dougalk / conversational_ai_music_bot

conversational_ai_music_bot

This repo is a work in progress. It demonstrates a music chatbot that can converse with users about music use cases.

The following functionality is currently coverered:

For recommendations:

• music recommendations via artist
• music recommendations via song title

For music trivia questions:

• Album Release Date Questions
• Album Tracklist Questions
• Artist Biography Questions

Pre-requisites

Check if your Python env is already configured python3 --version pip3 --version

Mac Instructions. For Windows or Linux, please see here: https://rasa.com/docs/rasa/installation/environment-set-up

1. Install homebrew if you haven't already (https://brew.sh/)
2. Update homebrew: brew update
3. brew install python

Clone the repo

1. See https://docs.github.com/en/repositories/creating-and-managing-repositories/cloning-a-repository for more details.

How to set up your RASA environment, using Python version 3.9

How to set up

1. Create a virtual env using python 3.9: python3.9 -m venv ./venv
2. Activate the virtual env: source ./venv/bin/activate
3. Make sure pip is up to date: pip3 install -U pip
4. Install rasa: pip3 install rasa
5. Install other requirements: pip install -r requirements.txt
6. Next you need to set up your client credentials, see the next section

How to set up client credentials. For example, for use of the Spotify API

1. Go to the .env.example file in the root directory
2. Rename .env.example to .env
3. Add the credentials in for each. The services needed are listed below:
   • Spotify: https://developer.spotify.com/documentation/web-api
   • Last FM: https://www.last.fm/api/authentication

Run rasa train to train the model

Install Streamlit

1. pip install streamlit
2. Check it is working (streamlit will launch in a browser window): streamlit hello
3. To stop the Streamlit server, press: ctrl-C
4. Press deactivate if you want to exist the venv

How to run and test the bot manually

1. Run the rasa server: rasa run -m models --enable-api --cors="" or rasa run -m models --enable-api --cors="" --debug

   • we use --cors"*" (Cross Origin Resource Sharing) to allow all traffic to connect. In production we set cors to only accept requests from specific origins

2. Split terminal

3. Run the action server rasa run actions (make sure you are in the correct folder!)

4. Split terminal again

5. In the new split terminal, run the streamlit app (make sure you are in the correct folder, cd streamlit): streamlit run main.py

6. The app will launch in a new window!

7. Alternatively; just run this to test the bot in the command line: rasa shell nlu

How to train the model

1. Make changes
2. Run: rasa train
3. Use above steps

How to run tests

Unit Tests

1. How to test each function using pytest

1. Run: pytest

Model Testing - Natural Language Understanding (NLU)

1. Testing each story using rasa test

1. Run: rasa test --stories tests/test_stories.yml

2. Cross Validating the model

This will perform NLU model evaluation with cross-validation by splitting the training data into multiple sets and training/testing the model several times to get an average performance metric

1. run: rasa test nlu --cross-validation

3. Test the the entire project (NLU, stories/rules, end-to-end conversations)

This will test:

• NLU evaluation: how well the model classifies intents and extracts entities based on the provided examples in the NLU training data using metrics such as precision, recall, and f1-score for intents and entities
• Dialogue Management (Core): tests stories and rules
• [Not utilising] End to End testing: test conversations if end-to-end examples are place in the tests/ directory

1. run: rasa test

conversational_ai_music_bot's Issues

Improve intent recognition (CALM)

Use Rasa CALM or LLMs more generally to get intents, for example prompt engineering.

Fix album release date questions with Rasa Calm

Utilise Hugging Face Open Source models if possible. Mistral an option.

https://rasa.com/blog/how-to-put-next-generation-ai-assistants-in-production-today-with-calm/

Or look into Autogen