causal-abstraction-eval 1.0.0__py3-none-any.whl

causal_abstraction/__init__.py

@@ -0,0 +1,177 @@
+"""
+Causal abstraction library
+
+A framework for validating causal explanations of complex systems.
+"""
+
+from importlib.metadata import PackageNotFoundError, version as _version
+
+try:
+    # Read from the installed distribution metadata (distribution name
+    # "causal_abstraction_eval"); pyproject.toml is the single source of truth.
+    __version__ = _version("causal_abstraction_eval")
+except PackageNotFoundError:  # running from an un-installed source checkout
+    __version__ = "0.0.0+unknown"
+
+# Analytical metrics
+from .analytical_metrics import (
+    AnalyticalMetric,
+    StructuralDeviationMetric,
+    CausalSensitivityIndexMetric,
+    MallowsCpMetric,
+    IBLagrangianMetric,
+    CIBLagrangianMetric,
+    MacroscopicInvarianceMetric,
+    ComplexityShiftMetric,
+    SobolSensitivityMetric,
+    IIAMetric,
+    BCCMetric,
+    ProbingMetric,
+    InfidelityMetric,
+    SymbionMetric,
+    DCCMetric,
+    RelationalFidelityMetric,
+)
+from .config import EvaluationConfig
+from .engine import EvaluationEngine
+from .experiment import ExperimentResults, AnalyticalResults
+# Metrics
+from .metrics import (
+    EvaluationMetric,
+    SubspaceCheckMetric,
+    L2Metric,
+    MSEMetric,
+    RMSEMetric,
+    NMSEMetric,
+    R2Metric,
+    JSDivergenceMetric,
+    KLDivergenceMetric,
+    PrecisionMetric,
+    MMDMetric,
+    TrajectoryMSEMetric,
+    DTWMetric,
+    TemporalAutocorrelationMetric,
+    SpectralMetric,
+    ConditionalIndependenceMetric,
+    VarianceDecompositionMetric,
+    get_metric,
+)
+from .models.high_level import CausalGraph
+from .models.low_level import LowLevelModel, NoisyLowLevelModel
+# NeuralModel is imported lazily (see __getattr__ below) so that a bare
+# `import causal_abstraction` does not eagerly load torch.
+from .paths import DiagramBuilder, CausalPath
+from .primitives import SystemState, AbstractVariable, ProbabilityDistribution, EmpiricalDistribution, UNMAPPED, \
+    DiscreteDistribution
+from .sampling import (
+    TopDownSampler, BottomUpSampler, InterventionSampler,
+    PairedSampler, NoisyMeasurementSampler, CombinedFaithfulnessSampler,
+    PHI_DUMMY_NAME,
+)
+from .schema import MicroVariableSchema, CoarseGrainingMap, MicroSelector, Variable
+from .spaces.base import (
+    Subspace, RectSubspace, SphereSubspace, UnionSubspace,
+    ComplementSubspace, FullSubspace, UniformSubspace,
+    GaussianSubspace
+)
+from .tasks import StandardTasks, EvaluationTask, TaskResults
+from .valuemap import ValueMap, ContinuousValueMap
+
+__all__ = [
+    # Version
+    "__version__",
+    # Core primitives
+    "SystemState",
+    "AbstractVariable",
+    "Variable",
+    "ProbabilityDistribution",
+    "EmpiricalDistribution",
+    "UNMAPPED",
+    "DiscreteDistribution",
+    # Schema & mapping
+    "MicroVariableSchema",
+    "CoarseGrainingMap",
+    "MicroSelector",
+    "ValueMap",
+    "ContinuousValueMap",
+    # Config & results
+    "EvaluationConfig",
+    "ExperimentResults",
+    "AnalyticalResults",
+    # Engine
+    "EvaluationEngine",
+    # Samplers
+    "TopDownSampler",
+    "BottomUpSampler",
+    "InterventionSampler",
+    "PairedSampler",
+    "NoisyMeasurementSampler",
+    "CombinedFaithfulnessSampler",
+    "PHI_DUMMY_NAME",
+    # Models
+    "LowLevelModel",
+    "NoisyLowLevelModel",
+    "CausalGraph",
+    "NeuralModel",
+    # Spaces
+    "Subspace",
+    "RectSubspace",
+    "SphereSubspace",
+    "UnionSubspace",
+    "ComplementSubspace",
+    "FullSubspace",
+    "UniformSubspace",
+    "GaussianSubspace",
+    # Paths
+    "DiagramBuilder",
+    "CausalPath",
+    # EvaluationMetric classes
+    "EvaluationMetric",
+    "SubspaceCheckMetric",
+    "L2Metric",
+    "MSEMetric",
+    "RMSEMetric",
+    "NMSEMetric",
+    "R2Metric",
+    "JSDivergenceMetric",
+    "KLDivergenceMetric",
+    "PrecisionMetric",
+    "MMDMetric",
+    "TrajectoryMSEMetric",
+    "DTWMetric",
+    "TemporalAutocorrelationMetric",
+    "SpectralMetric",
+    "ConditionalIndependenceMetric",
+    "VarianceDecompositionMetric",
+    "get_metric",
+    # AnalyticalMetric classes
+    "AnalyticalMetric",
+    "StructuralDeviationMetric",
+    "CausalSensitivityIndexMetric",
+    "MallowsCpMetric",
+    "IBLagrangianMetric",
+    "CIBLagrangianMetric",
+    "MacroscopicInvarianceMetric",
+    "ComplexityShiftMetric",
+    "SobolSensitivityMetric",
+    "IIAMetric",
+    "BCCMetric",
+    "DCCMetric",
+    "ProbingMetric",
+    "InfidelityMetric",
+    "SymbionMetric",
+    "RelationalFidelityMetric",
+    # Tasks
+    "StandardTasks",
+    "EvaluationTask",
+    "TaskResults",
+]
+
+
+def __getattr__(name):
+    # PEP 562 lazy attribute access: defer the torch import behind
+    # NeuralModel so a bare `import causal_abstraction` stays lightweight.
+    if name == "NeuralModel":
+        from .models.neural import NeuralModel
+        return NeuralModel
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")