* checkpoint
* bugfixes and added dynamic production flow calculator
* added lua implementation of the achievement tracking
* cleanup
* movedprod stats to a new file
* created a test
* removed unnecessary file
* tidied up
* removed change
* Add blueprint based initial states. Add diversity sampling.
* Fixed namespace reinitialising bug
* Give reward for harvesting. Increase complexity exponent in calculating reward
* Implemented planning MCTS with the parallelisation infrastructure.
* added new prompts
* first version of planning executor
* checkpoint
* first draft
* added plan parent id
* removed changes
* fixed a small error
* merged with newest
* fixedsome bugs
* added meta and outputs
* added mew logging
* updated prompts
* improvements and bugfixes
* entrypoint for planning
* removed testing code
* removed testing code
* made additions to get them working with generation paramaters
* revoked some changes
* removed file
* Added a formatter class to simplify long conversational chains before going to the LLM to save tokens / context.
* Instance logging using Rich. More data and some bug fixes.
* Partial implementation of a chunked MCTS searcher
* Retain comments and whitespace in chunk parser
* Add inventory and entities to the observation state if missing via modifying agent generated code.
* added the plan sampler
* Ensure holdout runs for chunked mcts. Ensure all 3 servers process tasks.'
* added example how to use
* added correct stop sequence parsing
* added planning
* added docstring
* Fixed error serialization bug. Added objective seeding.
* Add recursive crafting support.
* Add refined system policy. Fixed pipe connect when invalid connection points or jagged path
* Removing connection points after setting them in serialized
* Add support for running MCTS in full parallelization, with instance grouping parameters
* Bug fixes, better logging.
* Holdout scores are retrieved after each chunk is processed.
* Sleep action is now subdivided by the game speed.
* Better holdout
---------
Co-authored-by: Jack Hopkins <jackhopkins@192.168.1.116>
Co-authored-by: Jack Hopkins <jackhopkins@Jacks-MacBook-Pro.local>
Co-authored-by: Mart <mart@paperplane.ai>
* Added a formatter class to simplify long conversational chains before going to the LLM to save tokens / context.
* Instance logging using Rich. More data and some bug fixes.
* Partial implementation of a chunked MCTS searcher
* Retain comments and whitespace in chunk parser
* Add inventory and entities to the observation state if missing via modifying agent generated code.
* added the plan sampler
* Ensure holdout runs for chunked mcts. Ensure all 3 servers process tasks.'
* added example how to use
* added correct stop sequence parsing
* added planning
* added docstring
* Fixed error serialization bug. Added objective seeding.
* Add recursive crafting support.
---------
Co-authored-by: Jack Hopkins <jackhopkins@192.168.1.116>
Co-authored-by: Jack Hopkins <jackhopkins@Jacks-MacBook-Pro.local>
Co-authored-by: Mart <mart@paperplane.ai>
