First of all, thank you for using SigOpt; we have had many exciting opportunities arise since our November 2020 acquisition by Intel and we are looking forward to serving both users who are new to the platform and those who have been with us since we started on this journey many years ago. Thanks to the support of Intel, we have been able to expand our vision of empowering the world’s experts through our earlier launched Free Usage Tier. Tell your friends!
Second, we want to update everyone regarding some recent product decisions – the result of these decisions should require no change in behavior for most of you.
In light of feedback we have received, we will NOT be deprecating our classic SigOpt client or its functionalities. Moving forward, our documentation will refer to our two main API interaction models as the Core Module and AI Module. Everybody is free to use either Module as it fits with your workflow. However, a given Experiment must either be a Core Experiment (Suggestions/Observations) or an AI Experiment (Runs).
In the past 12 months, if you have used our Python client and manually created a Connection object, you have received a warning about an upcoming loss of support for that behavior. In the future, feel free to disregard the warning – we will continue to support the usage of these objects. If you have never seen that message, you do not have to change your behavior and can continue to use SigOpt as you have been.
What makes these modules different?
While both modules support the same workflow of passing parameter and metric information to SigOpt and using an iterative loop to arrive at an optimal solution, the difference lies in how you will pass us these objects during the experiment create process, and how data will appear on the SigOpt dashboard for analytics and experiment tracking purposes.
- Core Module – You will use the Suggestion/Observation objects (through Python, Java, Bash, Matlab, etc) to interact directly with the SigOpt API. The use of tags will replace the project organization functionality.
- AI Module – You interact with SigOpt using Runs objects (through the Python client or CLI). The project functionality will be preserved and you’ll be able to analyze your runs within a project as originally intended.
For more information on which module fits best with your needs, please refer to our documentation page. For examples of how to create experiments using each module, check out their respective quick start tutorials here: Core, AI.
Which module is right for you?
Both the Core and AI modules share the main concepts of a SigOpt optimization loop, but it is still important to think about which one better fits your needs before launching into a large series of experiments. While you’re free to try them both out, keep in mind that Core and AI modules will be organized separately within the SigOpt web dashboard, will not share analysis pages, and may have different compatibilities with advanced optimization features.
The Core module API was designed for general usage of our optimization features. It’s lightweight and offers more flexibility, with the ability to create, delete, and update almost all available API objects throughout the experimentation process. This module is supported in Python, Java, and Bash – in particular, Bash users will be able to access the API without needing to install a client library.
The AI module API was designed with machine learning and AI applications in mind. It addresses the need for experiment and model artifact tracking and organization, whether it be learning curves, metadata, or other training run data. This module is only available in Python and also includes lightweight integrations with ML libraries like XGBoost and Hyperopt.
Let us know what you think!
Both modules and their respective Advanced Features are available for all SigOpt users. If you don’t already have a SigOpt account, you can sign up here for free!
Check out our documentation here, and post your questions on our community forum to get help! Thank you for your continued use of SigOpt, and we look forward to many more years of intelligent experimentation.