Meta-repository in which I keep environments and repositories from which to obtain trained predictors where to estimate SCE on.
- keep repositories,
- within each keep trained models
- within each make a predict / logits script
- align data loaders to compare at instance-level between models
- second best decoding
We need to get some logit space that stands for all possible label sequences So
N_test x L x K x K % x K #nesting depth, at most K
Combination-wise: L^(K x K)
=> use K-level CRFs for decoding, so average the energy over K CRFs.
These are now our logits! The decoding and predictions are special in its own right, will have to use the repo :/ => check which elements required to form TouristLeMC object
% have to create a new object containing this? % can simplify a lot [only nonbayesian] %% whatever gets me started fastest %%% also will have to create a vector/tensor representing y_true & correctness
- have to adapt to Transformers
- already converted ACE to conll format
- next step to get embeddings
This: https://github.com/Adaxry/get_aligned_BERT_emb might come in handy
- XML format (original)
- nested format (secondbest-decoding-NER)
- conll format (seq2seq-NER)
- jsonlines format (biaffine-NER)
- (rasa NLU format) (pyramid-NER)
Have converters from 1-2, 2-3.
######################################
var="XXX" % https://git-scm.com/book/en/v2/Git-Tools-Submodules
- install virtualenv $var
sudo apt-get install python3-pip
sudo apt-get install python3-venv
sudo pip3 install virtualenv
sudo pip3 install virtualenvwrapper
# set the following .bashrc
export VIRTUALENVWRAPPER_PYTHON=/usr/bin/python3.8
export WORKON_HOME=$HOME/.virtualenvs
source /usr/local/bin/virtualenvwrapper.sh
# optional bashrc values
alias py='python3'
ulimit -n 9048
export PYTHONIOENCODING=utf8
#FINALLY run
mkvirtualenv -p /usr/bin/python3.8 -a $HOME/code/SP-calibration-NER/$var $var
- install poetry and finish virtual environment
#Using poetry and the readily defined pyproject.toml, we will install all required packages
workon $var
pip3 install poetry
cd $HOME/code/$var
poetry install