Crate deep_causality_discovery 

Causal Discovery Language (CDL)

CDL hosts two discovery algorithms — SURD and BRCD — as two compile-time-isolated typestate sub-pipelines that converge on a shared analyze/finalize tail. A sub-pipeline is seeded by CdlBuilder::build_surd / CdlBuilder::build_brcd from a run config built by CdlConfigBuilder.

Compile-time sub-pipeline isolation

Each algorithm’s stage methods exist only on its own states, so crossing the sub-pipelines does not compile. Calling a BRCD stage on a SURD state:

use deep_causality_discovery::{CDL, SurdData};
fn cross(c: CDL<SurdData<f64>>) {
    let _ = c.brcd_discover(); // error[E0599]: no method named `brcd_discover`
}

Calling a SURD analyze on a BRCD state:

use deep_causality_discovery::{CDL, BrcdLoaded};
fn cross(c: CDL<BrcdLoaded<f64>>) {
    let _ = c.surd_analyze(); // error[E0599]: no method named `surd_analyze`
}

Feature selection is unavailable to the BRCD sub-pipeline:

use deep_causality_discovery::{CDL, BrcdConfigured};
fn cross(c: CDL<BrcdConfigured<f64>>) {
    let _ = c.feature_select(); // error[E0599]: no method named `feature_select`
}

Modules

brcd_analyze_config

brcd_loader_config

causal_discovery_config

data_csv_config

data_loader_config

data_parquet_config

data_preprocess_config

feature_selector_config

mrmr_config

surd_analyze_config

surd_config

surd_loader_config

Structs

BrcdAnalyzeConfig

Configuration for the BRCD analysis phase of the CDL pipeline.

BrcdConfig

Configuration for a BRCD run.

BrcdConfigured

BRCD entry state, carrying the run config from CdlBuilder::build_brcd.

BrcdError

The error type returned by every fallible BRCD operation.

BrcdInput

The fully-prepared input for a BRCD discovery run.

BrcdLoaded

BRCD state after the input bundle has been loaded.

BrcdLoaderConfig

The fully-specified configuration for a BRCD CDL run.

BrcdResult

A structured result for a BRCD root-cause discovery run.

BrcdResultAnalyzer

An implementation of ProcessResultAnalyzer for BrcdResult.

BrcdResults

BRCD state after the BRCD algorithm has run.

CDL

The core builder for the Causal Discovery Language (CDL) pipeline.

CdlBuilder

Entry point for constructing and running a CDL pipeline.

CdlConfigBuilder

Entry point for the staged CDL config builders.

CdlEffect

CdlEffectWitness

A witness type that “fixes” the Error (E) and Warning Log (WLog) types for CdlEffect, implementing the HKT and HKT3 traits.

CdlReport

Aggregates all significant findings from a CDL pipeline execution.

CdlWarningLog

ConsoleFormatter

A concrete implementation of ProcessResultFormatter that formats an analysis for display in the console.

CsvConfig

Configuration for loading data from a CSV file.

CsvDataLoader

A concrete implementation of ProcessDataLoader for reading data from CSV files.

DataDiscretizer

A concrete implementation of DataPreprocessor that discretizes continuous data into bins.

MissingValueImputer

A pre-processor for handling missing numerical data in a CausalTensor.

MixedGraph

A mixed graph over num_vertices nodes with typed edge endpoints.

MrmrConfig

Configuration for the MRMR (Max-Relevance, Min-Redundancy) feature selector.

MrmrFeatureSelector

A concrete implementation of the FeatureSelector trait that uses the MRMR algorithm.

OptionNoneDataCleaner

ParquetConfig

Configuration for loading data from a Parquet file.

ParquetDataLoader

A concrete implementation of ProcessDataLoader for reading data from Parquet files.

PreprocessConfig

Configuration for a data discretization pre-processing step.

ProcessAnalysis

A wrapper struct holding the results of an analysis as a vector of strings.

ProcessFormattedResult

A wrapper struct holding the final, formatted output of the CDL pipeline.

SurdAnalyzeConfig

Configuration for the SURD analysis phase of the CDL pipeline.

SurdCausalDiscovery

A concrete implementation of the CausalDiscovery trait using the SURD algorithm.

SurdCleaned

SURD state after data has been cleaned (masked to Option<T>).

SurdConfig

Configuration for the SURD (Synergistic, Unique, Redundant Decomposition) algorithm.

SurdConfigured

SURD entry state, carrying the run config from CdlBuilder::build_surd.

SurdData

SURD state after data has been loaded.

SurdFeatures

SURD state after feature selection has been applied.

SurdLoaderConfig

The fully-specified configuration for a SURD CDL run.

SurdResultAnalyzer

An implementation of ProcessResultAnalyzer for SurdResult.

SurdResults

SURD state after the SURD algorithm has run.

WithAnalysis

State after the raw discovery result has been analyzed. Both sub-pipelines converge here, carrying the polymorphic CdlDiscoveryOutcome.

Enums

AnalyzeError

BinningStrategy

BrcdLoadError

Failure cases for [crate::BrcdDataLoader] building a crate::BrcdInput.

CausalDiscoveryConfig

An enum representing the configuration for a specific causal discovery algorithm.

CausalDiscoveryError

CdlDiscoveryOutcome

The algorithm-specific result of a CDL discovery run.

CdlError

CdlWarning

ColumnSelector

CpdagError

Failure cases for reading or writing a CPDAG CSV file.

DataCleaningError

DataLoaderConfig

An enum representing the configuration for a specific data loader.

DataLoadingError

FamilyKind

The likelihood family BRCD scores each node with.

FeatureSelectError

FeatureSelectorConfig

An enum representing the configuration for a specific feature selection algorithm.

FinalizeError

Mark

The mark at one endpoint of an edge, in the typed-endpoint calculus shared by CPDAGs, MAGs, and PAGs.

MaxOrder

Defines the maximum order of interactions to consider in the SURD analysis.

PreprocessError

Traits

CausalDiscovery

Defines the contract for causal discovery algorithms.

DataCleaner

DataLoader

Defines the contract for loading data from a source into a CausalTensor.

DataPreprocessor

Defines the contract for data pre-processing steps.

FeatureSelector

Defines the contract for feature selection algorithms.

Precision

Precision bound for the CDL compute pipeline.

ProcessResultAnalyzer

Defines the contract for analyzing the raw results of a causal discovery algorithm.

ProcessResultFormatter

Defines the contract for formatting an analysis into a final output string.

Functions

brcd_run

Runs BRCD on the normal/anomalous datasets and returns the candidate root-cause sets ranked by descending posterior.

load_cpdag_csv

Reads a CPDAG CSV file from path into a unit-payload MixedGraph.

mrmr_features_selector

Selects features using the mRMR (Maximum Relevance, Minimum Redundancy) algorithm.

save_cpdag_csv

Writes graph to path as a CPDAG CSV file.

surd_states_cdl

Decomposes mutual information into its Synergistic, Unique, and Redundant components for each state of the target variable, based on the SURD-states algorithm, specifically designed to handle input tensors with Option<T> values.