How to get started with Language Understanding on the NeuralSpace Platform
In this short tutorial, we will explain to you how to get started with combined Intent Classification and Entity Recognition, called Language Understanding, on the NeuralSpace Platform.
It’s an easy 10-step process and you will have trained your first Language Understanding model with AutoNLP in one of NeuralSpace’s 87 natively supported languages. Best of all, we will use NeuralSpace’s no-code web interface and achieve state-of-the-art results without writing a single line of code.
The NeuralSpace Platform is a collection of pre-trained Natural Language Processing (NLP) models that can be fine-tuned to each unique use case with AutoNLP.
Different models for Language Understanding, Entity Recognition, Machine Translation, Transliteration, Data Augmentation, Language Detection, Speech to Text, Text to Speech, and many more are available in a no-code web interface, through APIs and the CLI. Thus, users do not need to worry about building state-of-the-art transformer-based deep learning models or finding sufficiently large amounts of data in low-resource languages.
In fact, we will train a Language Understanding model with only 20 examples in code-mixed Gujarati in this tutorial.
10 Steps to Get Started
Without any credit card or other payment details, sign up to NeuralSpace here, activate your account through your verification email and log in.
2. TAKE THE TOUR
After logging in, you will be welcomed by a virtual tour. We know the Skip button is enticing but just take it and you will save yourself a ton of time later on.
3. SELECT/INSTALL APPS
You are on the Dashboard page which is currently empty and will show its value later once your models are deployed. Navigate to the column menu on the left and select one of our services (Language Understanding, Entity Recognition, etc.) or click on Install Apps.
4. SELECT/INSTALL LANGUAGE UNDERSTANDING
Select Language Understanding or install it by clicking on the Install button below the explainer video. Once installed, Language Understanding will pup up in Your Apps in the column menu on the left. Click on it.
5. CREATE PROJECT
The first thing you will see is the Language Understanding Dashboard. As it was with your Platform Dashboard, you can ignore it for now and will appreciate its value once you have models in production. Navigate instead to the right top corner and click on Create project. If you want to read more about the concepts behind projects, check out our Docs.
Give your project a name and select the language of your data.
Tip: Choose Multi-Lingual/Code-Mixed if you expect text in more than one language or a mix of languages and alphabets, for example, Hinglish, Arabizi, etc.
I will call my project Gujaratish because I will create a dataset in Gujarati using the English/Latin alphabet that can be used for a food order chatbot.
Once you have created the project, you will receive a copy of a pre-trained model in the language you selected and can customize it to your unique use case now!
6. DATASET PREPARATION
It’s time to fine-tune your pre-trained model. Navigate to the Data Studio where all of your data is uploaded, annotated, and modified.
7. ADD/UPLOAD/IMPORT EXAMPLES
Let’s add some example sentences now. You will see a field titled Type your examples here.
You can either navigate to the Upload page in the top bar to upload existing datasets, import one of NeuralSpace’s datasets, or write your own examples directly in the Data Studio. Generally, we recommend using one of our pre-uploaded datasets to start with, that can be accessed on the Import page. Currently, there are more than 100 datasets available and we constantly add new ones.
If you want to create your own data in the Data Studio (can be on top of the imported data), you will see a field titled Type your examples in the Data Studio.
The sentence I wrote here means “will my pizza be warm?” and we will first add an intent to it before we tag an entity. Click on Add intent and add this example to your custom intent, here order_query.
Next, we will tag an entity in this sentence. The word pizza is a food item and is important to be recognized in a food order chatbot. Let’s tag it as food_item. Click on Create entity to add a trainable entity that’s not part of the pre-trained entities that come with every pre-trained model.
Add a couple more of such examples and you are prepared to train your custom model with AutoNLP. Scroll a bit up to the top of the Data Studio, select how many training jobs (read more about what training jobs are here), click on Train with AutoNLP and you are ready to — CLICK, TRAIN, CHILL at its best.
You can follow the progress in the Model List at the bottom of your Project Details page.
The Training Status will go from Queued to Training to Completed within a few seconds, but you may need to wait a bit longer for larger datasets. Whereas a small dataset with 20 examples only takes 3–5 seconds to train, a dataset with 5000 examples will be trained in about 15 minutes.
Your model is trained and you can deploy it now by clicking on the Deploy button. Select the number of replicas that can accommodate your expected throughput in terms of Requests per Second. We select 1 here.
Once it’s deployed, you can evaluate it by clicking on the model in the Model List and then Test model in the right menu. Type in any sentence you want to test your model on and see its output. We will try a different food item here.
Click on Parse text and get an output below. The classified intents and recognized entities are ranked by a confidence score.
That was it! 10 simple steps to prepare your dataset, train a state-of-the-art transformer-based deep learning model and evaluate it in code-mixed Gujaratish. (Gujarati+English)
Get started now: NeuralSpace Platform