UiPath AI Fabric Hands-On Exercise & Tutorial 2020

(Complete Working Example with Different ML Models)

Today I’m going to show you exactly how to work with UiPath AI Fabric.

In this comprehensive guide of 3 Blog posts, I’ll cover:

How to Enable AI Fabric
Key terms used in AI Fabric (Projects/Datasets/ML Packages/Pipelines/ ML logs
How to use Out of Box ML Models
How to create the first project using Out of Box ML Models
How to train the model using your datasets
How your own models can be deployed on AI Fabric
And; Lots more insight about AIFabric tool.

So if you want to start building AI-Enabled RPA for the next project, you’ll love this updated blog series.

Let’s get started.

Pre-Requisite

You will need –

UiPath Studio v 2019.10+, with UiPath Robot v 2019.10+ and UiPath Orchestrator 20.4
AI Fabric Enabled Orchestrator Instance (You can request for Free Trial for Enterprise Edition)

How this tutorial is organised

Whats is covered –

Introduction to AI Fabric and How to enable AI Fabric
Understand the key terms used in AI Fabric (Projects/Dataset/ML Packages/ pipelines/ ML logs
Create a basic project using Out of Box ML Models
How your own models can be deployed on AI Fabric
Challenges and Issues faced during deployment

Whats not covered –

Concepts related to Artificial Intelligence.
Setting-Up UiPath Studio, Orchestration and Other Configuration
Setting Up Python Working environment for creating New Deploy-able Model
How to Use C# & Visual Basic Expressions in UiPath

Related Blog Post links –

UiPath AI Fabric Tutorial | Build Your ML Based RPA Project [This Article]
UiPath AI Fabric Tutorial | How to Use ML Skill in UiPath Studio [Part 2]
UiPath AI Fabric Tutorial | How to Deploy Own Model in AI Fabric [In Draft facing some challenges with the deployment will post soon ]

Table of Contents

Introduction to UiPath AI Fabric

What is AI Fabric? Where does it fit in…

Well… AI Fabric is the product in the UiPath Enterprise Platform that enables deploying, consuming, managing and improving Machine Learning models. It can manage models built in-house, by UiPath and our partners, and even open-source models.

You can think AI Fabric as another container(Not wise to say ML Lab) where you can create/manage and deploy your own machine learning model for consumption into intelligent automation workflow.

Once you deploy/enable ML model into AI Fabric it will become available for RPA developer to consume inside workflow using UiPath Studio.

Its become super easy inside the studio to use them as you can select the auto-populated drop-down list with all the successfully deployed ML Skills from the AI Fabric tenant.

After that, all you need to do is pass the data as input to given ML Skill and perform JSON operation on Output and perform Post action based on output received.

Let’s park the idea for now that how you will be able to build the automation using AI Fabric model inside Uipath Studio.

We will cover that in the next part of this tutorial.

Lets first see how we can deploy ML Skill in AI Fabric from scratch using step by step guide.[In Next Section]

Component of AI Fabric

Types of Machine Learning models on AI Fabric

You can reuse the models built by your data science team. You can upload your model into AI Fabric and deploy ML skills.
You will be also able to manage Open-source models built by the data scientist community in AI Fabric, train and deploy them in their workflows as per your use case.
Models built by UiPath’s technology partners, such as H2O.ai (As of now only limited information is available; let’s wait for product update.)
Out-of-the-box models built by UiPath.(Many Quick deployable models such as sentiment analysis, document understanding, language translator etc.)

Key Concepts of AI Fabric

AI Fabric makes it extremely simple to use Machine Learning in the RPA workflows built with UiPath Studio. This way, robots can process unstructured data, better handle uncertainty in decision making, and work with use cases which require human intelligence.

Lets discuss about, the process of creating ML projects in AI Fabric, uploading datasets and training ML packages, deploying them as ML skills in RPA workflows in Studio and configuring the feedback loop by sending data to AI Fabric.

Projects – You can think of the project as a group of isolated resources for a specific ML use case.
Datasets – A folder containing files and folders to train your model.
ML Packages -An ML Package is a folder with all the code and metadata needed to train and serve a Machine Learning model. It can have multiple versions.
Pipeline -A Pipeline is a description of an ML workflow, including all of the functions in the workflow and the order of execution of these functions.
ML Skills – A trained ML model with an acceptable level of performance is ready to be used in an automation
ML Logs – Dashboard to perform analysis on logs when something goes wrong with ML Package.

Types of Pipelines

There are three types of pipelines:

A Training Pipeline takes as input an ML Package and a dataset, together producing a new ML package version;
An Evaluation Pipeline takes as input an ML Package Version and a dataset, together producing a set of metrics and logs;
A Full Pipeline essentially runs a data processing function, and a Training Pipeline and an Evaluation Pipeline in succession.

Create your first project (Step by Step)

#1. How to Access AI Fabric

To experience the full features of AI Fabric and follow along the example in this blog-post, you need an Enterprise account in UiPath Automation Cloud.

The AI Fabric app is licensed as a separate service.

If you don’t have such an account, you can request your UiPath account manager for AI Fabric trail.

As per UiPath product road-map, AI Fabric will become available both on-premises and through the Automation Cloud for community by end of 2020.

As of now its only available to Enterprise account in UiPath Automation Cloud.

To gain access to AI Fabric, change your licensing plan to Enterprise Trial or Enterprise and then allocate AI Robot licenses to your AI Fabric service within a tenant in your Automation Cloud organization.

You will get an error pop-up window like above in case you don’t have required licenses.

The AI Fabric menu option is displayed in the Cloud Portal left navigation bar. You can access the app if your tenant has the necessary licenses and you have the corresponding permissions.

#2. Create ML Projects in AI Fabric

To create a new Project navigate to the Dashboard menu in the left navigation bar and click on AI Fabric . This page will display all your existing project if any. If you need to create new project you can click “Create Project” context menu from top right corner of screen. (Shown below)

In next window, All you need to do is provide below details.

Project name – Give name of the project..This is for reference purpose only (For my example I call it GoogleAppReview )
Project description – Provide meaning full description for your project .

Once done you will land up on dashboard which provides details of your project. as of now we have not added any datasets or ML packages so it would be shown as blank. However this will be useful when we have something up and running.

In the next step, we need to define the datasets for our project.

#3 Create a new Datasets

To create a new Datasets navigate to the Datasets menu in the left navigation bar and create an empty Datasets by clicking “Create New”.

This step is optional.

This depends on the type of ML Package you wish to use. If your model is trainable you can use this screen to upload train and test data.

Once uploaded dataset can be used in Pipeline to train your model.

for this example; I am going to use out-of-box sentiment analysis, non-trainable model(already trained model) to create the project.

we will talk about this in more details in the next part of the tutorial.

#4 Upload/Use Out of Box ML Package

As discussed above ML Package is a folder with all the code and metadata needed to train and serve a Machine Learning model.

You will have two choices here –

Upload a Zip file
Out of the Box Package

Out-of-the-Box ML Packages are Ready-to-use packages in AI Fabric. They are provided by UiPath, by partners or by the community as Open-Source (OS) packages.

for this example, we will use Out-of-the-Box ML Packages.

When you click on Out-of-the-Box ML Packages, You will land up on category page of different types of models and you can select your choice as per requirement of your project.

Out-of-the-box models are appropriate in some contexts. Consider sentiment analysis, where understanding sentiment is similar across different industries and organizations;

But be careful, something working well for others may not work well with you…as it depends on datasets…

Then …You must be thinking why we need Out of Box ML Packages when we can use API provided by Google/Microsoft/IBM/AWS etc. ?

Well, frankly speaking, At the moment I don’t see any benefit unless Uipath has plans to include thousands of trained models which are available for free to use by different big players.

Here you can access the ML Package Details page by first navigating to the ML Packages page and then double-clicking on the name of an ML Package, as shown below:

The important point to note here is –

Input Description – for this example its text (English text for sentiment analysis )
Output Description – what you will get as output. You will receive JSON as output with the classification along with the confidence score

These details would be required inside your workflow to define input and output of ML Skills in UiPath Studio. Good point is you can test that in studio as well.

#5. Create Pipeline

A Pipeline Run is an execution of a pipeline based on code provided by the user. Once completed a Pipeline Run will have associated outputs and logs.

There are three types of pipelines:

A Training Pipeline takes as input an ML Package and a dataset, together producing a new ML package version;

An Evaluation Pipeline takes as input a ML Package Version and a dataset, together producing a set of metrics and logs;

A Full Pipeline essentially runs a data processing function, and a Training Pipeline and an Evaluation Pipeline in succession.

Again an optional step as this would be required when you train your model in AI Fabric.

In case your model is already trained, You can skip this step.

For this example, we can skip this part.

You can use below the screen to Pipeline based on your requirement. You need to select the ML Package you are using, Whats is the version you are using and input dataset.

we will cover this part in the next part of the tutorial.

The Pipeline includes the definition of the inputs required to run the pipeline and outputs to get from this pipeline.

#6 Deploying ML skills

# 7. Check ML Logs

Ml Logs page contains the details of ML Package Validation Events and other related events to the project.

Logs related to the pipeline events are also displayed here.

You can sort logs on the basis of ML Skill Name, Severity and date-time.

Congratulation!

You have successfully created the first project on AI Fabric.

Now its time to consume the AI Fabric ML Skill Deployed by you inside the workflow to create your first ML Skill-based automation.

This is covered as the next part of Blog you can access below.

See you there.

Happy Automation.

Related Blog Post links

UiPath AI Fabric Tutorial | Build Your ML Based RPA Project [This Article]
UiPath AI Fabric Tutorial | How to Use ML Skill in UiPath Studio [Part 2]
UiPath AI Fabric Tutorial | How to Deploy Own Model in AI Fabric [In Draft facing some challenges with the deployment will post soon ]

References –

Image Credit – UiPath Documentation
AI Fabric Docementation – Product Documentation