- Docker-compose [install guide -> https://docs.docker.com/compose/install/]
After installing docker-compose, go to the root folder of this project and run
docker-compose up
Once the server is up, use the following endpoint.
POST 'localhost:5005/classify?model=[MODEL_NAME]'
model: different models can be used, specify a model name
samples: a json with the samples with the following format
{'samples': [{'seq': 'hello world'}, {'seq': 'other sentence}...]}
It returns an object with the labels for each sample with the following format.
{'samples': [{'seq': '', 'model_version': '1234', 'predicted_labels': {'quality':0.96, 'topic': 'topid_id'}, ...]}
curl -H "Content-Type: application/json" -X POST -d '{"samples":[{"seq": "85.50. Ensure that children living and working on the street are provided with adequate protection, assistance, health care, education and shelter (Hungary);"}]}' 'localhost:5005/classify?model=SDGs'
Learn and apply paragraph to topic training.
This code requires Python 3.7 venv and pip.
To install, run ./run install.
Optional: Install GPU support with ./run pip install tensorflow-gpu==1.15.0.
The code requires a BERT(-like) model to produce sequence (sentence / paragraph) embeddings.
A good starting point is "https://tfhub.dev/google/bert_uncased_L-12_H-768_A-12/1".
Model names starting with "http" are retrieved from tfhub, others are loaded from the given local path.
To run the web server, ./run server.
During development, use ./run devserver.
To run operations from the command line, ./run local --help.
To run nose tests, ./run test.
To run pycodestyle, ./run lint.
This package uses MyPy for Python type checking and intellisense.
To install mypy in vscode, install the 'mypy' plugin and run these:
sudo apt install python3.8-venv python3.8-dev
python3.8 -m venv ~/.mypyls
~/.mypyls/bin/pip install -U wheel
~/.mypyls/bin/pip install -U "https://github.com/matangover/mypyls/archive/master.zip#egg=mypyls[patched-mypy]"