Skip to content

Latest commit

 

History

History
356 lines (257 loc) · 14.3 KB

README.md

File metadata and controls

356 lines (257 loc) · 14.3 KB
Raw

Release Build Status codecov

About

   ____                          __      __        _____       _____   
  / _  \____  __  ______  ______/ /_____/ /___    /_   _\     /_   _\  
 / /_/ / __ \/ / / / __ \/ ___\, / __ \, / __ \    0>  0>     <0  <0   
/ ,\ ,/ /_/ /\ \/ / /_/ /\__, / / /_/ / / /_/ /   \__⊽__/     \__⊽__/  
\/  \/\__/\/  \__/ ,___/\____/\/\__/\/\/ ,___/       ⋂  - Hey!   ⋂     
                 \____/                \____/             Olà! -

Ravestate is a reactive library for real-time natural language dialog systems. It combines elements from event-based and reactive programming into an API, where application states are defined as functions that are run when a certain boolean set of criteria (signals) in the current application context is satisfied. It is the first reactive API to allow for boolean combinations of events. You may find a short introductory video here.

Docs 📖

Pretty docs can be found at roboy.github.io/ravestate.

Reactive Hello World

import ravestate as rs

# We want to write some text output, so we
# need the raw:out context property from ravestate_rawio.
import ravestate_rawio as rawio

# Make sure that we use some i/o implementation,
# so we can actually see stuff that is written to rawio:out.
import ravestate_conio as conio

# Ravestate applications should always be wrapped in a Module.
# This allows easier scoping, and enables separation of concerns
# beyond states.
with rs.Module(name="hi!", depends=(rawio.mod, conio.mod)):

    # Create an application state which reacts to the `:startup` signal,
    # and writes a string to raw:out. Note: State functions are
    # always run asynchronously!
    @rs.state(cond=rs.sig_startup, write=rawio.prop_out)
    def hello_world(context):
        context[rawio.prop_out] = "Waddup waddup waddup!"

# Run context with console input/output and our 'hi!' module.
rs.Context("hi!", "conio").run()

Raveboard 🏄

Ravestate has an angular/socket.io-based interactive (beta) UI called Raveboard. It shows the events (spikes) that are currently relevant, as well as potential state activations that are referencing these spikes.

When using raveboard.UIContext instead of Context, or python3 -m raveboard instead of python3 -m ravestate, a real-time visualization of all spikes/activations, as well as a chat window, will be hosted on a configurable port. You can find dedicated docs here.

The following GIF shows raveboard together with ravestate_visionio:

Raveboard

Installation 🔥

Via PIP

The easiest way to install ravestate is through pip:

pip install ravestate

Note: Ravestate requires Python 3.6 or higher. It is tested on Ubuntu 18.04 and 20.04, as well as macOS > High Sierra. It is currently not tested on Windows, but seems to run fine in WSL2.

For reliability, we recommend using an environment virtualization tool with Ravestate, like virtualenv or conda.

Via Docker/Docker-compose

Why Docker?

Ravestate offers a docker image that bundles runtime dependencies that are required for advanced cognitive dialog systems/chatbots:

  • 📦 Neo4j: The Neo4j Graph DBMS is used by Scientio for long-term memory.
  • 💡 Redis: A Redis in-memory DB is used for fast short-term memory, e.g. to store/recall facial feature vectors.
  • 🤦 FaceOracle: A Roboy-developed server-client architecture used by ravestate_visionio for real-time face recognition.
  • 🤖 ROS Melodic: Version 1 of the Robot Operating System for distributed real-time communication. This version of ROS requires a broker process (roscore), which is started automatically inside the container.
  • 🤖 ROS2 Dashing: Version 2 of the Robot Operating System for distributed real-time communication.
  • 🤗 HuggingFace Transformer Models: Language models (ConvAI GPT/OpenAI GPT2) for neural-network-generated conversation.
  • 💌 Roboy ROS Messages: Message defs. that are required to interact with Roboy hardware.

Installing these dependencies by hand is time-consuming and error-prone, so using Docker to ship them makes everyone's lives easier!

How to get?

Option 1: Pulling from Dockerhub

docker pull realitivity/ravestate

Note: You will still need to clone ravestate for the docker-compose file.

Option 2: Building the image yourself

Clone ravestate:

git clone [email protected]:roboy/ravestate && cd ravestate

You can build the ravestate container using the provided Dockerfile:

docker build -t realitivity/ravestate .

Note: Building the container takes time and requires a good internet connection, since all of the dependencies are several Gigabytes in size.

How to run?

Use one of the following docker-compose commands to run ravestate in Docker:

Platform Command
Linux docker-compose up -d rs-linux
macOS docker-compose up -d rs-macos
Windows Not supported yet.

The container is now running and a shell inside the container can be opened with:

docker exec -it rs bash

You can now start ravestate or raveboard as described in the section Running Hello World.

python3 -m ravestate [...]

Which services are exposed from the container?

Service Port Description
Neo4j UI 7474 Neo4j UI for DB stored under <ravestate>/db/neo4j
Neo4j Bolt Interface 7687 Communication with Neo4j DBMS
Redis Database Dump - A dump of the Redis DB in the container can be found under <ravestate>/db/redis
FaceOracle Client Interface 8088 Visualisation for the FaceOracle client.
Raveboard 42424 Default port for raveboard, the ravestate debug UI.

Developing ravestate

Initial configuration and setup

Step 1: Clone the repository and install dependencies:

cd ~

# Create a virtual python environment to not pollute the global setup
python3 -m virtualenv -p python3 python-ravestate

# Source the virtual environment
. python-ravestate/bin/activate

# Clone the repo
git clone [email protected]:roboy/ravestate && cd ravestate

# Install normal requirements
pip install -r requirements.txt

# To run tests & build docs, install pytest, mocking, fixtures, pydoc, ...
pip install -r requirements-dev.txt

# Link your local ravestate clone into your virtualenv
pip install -e .

Step 2: Launch the ravestate docker container as described above. It will serve you Neo4j, which is a backend for Scientio, Roboy's long-term memory system.

Step 3: In the config folder, create a file called keys.yml. This is where your secrets, such as database credentials or your telegram token will go. For example,:

module: telegramio
config:
  telegram-token: <sexycactus>  # This is where your own telegram bot token
                                # will go later

Step 4: You may now conduct your first conversation with ravestate:

python3 -m raveboard -f config/generic.yml -f config/keys.yml

Open raveboard on localhost:42424/ravestate/index.html?rs-sio-url=http%3A//localhost%3A42424 to conduct your first conversation with ravestate.

After the conversation, check the Neo4j interface under localhost:7474. It should now contain some nodes!

Reminder: Whenever you use ravestate from the command line, source the virtual environment first!

Running your Telegram bot

To test your telegram bot with a custom bot token in your keys.yml, just run telegram_test.yml instead of generic.yml. This will load the ravestate_telegramio module.

Setting up PyCharm

  1. Open your local ravestate clone as a project in pycharm.
  2. Under Project Preferences > Python interpreter, set your virtual environment.
  3. Mark the modules folder as sources root via the right-click context menu.
  4. Create a run config via the "Edit configurations menu":
    • Create a new Python configuration.
    • Set raveboard as the module to execute
    • Set the working directory to the git clone directory.
    • Set parameters to -f config/generic.yml -f config/keys.yml.
  5. You should now be able to run the generic ravestate config from PyCharm.

Running Hello World

Ravestate applications are defined by a configuration, which specifies the ravestate modules that should be loaded.

To run the basic hello world application, run ravestate with a config file or command line arguments:

Running with command line spec

You can easily run a combination of ravestate modules in a shared context, by listing them as arguments to python3 -m ravestate:

python3 -m ravestate \
    ravestate_wildtalk \
    ravestate_conio \
    ravestate_hibye \
    ravestate_persqa

Run python3 -m ravestate -h to see more options!

Running with config file(s)

You may specify a series of config files to configure ravestate context, when specifying everything through the command line becomes too laborious:

# In file hello_world.yml
module: core
config:
  import:
    - ravestate_wildtalk
    - ravestate_conio
    - ravestate_hibye
    - ravestate_persqa

Then, run ravestate with this config file:

python3 -m ravestate -f hello_world.yml

Modules

Ravestate offers a landscape of fine-grained modules for different aspects of dialog application tasks, which may be seen in the following dependency diagram. Broadly, the modules are categorized into Core (Blue), I/O (Yellow), External (Red) and Skills (Green):

Core Modules

Module name Description
ravestate Core ravestate library.
ravestate_rawio Provides raw_in, raw_out, pic_in properties, which are served by the IO modules.
ravestate_ontology Connects to scientio to serve a built-in ontology.
ravestate_interloc Provides the all_interlocutors property, where present interlocutors are registered by the IO modules.
ravestate_idle Provides bored and impatient signals, as specified here.
ravestate_verbaliser Utilities for easy management of conversational text, documented here.
ravestate_nlp Spacy-based NLP properties and signals, documented here.
ravestate_emotion Generates signals for, and recognizes specific emotions (sig_shy, sig_surprise, sig_happy, sig_affectionate).
ravestate_ros1 Provides specific Ros1PubProperty, Ros1SubProperty and Ros1CallProperty context properties, which greatly simplify working with ROS1 in ravestate. Documentation here.
ravestate_ros2 Provides specific Ros2PubProperty, Ros2SubProperty and Ros2CallProperty context properties, which greatly simplify working with ROS2 in ravestate.

IO Modules

Module name Description
ravestate_conio Simple command-line based IO for development purposes.
ravestate_telegramio Single- or Multi-process Telegram server module, documented here.
ravestate_roboyio PyroBoy -based STT/TTS with ROS2.
ravestate_visionio See dedicated docs here. Enables face-recognition based dialog interactions.

Skill Modules

Module name Description
ravestate_wildtalk See docs here - runs generative language models (GPT-2, ConvAi, ParlAi)!
ravestate_hibye Simply voices Hi! (or the likes thereof) when an interlocutor is added, and Bye when one is removed.
ravestate_persqa Conducts personalized smalltalk with interlocutors, interacts with Scientio to persist trivia.
ravestate_genqa DrQA -based general question answering module.
ravestate_roboyqa QA module which provides answers to questions about Roboy, such as Who is your dad?
ravestate_akinator (*) Enables dialog-based play of Akinator!
ravestate_sendpics (*) Uses face recognition to extract facial features and an assiciated Person with pic_in and ontology, which are then persisted in Redis and Scientio.
ravestate_fillers Recognize when the dialog context is taking a long time to produce an answer, and voice a filler like "Uhm" or "Let's see...".

Note: (*) = deprecated.

Running tests

If you have built the ravestate docker image as described above, you may run the test suite as follows:

docker run -t -v $(pwd):/ravestate -w /ravestate realitivity/ravestate ./run_tests.sh

Building/maintaining the docs

If you have installed the dependencies from requirements-dev.txt, generate the docs by running this command at project root:

export PYTHONPATH=$PYTHONPATH:$(pwd)/modules
git rm -rf docs
rm -rf _build docs
pydocmd build
mkdir -p docs/resources/docs && cp resources/docs/*.png docs/resources/docs && cp resources/docs/*.gif docs/resources/docs
git add docs/*
# For inspection: python3 -m http.server --directory docs

The structure and content of the docs are defined in the file pydocmd.yml.