Experimental approaches to constructing knowledge graphs and using them for reasoning tasks
# download CoNLL-2003 tags and prepare train, validation, and test data using CoNLL-2003 scripts
python download_unzip.py \
--url https://www.clips.uantwerpen.be/conll2003/ner.tgz \
--save-directory /Users/tmorrill002/Documents/datasets/conll/raw \
--reuters-file-path /Users/tmorrill002/Documents/datasets/reuters/rcv1.tar.xz
# parse the data into csv files for use with Pandas
python parse.py \
--data-directory /Users/tmorrill002/Documents/datasets/conll/raw/ner \
--save-directory /Users/tmorrill002/Documents/datasets/conll/transformed
# run a baseline algorithm to evaluate performance of a naive approach
# this is the approach detailed in the CoNLL-2003 paper
python baseline.py \
--data-directory /Users/tmorrill002/Documents/datasets/conll/transformed
# run a vanilla LSTM and evaluate the results
python train.py \
--config configs/baseline.yaml
Still using Python 3.8 for now until more libaries are compatible with Python 3.9 (as of December 22, 2020).
# use Python 3.8 installed with homebrew
virtualenv .venv -p /usr/local/opt/python@3.8/bin/python3
source .venv/bin/activate
pip install -r requirements.txt