😀MetaBox-v2.0.2 is released！！😀

😀In the month since the release of MetaBox-v2.0.1, we have received a lot of user feedback. Thank you for your support! 🎉We have made targeted adjustments for each feedback, and now we are officially pushing MetaBox-v2.0.2! 🥳🥳The following are some of the main corrections:

1. Document support Chinese translation, and the improvement of the classification of problem difficulty, etc.
2. Some problems in LSGO (confusion between func and compute, etc.) are fixed.
3. Support more types of config, such as custom tensorboard names, etc.
4. Fix some bugs in madac and vdn in MTO, and some bugs in WFG1 and WFG2
5. Correction of bias in LGA and fix of cpu random number setting
6. Correction of return_info dimension error
   ...

In addition, there are many details to be improved. We are very grateful for your support and hope to further improve MetaBox in the future!🎆

😀MetaBox-v2.0 is officially released！！😀

🥳 Based on MetaBox’s fully automatic train-test-log workflow, Unification, Efficiency and Flexible are the core concepts of MetaBox-v2.0 🥳

• Unification: MetaBox-v2.0 provides a unified interface for various MetaBBO (MetaBBO-RL、MetaBBO-SL、MetaBBO-ICL、MetaBBO-NE) and test suites for different types of problems. 18 optimization problem sets (synthetic + realistic), 1900+ problem instances and 36 baseline methods (traditional optimizers + up-to-date MetaBBOs) are reproduced within MetaBox-v2 to assist various research ideas and comprehensive comparison.

• Efficiency: Parallelization enables MetaBox-v2.0 to train and test efficiently. As the first development exemplar to enable parallel training for MetaBBO, MetaBox-v2.0 supports training speeds up to 10 times faster than MetaBox-v1.0. Various Ray-based parallel modes allow users to fully utilize their computing resources to accelerate testing.

• Flexible: Taking the provided Evaluation Metrics as an example, by recording Metadata, users can easily and flexibly visualize these evaluation metrics across different MetaBBO methods for various types of problems, without being constrained by specific problems or approaches. There are even more flexible usages, including but not limited to leveraging existing resources from other platforms (such as EvoX) to extend MetaBox’s current methods and problem sets. For more information, please refer to the Gallery!

🚀🎉Based on the above three features, users only need to specify the target problem and the MetaBBO algorithm to be used. MetaBox will then provide fully automated training and testing, and ultimately allow flexible visualization of results using the recorded Metadata.🎉🚀