An implementation of Non-Axiomatic Causal Explorer in Julia

• [ x ] Make experiment no. 2 run.
  • [ x ] hypothesize and other unimplemented functions need to reflect the original logic flow diagram. Multiple functionalities condensed in the current functions need to be split out into those functions.
    • [ x ] hypothesize
    • [ x ] new_hypotheses
    • [ x ] verify_hypothesis
    • [ x ] max_truth_exp
    • [ x ] best_hypothesis
    • [ x ] highest_reward
    • [ x ] weakest_hypothesis
    • [ x ] oldest_observed
• [ x ] Remove old code (everything currently outside NACE.jl), preserving useful parts (mostly done).
• [ x ] Lastly, refactor NACE.jl into multiple files to have a clean library structure.
• Next on the agenda, make the agent actually learn something in the experiment no. 2.
  • [ x ] Score updates currently don't do anything, figure out why.
  • [ x ] Verify that the rule gen logic makes sense.
  • Verify the rest of the code.

I recommend using uv to make a Python virtual env (uv sync, uv sync --extra dev, etc.) Run the REPL with uv run julia --project, before installing dependencies (] instantiate) make sure to execute ENV["PYTHON"] = Sys.which("python") -- that will set PyCall to use your environment's Python. You don't have to run julia from the activated environment after that.

Alternatively, you can run uv run julia --project -e 'ENV["PYTHON"] = Sys.which("python"); using Pkg; Pkg.instantiate()' to do all of the above in one line.

In the REPL, using NACE should be enough, some things may be not exported (the API is not yet stable, needless to say) -- those you have to access under the packages namespace (NACE). There is a comment in NACE.jl that shows how to spawn a Gym environment in Julia.

There are experiment files that you can run with julia --project experiments/<exp name>.jl.

Under the hood, environments are provided by the Farama Foundation's Minigrid library, which relies on the Gymnasium package, also maintained by the Foundation.