Publishing an AI experiment

<< Click to Display Table of Contents >>

Navigation:  Cloud applications > Bizagi Artificial Intelligence > Working with AI experiments >

Publishing an AI experiment

Overview

Once you have created an AI experiment and generated the model, as described at Working with AI experiments, and when deciding to settle with its result, you may choose to publish it so that it can be used in a production environment.

 

Before you continue

Publishing an experiment means that you are completely sure that the results thrown by that experiment are satisfactory (its accuracy or standard error).

Before doing so, consider:

1. You should publish experiments whose models have been generated against production data (targeting a production-environment Dataset).

2. You should be aware that a published experiment will imply use in a production environment, and therefore you will not be able to modify or delete such experiment afterward.

3. You should maintain a published experiment, by periodically training/generating the model.

As it is common in machine learning capabilities, you need to look after the learning curve of your model so that you ensure it reaches an optimal behavior.

Note that even though you may have your Bizagi processes constantly feed the source Dataset, the AI experiment will not change or act upon new information unless you explicitly use its Generate model option.

4. You should ensure that the data is final and that it doesn't need further preparation once it resides in the Dataset.

This also means that as stated above, using the Generate model frequently will help you to train the model so that it is more reliable, however this does not guarantees that your data is exempt of having high variance or bias.

It is usually recommended to have large amount of information in the Dataset in order to produce reliable experiments.

 

Procedure

To publish an experiment, follow these steps:

 

1. Click the Publish button.

Do this where you are shown with the experiment's results:

 

Cloud_AI_publish

 

Bizagi Artificial Intelligence will prompt you if you are sure to publish this prediction, recall there is only one published experiment per model. Publishing a new experiment results in un-publishing the previous one and projects will no longer be able to use it.

Click Publish now > button.

 

2. Consider Testing Data if necessary.

At this point, your AI experiment has been published to both a testing environment and to a production environment (each with a different URL).

 

Cloud_AI_publish2

 

However and as a last, optional step, you may choose to include rules to apply for additional when using the testing environment.

To do this, click the Add rules button:

 

Cloud_AI_testingdata

 

In these settings, notice you may:

Add any number of rules, in which each rule may use any number of conditions.

According to conditions (e.g, the Number of times pregnant is greater or equal to 12), you may choose to directly influence the prediction.

Choose a default value for all predictions.

 

When done, close the publish window.

For a published experiment, you may at anytime go back to these Rules settings or to look up the connectivity information by clicking the View connectivity info button:

 

Cloud_AI_securityconn

 

Next steps

Note that detail about the HTTP request information, URL, access keys, and the connector is not relevant at this point to complete the publishing procedure.

Now that your experiment has been published, your Bizagi processes may rely on its service.

To learn how to do this, refer to Connecting Bizagi processes to use AI.