Skip to content

CausalIQ Analysis Trace

This module implements detailed tracing of structure learning processes, allowing researchers to record, analyze, and compare the step-by-step evolution of causal graphs during algorithm execution.

⚠️ NOTE: these functionality will be superseded by a more open format based on csv and GraphML standards.

Core Classes

Trace 

Trace(context: Optional[Dict[str, Any]] = None)

Class encapsulating detailed structure learning trace.

Parameters:

  • context

    (dict, default: None ) –

    Description of learning context.

Attributes:

  • context (dict) –

    Learning context.

  • trace (list) –

    Iteration by iteration structure learning trace.

  • start (float) –

    Time at which tracing started.

  • result (SDG) –

    Learnt graph.

  • treestats (TreeStats) –

    Statistics from a tree search.

Raises:

  • TypeError –

    If arguments have invalid types.

  • ValueError –

    If invalid context fields provided.

Methods:

  • __eq__ –

    Test if other Trace is identical to this one.

  • __str__ –

    Return details of Trace in human-readable printable format.

  • add –

    Add an entry to the structure learning trace.

  • context_string –

    Return a trace context as a human readable string.

  • diffs_from –

    Find differences of trace from reference trace.

  • get –

    Return the trace information.

  • read –

    Read set of Traces matching partial_id from serialised file.

  • rename –

    Rename nodes in trace in place according to name map.

  • save –

    Save the trace to a composite serialised (pickle) file.

  • set_result –

    Set the result of the learning GraphAction.

  • set_treestats –

    Set the statistics of a tree learning GraphAction.

  • update_scores –

    Update score in all traces of a series.

__eq__ 

__eq__(other: Any) -> bool

Test if other Trace is identical to this one.

Parameters:

  • other
    (Trace) –

    Trace to compare with self.

Returns:

  • bool ( bool ) –

    True if other is identical to self.

__str__ 

__str__() -> str

Return details of Trace in human-readable printable format.

Returns:

  • str ( str ) –

    Trace in printable form.

add 

add(activity: GraphAction, details: Dict[GraphActionDetail, Any]) -> Trace

Add an entry to the structure learning trace.

Parameters:

  • activity
    (GraphAction) –

    Action e.g. initialisation, add arc.

  • details
    (dict) –

    Supplementary details relevant to activity.

Returns:

  • Trace ( Trace ) –

    Returns trace after entry added.

Raises:

  • TypeError –

    If arguments have invalid types.

context_string classmethod 

context_string(context: Dict[str, Any], start: float) -> str

Return a trace context as a human readable string.

Parameters:

  • context
    (dict) –

    Individual context information.

  • start
    (float) –

    Start time information.

Returns:

  • str ( str ) –

    Context information in readable form.

diffs_from 

diffs_from(
    ref: Trace, strict: bool = True
) -> Optional[Tuple[Dict[Any, Any], List[int], str]]

Find differences of trace from reference trace.

Parameters:

  • ref
    (Trace) –

    Reference trace to compare this one to.

  • strict
    (bool, default: True ) –

    Whether floats are tested to be strictly the same or just reasonably similar. Defaults to True.

Returns:

  • Optional[Tuple[Dict[Any, Any], List[int], str]] –

    tuple or None: (major differences, minor differences, textual summary) or None if identical.

Raises:

  • TypeError –

    If ref is not of type Trace.

  • ValueError –

    If either trace is invalid.

get 

get() -> DataFrame

Return the trace information.

Returns:

  • DataFrame ( DataFrame ) –

    Trace as Pandas data frame.

read classmethod 

read(partial_id: str, root_dir: str) -> Optional[Dict[str, Trace]]

Read set of Traces matching partial_id from serialised file.

Parameters:

  • partial_id
    (str) –

    Partial_id of Trace.

  • root_dir
    (str) –

    Root directory holding trace files.

Returns:

  • Optional[Dict[str, Trace]] –

    dict or None: {key: Trace} of traces matching partial id.

Raises:

  • TypeError –

    If arguments are not strings.

  • FileNotFoundError –

    If root_dir doesn't exist.

  • ValueError –

    If partial_id is entry or serialised file is not a dictionary of traces.

rename 

rename(name_map: Dict[str, str]) -> None

Rename nodes in trace in place according to name map.

Parameters:

  • name_map
    (dict) –

    Name mapping {name: new name}.

Raises:

  • TypeError –

    If bad arg type.

save 

save(root_dir: str) -> None

Save the trace to a composite serialised (pickle) file.

Parameters:

  • root_dir
    (str) –

    Root directory under which pickle files saved.

Raises:

  • TypeError –

    If bad argument types.

  • ValueError –

    If no id defined for trace.

  • FileNotFoundError –

    If root_dir does not exist.

set_result 

set_result(result: SDG) -> Trace

Set the result of the learning GraphAction.

Parameters:

  • result
    (SDG) –

    Graph result from learning activity.

Returns:

  • Trace ( Trace ) –

    Current Trace to support chaining.

Raises:

  • TypeError –

    If result argument is not a SDG.

set_treestats 

set_treestats(treestats: Any) -> Trace

Set the statistics of a tree learning GraphAction.

Parameters:

  • treestats
    (TreeStats) –

    Statistics from tree learning activity.

Returns:

  • Trace ( Trace ) –

    Current Trace to support chaining.

Raises:

  • TypeError –

    If treestats argument is incorrect type.

update_scores classmethod 

update_scores(
    series: str,
    networks: List[str],
    score: str,
    root_dir: str,
    save: bool = False,
    test: bool = False,
) -> Dict[Tuple[str, str], Tuple[Optional[float], float]]

Update score in all traces of a series.

Parameters:

  • series
    (str) –

    Series to update traces for.

  • networks
    (list) –

    List of networks to update.

  • score
    (str) –

    Score to update e.g. 'bic', 'loglik'.

  • root_dir
    (str) –

    Root directory holding trace files.

  • save
    (bool, default: False ) –

    Whether to save updated scores in trace file. Defaults to False.

  • test
    (bool, default: False ) –

    Whether score should be evaluated on test data. Defaults to False.

Raises:

  • ValueError –

    If bad arg values.

Main class for recording and managing structure learning traces with detailed context, iteration-by-iteration recording, and comparison capabilities.

DiffType

Enumeration defining the different types of differences that can be detected when comparing traces.

CompatibilityUnpickler

Custom unpickler that handles module path changes for backward compatibility.

Maps specific classes that have been moved between modules.

Methods:

  • find_class –

    Override find_class to handle module path changes.

find_class 

find_class(module: str, name: str) -> Any

Override find_class to handle module path changes.

Parameters:

  • module
    (str) –

    Original module name from pickle.

  • name
    (str) –

    Class name.

Returns:

  • Any –

    The class object from the new location.

Custom unpickler for handling backward compatibility when loading trace files with module path changes.

Utility Functions

load_with_compatibility 

load_with_compatibility(
    file_handle: Any, compression: str = "gzip", **kwargs: Any
) -> Any

Load pickled data with module compatibility handling.

Parameters:

  • file_handle

    (Any) –

    File handle to read from.

  • compression

    (str, default: 'gzip' ) –

    Compression type.

Returns:

  • Any –

    Unpickled object.

Load trace files with backward compatibility support for older module structures.

Overview

Structure Learning Tracing

The Trace class provides comprehensive functionality for recording and analyzing causal graph structure learning:

Recording Capabilities: - Context Management: Store learning algorithm parameters, data characteristics, and environment information - Iteration Tracking: Record each step of the learning process with detailed action information - Result Storage: Capture final learned graphs and search statistics - Flexible Storage: Support for compressed file formats and cross-platform compatibility

Analysis Features: - Trace Comparison: Compare two traces to identify differences in learning paths - Score Updates: Retroactively update or recalculate scores using different metrics - Variable Renaming: Handle variable name changes across different datasets - Statistical Summaries: Generate summaries of learning algorithm behavior

Key Methods

Creation and Management: - __init__(): Initialize a new trace with optional context information - add(): Record a new iteration with action details and graph state - read(): Load trace data from files with compatibility support - save(): Save trace data to compressed files

Analysis and Comparison: - diffs_from(): Compare this trace with another to identify differences - update_scores(): Recalculate scores using different scoring functions - get(): Export trace data as a pandas DataFrame for analysis - rename(): Apply variable name mappings throughout the trace

Results and Context: - set_result(): Set the final learned graph - set_treestats(): Add tree search statistics - context_string(): Generate human-readable context descriptions

Trace Differences

The tracing system can identify several types of differences between learning runs:

  • MISSING/EXTRA: Iterations present in one trace but not another
  • ACTION: Different actions taken at corresponding iterations
  • ARC: Different arcs modified in corresponding actions
  • SCORE: Different scores recorded for the same actions
  • DETAILS: Differences in recorded action details

File Compatibility

The module includes robust backward compatibility features:

  • Module Migration: Handles changes in module structure over time
  • Class Mapping: Automatically maps old class locations to new ones
  • Version Tracking: Records software versions for reproducibility
  • Compressed Storage: Efficient storage using gzip compression

Integration with Graph Actions

Traces work seamlessly with the graph action enumerations to provide detailed records of:

  • Which specific arcs were added, deleted, or reversed
  • Score changes resulting from each modification
  • Alternative actions that were considered but not taken
  • Statistical constraints and prior knowledge influences
  • Debugging information for algorithm development

This comprehensive tracing capability is essential for:

  • Algorithm development and debugging
  • Reproducible research in causal discovery
  • Performance analysis and optimization
  • Comparative studies of different learning approaches
  • Educational demonstrations of structure learning behavior