Luis Powered Cli Application

11 Aug 2020

LUIS is one of Microsoft’s Cognitive Services and provides a way to perform text recognition and entity extraction on pieces of plain text.

One of the predefined models they offer illustrates how it can be used. This model allows you to enter a sentence such as “Book me a flight from Edinburgh to Las Vegas” and identifies the intent “Book a flight” and the entities “Source = Edinburgh” and “Destination = Edinburgh”

In order to become a bit more familiar with this service and learn a bit more about how it works I set out to create a console application that would allow developers to see all the pending pull requests and deploy a given pull request to a test environment. With this in mind I created a model that would support the following intents:

• List all Pull Requests
• List all Test Environments
• Deploy a Pull Request to a Test Environment

Use the authoring portal to create the application

Entities

The first thing to do is define the entities that form part of the model. In my use case I only have 1 entity, namely the Pull Request.

Intents and utterances

The next thing to do is to define what intents the bot can perform. If entities are the nouns of the model, then intents are the verbs. For each intent we have to provide examples of what a user might say, this is then used as the training set to teach the underlying Machine Learning algorithm how to classify a user’s input so that our client application can carry out the correct action.

I have 3 main intents that I want the bot to understand:

1. Display a list of all of the pull requests. This is the List Pull Requests intent.
2. Deploy a specific pull request to a test environment. This is the Deploy Pull Request intent.
3. Display a list of all of the test environments. This is the List Environments intent.

As an example of what utterances might look like, here is a list of utterances I entered for the Deploy Pull Requests intent.

From the image you can see that the portion of text that represents the Pull Request identifier has been highlighted and labelled as the Pull Request entity.

Train

We then have to train the model, this is the point where the service looks through each intent and utterance and creates the model that “understands” a user’s input. There isn’t really much to explain at this point, it’s just a case of clicking the button and waiting for the task to complete.

Test

There is a handy test utility in the portal that allows us to enter some text and see what the model predicts and what level of confidence the model has. Given the prediction we can then refine the model further by adding new utterances if necessary. Remember, that if any new utterances are added then the model must be retrained to pick the changes up.

In the above example we test an utterance of deploy pull request 1000 and we can see that the model has predicted the Deploy Pull Request intent with a confidence of 96.9% and has also extracted the PullRequestId entity with a value of 1000.

Publish

Calling the prediction API

Calling the API is really simple, all the information is passed in as a URL with the response being sent back as a JSON object.

Putting it all together