLL-2003 evaluation script is written in this language: http://strawberryperl.com. For Unix and Mac OSX systems, Perl should already be installed. For Windows systems, you may need to install it.
NeuroNER can either be run from the command line or from a Python interpreter.
To use NeuroNER from the command line, create an instance of the neuromodel with your desired arguments, and then call the relevant methods. Additional parameters can be set from a parameters.ini
file in the working directory. For example:
from neuroner import neuromodel
nn = neuromodel.NeuroNER(train_model=False, use_pretrained_model=True)
More detail to follow.
By default NeuroNER is configured to train and test on the CoNLL-2003 dataset. Running neuroner with the default settings starts training on the CoNLL-2003 dataset (the F1-score on the test set should be around 0.90, i.e. on par with state-of-the-art systems). To start the training:
# To use the CPU if you have installed tensorflow, or use the GPU if you have installed tensorflow-gpu:
neuroner
# To use the CPU only if you have installed tensorflow-gpu:
CUDA_VISIBLE_DEVICES="" neuroner
# To use the GPU 1 only if you have installed tensorflow-gpu:
CUDA_VISIBLE_DEVICES=1 neuroner
If you wish to change any of NeuroNER parameters, you can modify the parameters.ini
configuration file in your working directory or specify it as an argument.
For example, to reduce the number of training epochs and not use any pre-trained token embeddings:
neuroner --maximum_number_of_epochs=2 --token_pretrained_embedding_filepath=""
To perform NER on some plain texts using a pre-trained model:
neuroner --train_model=False --use_pretrained_model=True --dataset_text_folder=./data/example_unannotated_texts --pretrained_model_folder=./trained_models/conll_2003_en
If a parameter is specified in both the parameters.ini
configuration file and as an argument, then the argument takes precedence (i.e., the parameter in parameters.ini
is ignored). You may specify a different configuration file with the --parameters_filepath
command line argument. The command line arguments have no default value except for --parameters_filepath
, which points to parameters.ini
.
NeuroNER has 3 modes of operation:
- training mode (from scratch): the dataset folder must have train and valid sets. Test and deployment sets are optional.
- training mode (from pretrained model): the dataset folder must have train and valid sets. Test and deployment sets are optional.
- prediction mode (using pretrained model): the dataset folder must have either a test set or a deployment set.
A dataset may be provided in either CoNLL-2003 or BRAT format. The dataset files and folders should be organized and named as follows:
- Training set:
train.txt
file (CoNLL-2003 format) ortrain
folder (BRAT format). It must contain labels. - Validation set:
valid.txt
file (CoNLL-2003 format) orvalid
folder (BRAT format). It must contain labels. - Test set:
test.txt
file (CoNLL-2003 format) ortest
folder (BRAT format). It must contain labels. - Deployment set:
deploy.txt
file (CoNLL-2003 format) ordeploy
folder (BRAT format). It shouldn't contain any label (if it does, labels are ignored).
We provide several examples of datasets:
data/conll2003/en
: annotated dataset with the CoNLL-2003 format, containing 3 files (train.txt
,valid.txt
andtest.txt
).data/example_unannotated_texts
: unannotated dataset with the BRAT format, containing 1 folder (deploy/
). Note that the BRAT format with no annotation is the same as plain texts.
In order to use a pretrained model, the pretrained_model_folder
parameter in the parameters.ini
configuration file must be set to the folder containing the pretrained model. The following parameters in the parameters.ini
configuration file must also be set to the same values as in the configuration file located in the specified pretrained_model_folder
:
use_character_lstm
character_embedding_dimension
character_lstm_hidden_state_dimension
token_pretrained_embedding_filepath
token_embedding_dimension
token_lstm_hidden_state_dimension
use_crf
tagging_format
tokenizer
You are highly encouraged to share a model trained on their own datasets, so that other users can use the pretrained model on other datasets. We provide the neuroner/prepare_pretrained_model.py
script to make it easy to prepare a pretrained model for sharing. In order to use the script, one only needs to specify the output_folder_name
, epoch_number
, and model_name
parameters in the script.
By default, the only information about the dataset contained in the pretrained model is the list of tokens that appears in the dataset used for training and the corresponding embeddings learned from the dataset.
If you wish to share a pretrained model without providing any information about the dataset (including the list of tokens appearing in the dataset), you can do so by setting
delete_token_mappings = True
when running the script. In this case, it is highly recommended to use some external pre-trained token embeddings and freeze them while training the model to obtain high performance. This can be done by specifying the token_pretrained_embedding_filepath
and setting
freeze_token_embeddings = True
in the parameters.ini
configuration file during training.
In order to share a pretrained model, please submit a new issue on the GitHub repository.
You may launch TensorBoard during or after the training phase. To do so, run in the terminal from the NeuroNER folder:
tensorboard --logdir=output
This starts a web server that is accessible at http://127.0.0.1:6006 from your web browser.
If you use NeuroNER in your publications, please cite this paper:
@article{2017neuroner,
title={{NeuroNER}: an easy-to-use program for named-entity recognition based on neural networks},
author={Dernoncourt, Franck and Lee, Ji Young and Szolovits, Peter},
journal={Conference on Empirical Methods on Natural Language Processing (EMNLP)},
year={2017}
}
The neural network architecture used in NeuroNER is described in this article:
@article{2016deidentification,
title={De-identification of Patient Notes with Recurrent Neural Networks},
author={Dernoncourt, Franck and Lee, Ji Young and Uzuner, Ozlem and Szolovits, Peter},
journal={Journal of the American Medical Informatics Association (JAMIA)},
year={2016}
}