Skip to content

MassDistributionMedia/iris-agent

 
 

Repository files navigation

Iris: A Conversational Agent for Data Science

interface

Read more about the system in this article.

Command DSL

You can add commands to Iris using an in-editor GUI or work directly with the Python DSL. For example:

from iris import state_types as t
from iris import IrisCommand

class GenerateArray(IrisCommand):
    # what iris will call the command + how it will appear in a hint
    title = "generate a random array of {n} numbers"
    
    # give an example for iris to recognize the command
    examples = ["generate numpy array of size {n}"]
    
    # type annotations for each command argument, to help Iris collect missing values from a user
    argument_types = {"n":t.Int("Please enter size of array:")}
    
    # core logic of the command
    def command(self, n):
        import numpy
        return numpy.random.randint(100, size=n)
        
    # wrap the output of a command to display to user
    # by default this will be an identity function
    # each element of the list defines a separate chat bubble
    def explanation(self, result):
        return ["Here are the numbers", result]

Installation guide

Warning! The current version of Iris is an alpha release. We are still adding many new commands to the system and fixing bugs. A production-ready beta release will be out later this summer. Use at your own risk!

These are instructions to install and run Iris in debugging mode. A self-contained Electron app for OSX will be released later this summer.

You will first need to install anaconda. (Make sure to run source ~/.bash_profile after you have installed Anaconda, if it is not appearing in your path.)

Install and run the Python components:

cd iris-agent/backend
# create new conda env called iris with necessary packages
conda create --name iris --file packages.txt
# enter the conda env
source activate iris
# install the remaining pip packages
pip install -r requirements.txt
cd app/
# run the backend application
python app.py

Install and run the Javascript components. If you do not already have webpack, run npm install webpack -g:

cd iris-agent
npm install
# build JS app with webpack (can also run webpack --watch in seperate command window)
webpack
# start electron (this will open the application automatically)
npm start

Both the backend and frontend must be running for Iris to work!

Tests

To verify the backend is working:

cd iris-agent/backend/test
python test.py

About

A extensible conversational agent for data science tasks

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • TeX 34.0%
  • Python 29.6%
  • Tcl 29.2%
  • JavaScript 2.5%
  • Makefile 2.3%
  • C 1.8%
  • Other 0.6%