Creating models and experiments

<< Click to Display Table of Contents >>

Navigation:  Additional Services > Bizagi Artificial Intelligence >

Creating models and experiments

Overview

Artificial Intelligence for Automation Service, is an application which lets you explore machine learning capabilities directly in your Bizagi processes, so processes can rely on a predictive analysis service that is based upon reliable data that you have in a Bizagi Dataset.

Through predictive analysis, you can train models and carry out experiments that rely on renowned machine learning algorithms to determine a given outcome based on stored data with a given degree of certainty.

For such models, you can easily set your Bizagi processes to present a prediction once certain data has been received, either as a default value or to propose to the user.

For introductory information on this application, refer to Bizagi Artificial Intelligence.

 

This section describes how to get started with Artificial Intelligence, to create AI models and experiments.

 

AI models and experiments

To work with Artificial intelligence and its capabilities for Automation Service, you need a data source with which AI services will work.

As is usual in machine learning, you build your model based on sample inputs.

 

Define that data source by creating a model in Bizagi Artificial Intelligence's UI.

For that model, you can create any number of associated experiments. Within each experiment, you can evaluate possible combinations of input parameters and algorithms and the results they return.

 

The results you get will be in terms of a suggested algorithm for a use case statement with a predicted certainty for that given output (attribute).

For the prediction you define relevant attributes as input parameters.

 

The following image illustrates this concept:

 

Cloud_AI_experiment

 

You can create further experiments for different trials to refine the certainty of the prediction so that it is better in new experiments.

You need to know, and specify before the start of the experiment, the attribute on which you want a prediction to get meaningful results. (a supervised learning treatment).

 

Prediction certainty

An AI experiment presents a prediction's certainty or precision.

Precision is an accuracy percentage when you predict a categorical attribute. On the other hand a standard error value is used when you predict a numeric attribute.

 

Accuracy

The prediction's accuracy can be seen as the number of correct predictions divided by the total predictions made (then converted into a percentage).

Determining if a given accuracy is appropriate, depends on the specific use case.

However it is common that an accuracy higher than 70% can be considered as good and reliable (the higher the accuracy, the better).

 

Accuracy applies to attributes with categorical data type.

Examples of categorical data are:

oTypes of a rock: igneous, sedimentary or metamorphic.

oA named geographical place where people live (country, state or city).

oA medical condition which can be diagnosed as true or false.

 

Standard error

The prediction's standard error can be seen as the measure with which a value represents the complete population or data.

Determining if a given standard error is appropriate, depends on the specific use case.

For such decision, we recommend that you analyze the data distribution of the attribute to predict.

The lower the standard error, the better.

 

Standard error applies to attributes with continuous data types.

Examples of continuous data are usually numeric values, such as:

oA duration (e.g a measured elapsed time in seconds).

oA countable quantity of items.

oThe price of an item.

 

note_pin

Even though you should aim for good prediction certainty, this does not protect your model from having high variance or bias.

Make sure that your data is final and reliable and that it does not need further preparation.

We recommend having a large amount of information in the Dataset so you can run reliable experiments that produce meaningful results.

 

Procedure

To create a new model, follow these steps:

 

1. Log in to the Bizagi Artificial intelligence portal.

At https://ai-[CustomerName].bizagi.com/, provide your user credentials and log in.

 

2. Create a new Experiment.

Click New AI project to create an AI project:

 

Cloud_AI4_new

 

The Experiment Wizard opens. Click Let's Go!.

 

Cloud_AI5

 

4.1 Create an A.I. Project.

Create an AI project to help you classify and organize the AI elements you create.

Give the AI project a Name and a meaningful Description, and click Next when you are ready.

 

Cloud_AI._projectInfo_new

 

An AI project is not related to a Bizagi project holding processes.

The AI project's main purpose is to help you organize the set of AI models and experiments you create.

 

4.2 Create an A.I. Model.

Give it a Name and a meaningful Description, and configure the following adequately:

Datasource: Select the dataset in a dataset project that contains the data you want to use.

Environment: Select an environment of the dataset with the data you want to use.

You should always select a dataset with data from a production environment.

Experimenting with non-real data may yield results which aren't applicable to real use of your applications.

 

Cloud_AI_createModel_new

 

Click Next when done.

 

Set the data source from which sample data is taken.

You choose from Business Insights projects, Datasets and their environments, which belong to your Automation Service subscription.

 

The image below shows the Business Insights project and Dataset employed for the AI model's data source:

 

Cloud_AI_datasets

 

6. Create an A.I. Experiment.

Finally, create the experiment. Give it a Name and a meaningful Description.

 

Cloud_AI_createExp2_new

 

Click Finish when done.

 

Cloud_AI6

 

At this point, your experiment is ready help you make predictions.

To learn how to configure the experiment and work with its results, refer to Working with AI experiments.