How xvado00/aiq works
AIQ evaluates a chosen agent by running episodes in programs drawn from a reference-machine sample, then estimates performance across strata. The README also documents sample generation, resumable runs, log-based analysis, and several agent families. The graph reflects the sampled runtime code and those documented capabilities; unsampled subsystem links are limited to what the README supports.
Main entry points: AIQ.py AIQ runner · agents Learning agents · BF.py BF machine · Agent.py Agent base
Components 17
AIQ evaluation
- AIQ runner
AIQ.py - Log restoration
AIQ_continue_from_log.py
Reference machines
- Machine base
refmachines/ReferenceMachine.py - BF machine
refmachines/BF.py
Agent implementations
- Agent base
agents/Agent.py - Learning agents
agents - Other agents
agents - Observation encoder
agents/utils/observation_encoder.py - Epsilon decay
agents/utils/epsilon_decay.py - MC-AIXI wrapper
agents/MC_AIXI.py
Sampling and analysis
- BF sampler
refmachines/BF_sampler.py - Program samples
- Run logs
- Log analysis
ComputeFromLog.py
Other components
- Evaluator
- mc-aixi executable
- AIQ estimate
Connections 18
- Evaluator → AIQ runner: configures
- AIQ runner → Agent base: constructs agent
- AIQ runner → BF machine: constructs environment
- AIQ runner → Program samples: reads programs
- AIQ runner → Run logs: writes results
- AIQ runner → AIQ estimate: estimates AIQ
- Learning agents → Agent base: extends base
- Learning agents → Observation encoder: encodes observations
- Learning agents → Epsilon decay: decays exploration
- Other agents → Agent base: agent interface
- BF machine → Machine base: implements
- BF sampler → BF machine: tests programs
- BF sampler → Program samples: generates samples
- AIQ runner → Log restoration: loads prior results
- Log restoration → Run logs: reads
- Log analysis → Run logs: reads
- Log analysis → AIQ estimate: reports statistics
- MC-AIXI wrapper → mc-aixi executable: invokes