Skip to content

01_single_agent.md

Latest commit

 

History

History
33 lines (26 loc) · 945 Bytes

01_single_agent.md

File metadata and controls

33 lines (26 loc) · 945 Bytes
Raw

Single-agent model

This tutorial builds one validated generative model and inspects its transition semantics. It assumes the package is installed from the repository root:

uv sync --locked --extra dev
from blockference import ActiveGridference, make_grid

agent = ActiveGridference(
    make_grid(3),
    planning_length=2,
    env_state=(0, 0),
    affordances=["UP", "RIGHT", "STAY"],
)
agent.get_C((2, 2))
agent.get_D((0, 0))

print(agent.B.shape)  # (9, 9, 3)
print(agent.E)       # ['UP', 'RIGHT', 'STAY']
print(agent.D.sum()) # 1.0

B has one column-stochastic transition matrix per configured affordance. A is the identity likelihood for the fully observed grid, while C and D are one-hot vectors after get_C and get_D.

What to try next: compare the single-agent inference loop in theory.md with the complete persisted run in 03_cadcad_pipeline.md.