rasahq/crf-demo

demo code to the conditional random field video

Conditional Random Field Demo

This project demonstrates named entity recognition with a conditional randomfield. It also shows an action server responding to entities with custom python code.

It is maintained by Vincent D. Warmerdam, Research Advocate at Rasa.

Youtube

This repository is part of a youtube video that explains how conditional random fields work for entity exatraction. You can watch this video here

Contents

The project was created with a standard install.

pip install rasa
rasa init --no-prompt

After this setup the nlu.md file was changed to also have entities to detect, the domain.yml file was updated to reflect these changes and the actions.py server was implemented so that you can see the effects while chatting.

In particular the bot can now also detect programming languages.

Start

You can run the entire setup via the Makefile we've provided. The commands can be explained via make help. If you prefer to run without this, you can also run everything by running these commands in the terminal. You will need two terminals.

# terminal one
rasa run actions
# temrinal two 
rasa train
rasa shell --endpoints endpoints.yml

You can see what the action server receives by looking at the action server logs.

Test

If you want to change the NLU pipeline and see the effect. You can do so via;

# before changing `config.yml`
rasa train; rasa test --out before
# after changing `config.yml`
rasa train; rasa test --out before

You will now have two folders with results.

Play

Feel free to play around with this! Happy hacking!

Relevant Articles

There may be some online material that might help you appreciate some details.

• The implementation of CRF in Rasa directly in github.
• Extra maths that details the similarity between CRFs and Logistic Regression