R-AIF#
A replication of R-AIF (arXiv:2409.14216, “a practical
implementation of active inference with continuous variables and learned preferences”), pinned to
the authors’ code and its logged runs as the oracle: their tiny RSSM world model and CRSPP
preference model as declared Systems, the assembled agent and its training step, self-revision,
their three metrics, and the resumable multi-seed Mountain Car harness. One module per concern.
Reference world models as declared Systems: R-AIF's |
|
The assembled R-AIF agent, and its training step. |
|
The CRSPP (prior-preference model): its recognition scan, its loss and its scoring row. |
|
Self-revision: the per-episode backward pass that produces ρ and the imitation mask. |
|
R-AIF's three metrics: ACR, SR and R-S, folded from recorded training episodes. |
|
R-AIF's episode columns, as task columns of the engine's |
|
Seeding M⁺: episode collection in their driver's row convention, and the bang-bang fallback controller for fixture-less runs. |
|
The resumable multi-seed Mountain Car run, and its report and benchmark modes. |