Demo: learning with few labeled data and many unlabeled

Demo for learning with few labels webinar. The idea is inspired from (but not the same) Graph Agreement Models.

How to run this

Get elasticsearch: https://www.elastic.co/guide/en/elasticsearch/reference/current/install-elasticsearch.html Run elasticsearch in a separate terminal. If you have mem issues run it like (for example with 1G mem):

ES_JAVA_OPTS="-Xms1g -Xmx1g" ./bin/elasticsearch

Install requirements:

python3.8 -m venv venv
source venv/bin/activate
pip install -r requirements

To train MLM:

python -m src.train_agree_add_loop --only_mlm 1

To train on top default base model:

python -m src.train_agree_add_loop

To train on top of your trained MLM:

python -m src.train_agree_add_loop --base_model_name models/nlu_evaluation_data/mlm/epoch_SOMENUMBER

To run with fasttext as classifier for computing the baseline:

python -m src.train_agree_add_loop --use_fasttext 1

Formatted with black!

pip install black
black -l 80 -t py38 src/*

Contributions from:

License: have fun and share!