Chapter 9: Testing and Production Deployment

In this chapter, we will introduce how to test Rasa projects. We will then discuss how to verify NLU data and stories, as well as how to evaluate the performance of NLU models and Dialogue management models. Through testing, we can find errors in projects as early as possible. We can also comprehensively evaluate the performance of bots.

Moving on, we will discuss how to deploy Rasa applications in production environments. We will discuss the choice of deployment methods, model storage, tracker stores, and locker stores. By properly deploying Rasa applications, we can implement model version management, load balancing, service expansion, and other functions in production environments.

We will cover the following topics:

• Testing Rasa projects
• Deploying your Rasa assistant to production

Let's talk about validation and evaluation first, because they are executed before deployment in the software development process...

Testing Rasa projects

In this section, we will start by discussing how to validate data and stories. This step is used to find obvious bugs. Later, we will discuss how to evaluate NLU performance and how to read the corresponding reports. Finally, we will introduce the test story format and learn how to use test stories to evaluate the performance of Dialogue management.

Validating data and stories

If developers can quickly detect whether there are errors and where these potential errors are in NLU data and stories, this can help developers greatly improve work efficiency. In Rasa, there is a command for this purpose:

rasa data validate

The preceding command will detect errors in the data and configuration. Common errors include the following:

• Inconsistency of the training data (the same training data appearing in two or more different intents)
• The intents in the training data being inconsistent with the intents in the domain file (fewer or more intents)
...

Deploying your Rasa assistant to production

Here we introduce how to deploy your Rasa assistant to production.

When to deploy

It is common to use the minimum viable product or MVP strategy during the product development process. MVP is all about building a usable product prototype that fulfills the key requirements in the most efficient and simple way and then iterating to fine-tune the product details.

In Rasa, the official recommendation is that a product can be put into production as an MVP as soon as it can handle the most important (but not every) "happy path" of Dialogue. It's recommended to use Rasa X to have early users test the product prototype. This is in order to continuously improve the model until the product prototype reaches the MVP standard and you are ready to deploy it to a production environment.

Deployment options

When we want to deploy a Rasa assistant on a large scale, we normally use solutions based on Kubernetes or OpenShift. Rasa...

Summary

In this chapter, we discussed two very important stages in the development of a Dialogue system: testing and deployment. Testing is very important for us to ensure the intelligence of a Dialogue system. We must find the current problems of the Dialogue system through testing and correct these problems. We also discussed how to deploy Rasa projects to production environments. A real large-scale Dialogue system needs to be accessed by tens of thousands or even millions of users at the same time. Such a Dialogue system must have very good horizontal scalability. Fortunately, Rasa considered these issues at the beginning of the design and provided corresponding solutions. By using a central storage system, tracker store, and lock store, we are able to extend our service smoothly.

In the next chapter, we will discuss a user-centered methodology and the tools required for developing Dialogue systems.