Rasa as open source alternative to Facebook’s Wit.ai - Migration Guide¶

SUGGEST EDITS

This guide shows you how to migrate your application built with Facebook’s Wit.ai to Rasa. Here are a few reasons why we see developers switching:

Faster: Runs locally - no http requests and server round trips required
Customizable: Tune models and get higher accuracy with your data set
Open source: No risk of vendor lock-in - Rasa is under the Apache 2.0 licence and you can use it in commercial projects

In addition, our open source tools allow developers to build contextual AI assistants and manage dialogues with machine learning instead of rules - learn more in this blog post.

Let’s get started with migrating your application from Wit.ai to Rasa:

Step 1: Export your Training Data from Wit.ai¶

Navigate to your app’s setting page by clicking the Settings icon in the upper right corner. Scroll down to Export your data and hit the button Download .zip with your data.

This will download a file with a .zip extension. Unzip this file to create a folder. The file you want from your download is called expressions.json

Step 2: Create a Rasa Project¶

To create a Rasa project, run:

rasa init

This will create a directory called data. Remove the files in this directory, and move the expressions.json file into this directory.

rm -r data/*
mv /path/to/expressions.json data/

Step 3: Train your NLU model¶

To train a model using your Wit data, run:

rasa train nlu

Step 4: Test your NLU model¶

Let’s see how your NLU model will interpret some test messages. To start a testing session, run:

rasa shell nlu

This will prompt your for input. Type a test message and press ‘Enter’. The output of your NLU model will be printed to the screen. You can keep entering messages and test as many as you like. Press ‘control + C’ to quit.

Step 5: Start a Server with your NLU Model¶

To start a server with your NLU model, run:

rasa run nlu

This will start a server listening on port 5005.

To send a request to the server, run:

    
curl 'localhost:5005/model/parse?emulation_mode=wit' -d '{"text": "hello"}'
    
    
    copied!

The emulation_mode parameter tells Rasa that you want your json response to have the same format as you would get from wit.ai. You can also leave it out to get the result in the usual Rasa format.

Join the Rasa Community Forum and let us know how your migration went!