CounterfactualSearchScenario
Agents scenarios · Implementation: mlsysim/engine/scenario_definitions/agents.py.
Construct this scenario with keyword inputs. The fields below show the types and defaults declared by this class.
Replay paths that blame for a failure must search: a whole horizon, or one checked window.
Notes
Equation: Leaf paths number branching**depth; serial replay time is checked-window path count times time per path.
Assumptions: Every step has the same branching factor and each full path is replayed separately.
Excludes: Pruning, shared prefixes, parallel replay, and proof that a replay identifies causal responsibility.
References
Derivation: Leaf paths number branching**depth; serial replay time is checked-window path count times time per path. This is declared scenario accounting rather than a fitted empirical law.
Examples
>>> from mlsysim import Scenarios
>>> scenario = Scenarios.CheckedWindowBlameSearch
>>> value = scenario.full_paths
>>> round(float(value), 3)
1.2676506002282294e+30Attributes
| Name | Description |
|---|---|
| branching | |
| full_exponent | Base-ten logarithm of full leaf-path count, dimensionless. |
| full_paths | Leaf-path count for fixed branching at every step of the full horizon. |
| horizon | |
| replay_time | |
| window | |
| window_paths | Leaf-path count within the shorter checked replay window. |
| window_search_time | Serial minutes to replay every checked-window path at constant per-path duration. |
branching
Type: int.
Declared default: required.
horizon
Type: int.
Declared default: required.
window
Type: int.
Declared default: required.
replay_time
Type: Quantity.
Declared default: required.
full_paths
engine.scenario_types.CounterfactualSearchScenario.full_paths
Leaf-path count for fixed branching at every step of the full horizon.
Equation: branching ** horizon.
full_exponent
engine.scenario_types.CounterfactualSearchScenario.full_exponent
Base-ten logarithm of full leaf-path count, dimensionless.
Equation: _math.log10(full_paths).
window_paths
engine.scenario_types.CounterfactualSearchScenario.window_paths
Leaf-path count within the shorter checked replay window.
Equation: branching ** window.
window_search_time
engine.scenario_types.CounterfactualSearchScenario.window_search_time
Serial minutes to replay every checked-window path at constant per-path duration.
Equation: (window_paths * replay_time).to('minute').