How To Test Or Upgrade New Version Of ML Model?

How to Test or Upgrade New Version of ML Model From Pre-Existing Model Package?

Step 1: Upload a New Model Version

(These steps only apply if a model has already been chosen and uploaded previously to the AI Center Project. If no models have ever been used in this AI Center project, there will be no packages in the list as mentioned in the first bullet point below.)

In this step, make a new version of the desired model available to be used for training and/or deployment via an ML Skill. This step does not modify ML Skills that are currently deployed.

  1. In AI Center, navigate to ML Packages. Towards the bottom of this page, a list of packages that have been previously uploaded or created by training should be visible.
  2. Click on the ML Package for the model that is desired to be upgraded.
  3. After clicking on the ML Package, click the button in the top right corner that will say "Upload new version".
  4. On the new page that comes up, choose the desired package version that is to be tested: (See example below) **Important** If On-Prem, please validate that the version of the model that you are choosing is compatible with the version of AI Center installed. On the ML Packages page, when choosing the version number, there should be a warning at the very bottom of the page that lists what version of AI Center is required for use.
  1. Click Submit at the bottom of this page after choosing the desired Package Version.
  2. After clicking submit, chose the desired configurations for the package. (The screenshot below is only meant as an example. Configurations may vary.)
  1. After making the desired configuration changes, click Submit again.
  2. A new package version should now be listed under the specific ML package. (Note, this is not a deployed or trained version of the model. This is the base version of the model as provided by UiPath.)

Step 2: Training the New Version of the Model

Now that a new version of the model has been uploaded to the project for use, the model will need to be trained with specific datasets that have been carefully curated for training the model. This dataset can be the same as what was previously used for training an older version of the model.

  1. Click on Pipelines
  2. In the top left corner, click on the "Create new" button.
  3. Now configure the pipeline to train the new version of the model.
  1. Click Create
  2. Now, monitor the status of the Pipeline from the Pipelines page in the AI Center project. Note: The time it takes for the pipeline to complete will depend on the size of the dataset chosen, fields in the data, etc.
  3. Once the pipeline is successful, click on the pipeline. At the top of this page, the pipeline details will be displayed. In the details, a Generated version number should be visible. Example: Generated Version: (The last digit in the generated version is the training version produced.)

Step 3: Deploying a New Skill With the New Trained version of the Model

After the Pipeline completes successfully, the next step is to Create a new ML Skill. By choosing to create a New ML Skill vs Updating an existing Skill, you will not impact the DU processes that is currently using the older version in the old skill because the old skill will remain deployed while we create and test the new skill with the new version number.

  1. In the AI Center project, on the ML Skills page, click the "Create new" button at the top right of the page.
  2. Give the ML Skill a new name
  3. Choose the desired package
  4. Choose the new major version of the package that was previously made available back in Step 1
  5. Choose the package minor version number based on the last digit in the generated version number from the pipeline details. Example: If the Generated version in the pipeline details was The minor version chosen should be 1.
  6. Click Create
  7. After a few minutes, the Skill should show as available on the ML Skills page.

After all of the steps above have been completed, the new skill can be tested in RPA processes.

Step 4: Update the Older Existing Skill After Testing The New Skill

(Complete this step only after testing the new skill that was deployed in step 3 and the results seen are satisfactory.)

After completing testing for the new version of the model with the New ML Skill deployed, the older skill can be updated to the new version of the model by following these steps.

  1. Go to the ML Skills page in the AI Center Project
  2. Choose the older skill that has been using the old version in production
  3. On the ML Skills detail page, toward the bottom of the page,a list of available package versions should be listed that can be use to upgrade (or downgrade) the Skill.
  4. Select the new trained package version and click the update option to the far right of the package versions row. See the example below:
  1. After choosing the desired configurations (See example below. Note, if the ML Skill was made public, make sure this option is still enabled.), choose Confirm
  1. After some time, the ML skill should be updated and available for use again it should now be using the new trained model version.