cirrushuet / rasalit

Visualizations and helpers to improve and debug machine learning models for Rasa Open Source

RasaLit

A collection of helpful viewers that help with understand Rasa NLU components. Some of these views are made using streamlit, hence the wink in the name.

Feedback is welcome.

Contribute

There are many ways you can contribute to this project.

• You can suggest new features.
• You can help review new features.
• You can submit new components.
• You can let us know if there are bugs.
• You can let us know if the components in this library help you.

Feel free to start the discussion by opening an issue on this repository. Before submitting code to the repository it would help if you first create an issue so that we can disucss the changes you would like to contribute. You can ping the maintainer (Github alias: koaning) both in the issues here as well as on the Rasa forum if you have any questions.

Installation

You can install via pip by linking to this github repository.

python -m pip install git+https://github.com/RasaHQ/rasalit

Compatibility

The focus is to support the most recent version of Rasa. We do tag a release every time that we upgrade though. That means you can find compatible versions of rasalit for Rasa 1.10.

Usage

You can directly access the command line app.

> python -m rasalit --help
Usage: rasalit [OPTIONS] COMMAND [ARGS]...

  Helper Views for Rasa NLU

Options:
  --help  Show this message and exit.

Commands:
  diet-explorer  Allows you to explore the DIET settings.
  live-nlu       Select a trained Rasa model and interact with it.
  nlu-cluster    Cluster a text file to look for clusters of intents.
  overview       Gives an overview of all `rasa train nlu` results.
  version        Prints the current version of rasalit.

Features

The app contains a collection of viewers that each specialize in a seperate task.

nlu-cluster

This command allows you to cluster similar utterances in a text file.

Example Usage:

python -m rasalit nlu-cluster --port 8501

This will start a server locally. Internally it is using the whatlies package to handle the embeddings. This means that while the demo is only in English, you can extend the code to work for Non-English scenarios too! For more details, as well as a labelling tool, check out the notebook found here.

overview

This command shows an summary of the intent/entity scores from a rasa train nlu run.

Example Usage:

> python -m rasalit overview --folder gridresults --port 8501

This will start a server locally on port that will displace an interactive dashboard of all your NLU gridsearch data.

To fully benefit from this feature you'll need to run some models first. You can run cross validation of models in Rasa via the command line:

rasa test nlu --config configs/config-light.yml \
              --cross-validation --runs 1 --folds 2 \
              --out gridresults/config-light
rasa test nlu --config configs/config-heavy.yml \
              --cross-validation --runs 1 --folds 2 \
              --out gridresults/config-heavy

Then Rasa, in this case, will save the results in gridresults/config-light and gridresults/config-heavy respectively.

To get an overview of all the results in subfolders of gridresults, you can run the rasalit overview --folder gridresults command from the same folder where you ran the rasa test command. You'll get some simple charts that summarise the intent/entity performance.

live-nlu

This command gives you an interactive gui that lets you see the output of a trained modelling pipeline.

Example Usage:

> python -m rasalit live-nlu --folder models --port 8501

This will start a server locally on port that will displace an interactive playground for your trained Rasa NLU model. You can see the confidence levels as well as the detected entities. We also show some shapes of internal featurization steps.

diet-explorer

This command gives you an interactive visualisation of DIET that allows you to see the available hyperparameters from all the layers in the algorithm.

Example Usage:

> rasalit diet-explorer --port 8501

This will start a server locally on port 8501 that will display an interactive visualisation of the DIET architecture.