AutoML v.0.4.0: Redefining the Configuration of ML Workflows

Claudio Bruderer

Written by:
Claudio Bruderer (Product Manager at Modulos)

Our latest version of Modulos AutoML is out: v.0.4.0. With this version, we are giving you a completely redefined ML workflow creation process. Not only does it include new features and increased transparency into the process, it also makes the workflow creation straightforward and easy. Among other new features, we have also extended the README distributed with every AutoML Solution to give you more insights into your trained model.

Upgraded Workflow Creation


By loading the video, you agree to YouTube’s privacy policy.
Learn more

Load video

The new workflow creation process in action.

Sleek and Clean UI Design

We are convinced that it is crucial for an AutoML platform to be easy to use and transparent. We have adhered to this principle when redesigning the entire creation process (see video above). The fresh look and feel allows for a seamless navigation back and forth between configuration steps while letting you always know where you are.

Review & Launch to Skip Optional Steps

Selection of the objective to be optimized in the workflow creation process.

We distinguish in Modulos AutoML between mandatory and optional steps in the workflow creation. The mandatory steps require your domain expertise to configure a workflow suiting your use case. Hence, they include the choice of input data, what feature needs to be predicted (label), and with which objective the prediction is optimized for. The optional steps describe AutoML settings relevant for the model search for which AutoML will provide strong defaults. These steps include which model and feature engineering methods are considered in the search and what search algorithm for the high dimensional space is employed.

You can now skip optional steps with “Review & Launch” (see image above) making the creation process smoother and faster for you. In that case the platform will use the default settings suited to your problem. Of course, you can still complete all steps in the workflow creation and adjust the optional steps’ default settings.

List of Applicable Models Always at Your Fingertips

Pop-up containing all the applicable feature engineering methods and ML models accessible at any step of the workflow creation.

Another one of the core design principles for the new workflow creation is to make the impact of your choices transparent. At every step, you can review the list of applicable feature engineering methods and ML models. In addition, Modulos AutoML gives you further information on all the modules and discloses at which step their status changed.

Example Predictions in the Solution README

List of examples for good predictions included with the Solution README.

With every trained Solution, you get a README file which contains information on the setup, the requirements, and the deployment. It also gives you insights into the performance of the model on validation data. Besides showing plots, the README now also contains examples for comparatively good and bad predictions of the model. This allows you as a domain expert to look into the model, understand it, and potentially even refine it in future workflows.

Other Improvements and Bug Fixes

We additionally have made a range of other improvements and fixes to the latest version of Modulos AutoML, which include:

  • Extended the dataset upload confirmation step to display columns removed during the validation process due to zero variance. These columns do not contain any relevant information for the model and are hence not included in any Solution.
  • Refined the default sorting order of the tables containing all workflows, datasets, and candidate ML solutions in training to have the most recent ones on top. The default sorting order is now also highlighted.
  • Augmented the check for incomplete data at the dataset upload step to also include infinite values.
  • Introduced a configuration file including your licensing information. This will simplify your interactions with the platform and make sure that you are always starting AutoML with the licensed resources.
  • Increased the update frequency of the computation times shown in the UI.
  • Extended the support for non-latin characters (UTF-8) in confusion matrix plots in the Solution README.