assume-framework/assume: v0.6.0 - (18th March 2026)

Improvements: Deterministic behavior with seed setting: Simulations are now deterministic by default for improved reproducibility. This can be controlled via a seed setting in config.yaml files, therefore it only applies for scenarios loaded via load_scenario_folder. Note that complete determinism is not guaranteed for all hardware and software configurations, especially with PyTorch-based learning strategies. It may also decrease performance of reinforcement learning due to disabled non-deterministic optimizations. seed not set in top-level of config: Sets the seed to a fixed default value (42) for deterministic behavior. seed: <int>: Sets the seed for all random number generators to provided <int>. seed: null: Disables seed setting, allowing for non-deterministic behavior as before. May improve performance of reinforcement learning. Delete environment.yaml: The environment.yaml file has been removed from the repository to simplify maintenance and was completely redundant with the pyproject.toml. Users can as before create their own environment using the provided pip installation instructions, which allows for more flexibility and easier updates. Add validation for simulation setup: Added checks to validate the simulation setup for common issues, such as missing bidding strategies or inconsistent market configurations. Warnings are issued to inform users of potential problems that could affect simulation results. Added reward calculation for unit operators: Unit operators have now the opportunity to calculate rewards based on the returned orderbooks for their own purposes. This enables learning strategies on unit operator level / portfolio learning strategies. Structured Validation Error: Introduces the new ValidationError to represent a failing validation. Since it derives from the base ValidationError, all existing error handling remains compatible, but users can now also catch this specific error type to Add support for Pandas 3handle validation errors separately if desired. Add support for Python 3.14 Bug Fixes: Fix buffer and update order: Fixed the order of buffer writing and policy updating in the learning role to ensure that both have the exact same order, which is necessary so that during updates the correct data is used. Thisbug will have compormised learning with very heterogeneous units after the last release. Fix data loss in RL learning role: Fixed data loss in RL learning role by implementing atomic swap with carry-over for incomplete timesteps in cache Update notebooks to always install latest repo version from Google Colab: This ensures that the latest version is always used