Skip to main content

How I did customize "rasa-nlu-trainer" as my own tool


Check out my implementation here

Background

I wanted to have a tool for human beings to classify intents and extract entities of texts which were obtained from a raw dataset such as Rocket.chat's conversation, Maluuba Frames or here. Then, the output (labeled texts) could be consumed by an NLU tool such as Rasa NLU.

rasa-nlu-trainer was a potential one which I didn't need to build an app from scratch. However, I needed to add more of my own features to fulfill my needs. They were:

1. Loading/displaying raw texts stored by a database such as MongoDB
2. Manually labeling intents and entities for the loaded texts
3. Persisting labeled texts into the database

I firstly did look up what rasa-nlu-trainer's technologies were used in order to see how to implement my mentioned features.

At first glance

rasa-nlu-trainer was bootstrapped with Create React App. Create React App is a tool to create a React app with no build configuration, as it said. This tool is also recommended by the official React.js tutorial. I actually was very new to React.js. I wondered what problem Creat React App tried to solve so that I first wanted to set up a project without Create React App. I followed this post.

Well, let me say if I built a React app from scratch then I needed to do some manual configurations for Webpack and Babel including webpack.config.json and .babelrc

Then, I tried to use Create React App, and I followed this guide. All right, let me say Create React App helped me create a React boilerplate project with hiding Webpack and Babel's configuration so that as a developer I only needed to focus on writing production code.

The nuts and bolts

I went on starting to watch how rasa-nlu-trainer's source code was organized and what its techniques are used.

Folder structure

1. Auto-created stuff by Create React App
   - public/index.html
   - src/index.js
2. Overrides
   - They didn't use "react-script", they created their own way instead! That was the reason why some additional folder/files were added such as server.js, scripts and config.

Used technologies

- Programming language: Javascript
- Babel: ES6 transpiler
- Project bundle: Webpack
- Dependencies management: npm
- Backend: Node.js
- Fronted framework/project-based: React.js
- React model storing: React-Redux
- React UI extension components: Ant Design
- TypeScript

Hooray! Welcome me to Javascript world!

Architecture

- index.js: Provider{store} <---> LocaleProvider ---> App
- How "separation of concerns" works: import and  export
- Global data (store) is immutable and only changed by files actions.js and reducer.js

Proposed changes

Finally, I could start to add my needs to the existing rasa-nlu-trainer app. I listed out what my features:
- Creating an overview page as a custom React component to display all loaded texts
- Customizing the existing page for classifying intents and extracting entities.
- Implementing Node.js methods to load/stores texts (as JSON)  which stored in MongoDB

The rule of thumb was I needed to follow the mentioned architecture for modifying the global data which were managed by React-Redux.

You can take a look at the changes from this pull request.

Conclusion

Sometimes, the best way to learn new techs is to get down into an existing tool and customize it with your own needs. 💪
Labels:

Comments

Post a Comment

Popular posts from this blog

Only allow input number value with autoNumeric.js

autoNumeric is a jQuery plugin that automatically formats currency and numbers as you type on form inputs. I used autoNumeric 1.9.21 for demo code. 1. Dowload autoNumeric.js file from  https://github.com/BobKnothe/autoNumeric 2. Import to project <script src="http://ajax.googleapis.com/ajax/libs/jquery/1.11.0/jquery.min.js"></script> <script type="text/javascript" src="js/autoNumeric.js"></script> 3. Define a function to use it <script type="text/javascript"> /* only number is accepted */ function txtNumberOnly_Mask() { var inputOrgNumber = $("#numberTxt"); inputOrgNumber.each(function() { $(this).autoNumeric({ aSep : '', aDec: '.', vMin : '0.00' }); }); } </script> 4. Call the function by event <form> <input type="text" value="" id="numberTxt"/>(only number) </form> <script ty...
Post a Comment
Read more

Why Business Rules Engine Matter

"A good DSL minimizes the 'communication gap' between a domain concept and the code that implements it" - Robert C. Martin A Use Case It looks like adopting a new technology is always driven by a business need. The organizations, such as banks, have their own business processes (such as data gathering and document management). These processes are different from others but usually the differences are not big. How do we build a product which can be reused the similar features but still be adaptable with the specific requirements of each bank? Then, the answer is we should have a library/framework. There is an idea that we can implement this library/framework by using a business rules engine. Which has the following benefits: Be able to modify implementation of business domain at runtime (such as XML, CSV) S ource code is more readable for both developers and domain experts. Therefore, source code can be consider it as a document Since Business Rules Engine...
Post a Comment
Read more

The Evolution of Team Collaboration

I n my point of view, if a team has a good collaboration, team members will achieve the following: To be more effective To make work more enjoyable I have been working for a company for nearly four years as an software developer. Working on various projects from maintaining existing systems to developing a substantial product resulted in me moving to new teams three times. Actually, my most stable team lasted only around three years.  Every time I've moved to a new team, I have a chance to work with new members and a new team culture again. Indeed, I realize that there is a process of developing the team collaboration which gets better time by time. I think it is an evolution ! Phase 1: Poorly collaborate For example, that is when the team members have the following issues: Only work on his/her area of expertise Don't communicate to others Be not confident to take on new challages Don't listen to other members. Subsequently, the team ha...
Post a Comment
Read more

Building Axon.ivy Projects on Bitbucket Pipelines

Read me  if you don't know what Axon.ivy (Ivy) is. Motivation -  Ivy projects are designed to be built on a continuous integration (CI) server like Jenkins - Today, Bitbucket supports for CI with Bitbucket Pipelines - We're using Bitbucket. Then, why not? It must be very cool and convenient for us if we can centralize our CI and VCS (version control system) tools in one place. Here is an approach We have to use a maven plugin called project-build-plugin  to build ivy projects. This plugin requires an instance of Ivy engine during building time. Bitbucket Pipelines allows us to specify our own docker image as a build environment. What we need to do  is to prepare our docker image with needed stuffs such as JDK, Maven, Ivy engine, etc. Step 1. Prepare Docker images For testing purpose, I already created two docker images: Maven and Axon.ivy engine. They are now available on Docker Hub This image for Maven using Oracle JDK 8 This image for Axon.iv...
Post a Comment
Read more

Applying pipeline “tensorflow_embedding” of Rasa NLU

According to this nice article , there was a new pipeline released using a different approach from the standard one ( spacy_sklearn ). I wanted to give it a try to see whether it can help with improving bot’s accuracy. After applying done, I gave an evaluation of “tensorflow_embedding”. It seemed to work better a bit. For example, I defined intents “greet” and “goodbye” with some following messages in my training data. ## intent:greet - Hey! How are you? - Hi! How can I help you? - Good to see you! - Nice to see you! - Hi - Hello - Hi there ## intent:goodbye - Bye - Bye Bye - See you later - Take care - Peace In order to play around with Rasa NLU, I created a project here . You can have a look at this change from this pull request . Yay! When I entered message “hi bot”, then bot with “tensorflow_embedding” could detect intent “greet” with better confidence scores rather than bot with “spacy_sklearn”. The following are responses after executing curl -X POST loc...
Post a Comment
Read more