Erechtheus / ner_LSTM

Named Entity Recognition using LSTM/CRF in Keras

ner_LSTM

Named Entity Recognition using LSTM/CRF in Keras This is a rather simple and nowadays old-school implementation of NER using a BiLST and an optional CRF layer. The project uses rather outdated packages, maybe it works with recent packages as well?

prerequisite

Download embeddings from https://cloud.devmount.de/d2bc5672c523b086/ Install requirements.

This works for Python 3.8

conda create -n ner python=3.8 conda activate ner

conda install tensorflow==2.4.3 conda install scikit-learn==0.23.2 conda install gensim==3.8.3 conda install xopen==0.9.0 conda install numpy==1.19.5 pip install tensorflow_addons==0.11.2 pip install wandb==0.12.2

This works Python 3.9

conda create -n ner2 python=3.9 conda activate ner2

conda install tensorflow-gpu==2.6.2 conda install scikit-learn==0.23.2 conda install gensim conda install xopen==0.9.0 pip install tensorflow_addons pip install wandb

##Train your model using trainGenerator.py Please use this to train a new model and not train.py! It is much faster than the train.py und leads to better results

Evaluate your model test.py

Test a NER model dataset.

testGenerator.py

Tests a model on the dataset using a generator; but there seems to be some error in the implementation...