Class swarmauri_evaluator_subprocess.SubprocessEvaluator.SubprocessEvaluator

swarmauri_evaluator_subprocess.SubprocessEvaluator.SubprocessEvaluator

Bases: EvaluatorBase

Evaluator that runs programs in isolated subprocesses and measures their performance.

This evaluator executes programs in sandboxed subprocesses, capturing stdout, stderr, exit code, and runtime metrics. It provides security through resource limits and timeout constraints to prevent malicious or poorly written code from affecting the host system.

type class-attribute instance-attribute

type = 'SubprocessEvaluator'

timeout class-attribute instance-attribute

timeout = Field(
    default=30.0,
    description="Maximum execution time in seconds",
)

max_memory_mb class-attribute instance-attribute

max_memory_mb = Field(
    default=512, description="Maximum memory usage in MB"
)

max_processes class-attribute instance-attribute

max_processes = Field(
    default=64, description="Maximum number of processes"
)

max_file_size_mb class-attribute instance-attribute

max_file_size_mb = Field(
    default=10, description="Maximum file size in MB"
)

working_dir class-attribute instance-attribute

working_dir = Field(
    default=None,
    description="Working directory for execution",
)

env_vars class-attribute instance-attribute

env_vars = Field(
    default_factory=dict,
    description="Environment variables for the subprocess",
)

success_exit_codes class-attribute instance-attribute

success_exit_codes = Field(
    default=[0],
    description="Exit codes considered successful",
)

score_on_timeout class-attribute instance-attribute

score_on_timeout = Field(
    default=0.0, description="Score to assign on timeout"
)

score_on_error class-attribute instance-attribute

score_on_error = Field(
    default=0.0,
    description="Score to assign on execution error",
)

model_config class-attribute instance-attribute

model_config = ConfigDict(
    extra="allow", arbitrary_types_allowed=True
)

id class-attribute instance-attribute

id = Field(default_factory=generate_id)

members class-attribute instance-attribute

members = None

owners class-attribute instance-attribute

owners = None

host class-attribute instance-attribute

host = None

default_logger class-attribute

default_logger = None

logger class-attribute instance-attribute

logger = None

name class-attribute instance-attribute

name = None

resource class-attribute instance-attribute

resource = Field(default=EVALUATOR.value)

version class-attribute instance-attribute

version = '0.1.0'

aggregate_scores

aggregate_scores(scores, metadata_list)

Aggregate multiple evaluation scores and their metadata.

This implementation extends the base aggregation with subprocess-specific metrics.

PARAMETER	DESCRIPTION
scores	List of individual scores to aggregate TYPE: List[float]
metadata_list	List of metadata dictionaries corresponding to each score TYPE: List[Dict[str, Any]]

RETURNS	DESCRIPTION
Tuple[float, Dict[str, Any]]	A tuple containing: - float: The aggregated score - Dict[str, Any]: Aggregated metadata with execution statistics

Source code in swarmauri_evaluator_subprocess/SubprocessEvaluator.py

def aggregate_scores(
    self, scores: List[float], metadata_list: List[Dict[str, Any]]
) -> Tuple[float, Dict[str, Any]]:
    """
    Aggregate multiple evaluation scores and their metadata.

    This implementation extends the base aggregation with subprocess-specific metrics.

    Args:
        scores: List of individual scores to aggregate
        metadata_list: List of metadata dictionaries corresponding to each score

    Returns:
        A tuple containing:
            - float: The aggregated score
            - Dict[str, Any]: Aggregated metadata with execution statistics
    """
    # Use the base implementation for basic aggregation
    aggregated_score, aggregated_metadata = super().aggregate_scores(
        scores, metadata_list
    )

    # Add subprocess-specific aggregated metrics
    if metadata_list:
        # Count success/failure reasons
        reason_counts = {}
        for meta in metadata_list:
            reason = meta.get("reason", "unknown")
            reason_counts[reason] = reason_counts.get(reason, 0) + 1

        # Calculate timeout rate
        timeout_count = sum(
            1 for meta in metadata_list if meta.get("timed_out", False)
        )
        timeout_rate = timeout_count / len(metadata_list) if metadata_list else 0

        # Calculate exit code statistics
        exit_codes = [
            meta.get("exit_code") for meta in metadata_list if "exit_code" in meta
        ]
        success_exit_count = sum(
            1 for code in exit_codes if code in self.success_exit_codes
        )
        success_rate = success_exit_count / len(exit_codes) if exit_codes else 0

        # Add to aggregated metadata
        aggregated_metadata.update(
            {
                "reason_counts": reason_counts,
                "timeout_rate": timeout_rate,
                "success_rate": success_rate,
                "total_executions": len(metadata_list),
            }
        )

    return aggregated_score, aggregated_metadata

register_model classmethod

register_model()

Decorator to register a base model in the unified registry.

RETURNS	DESCRIPTION
Callable	A decorator function that registers the model class. TYPE: Callable[[Type[BaseModel]], Type[BaseModel]]

Source code in swarmauri_base/DynamicBase.py

@classmethod
def register_model(cls) -> Callable[[Type[BaseModel]], Type[BaseModel]]:
    """
    Decorator to register a base model in the unified registry.

    Returns:
        Callable: A decorator function that registers the model class.
    """

    def decorator(model_cls: Type[BaseModel]):
        """Register ``model_cls`` as a base model."""
        model_name = model_cls.__name__
        if model_name in cls._registry:
            glogger.warning(
                "Model '%s' is already registered; skipping duplicate.", model_name
            )
            return model_cls

        cls._registry[model_name] = {"model_cls": model_cls, "subtypes": {}}
        glogger.debug("Registered base model '%s'.", model_name)
        DynamicBase._recreate_models()
        return model_cls

    return decorator

register_type classmethod

register_type(resource_type=None, type_name=None)

Decorator to register a subtype under one or more base models in the unified registry.

PARAMETER	DESCRIPTION
resource_type	The base model(s) under which to register the subtype. If None, all direct base classes (except DynamicBase) are used. TYPE: Optional[Union[Type[T], List[Type[T]]]] DEFAULT: None
type_name	An optional custom type name for the subtype. TYPE: Optional[str] DEFAULT: None

RETURNS	DESCRIPTION
Callable	A decorator function that registers the subtype. TYPE: Callable[[Type[DynamicBase]], Type[DynamicBase]]

Source code in swarmauri_base/DynamicBase.py

@classmethod
def register_type(
    cls,
    resource_type: Optional[Union[Type[T], List[Type[T]]]] = None,
    type_name: Optional[str] = None,
) -> Callable[[Type["DynamicBase"]], Type["DynamicBase"]]:
    """
    Decorator to register a subtype under one or more base models in the unified registry.

    Parameters:
        resource_type (Optional[Union[Type[T], List[Type[T]]]]):
            The base model(s) under which to register the subtype. If None, all direct base classes (except DynamicBase)
            are used.
        type_name (Optional[str]): An optional custom type name for the subtype.

    Returns:
        Callable: A decorator function that registers the subtype.
    """

    def decorator(subclass: Type["DynamicBase"]):
        """Register ``subclass`` as a subtype."""
        if resource_type is None:
            resource_types = [
                base for base in subclass.__bases__ if base is not cls
            ]
        elif not isinstance(resource_type, list):
            resource_types = [resource_type]
        else:
            resource_types = resource_type

        for rt in resource_types:
            if not issubclass(subclass, rt):
                raise TypeError(
                    f"'{subclass.__name__}' must be a subclass of '{rt.__name__}'."
                )
            final_type_name = type_name or getattr(
                subclass, "_type", subclass.__name__
            )
            base_model_name = rt.__name__

            if base_model_name not in cls._registry:
                cls._registry[base_model_name] = {"model_cls": rt, "subtypes": {}}
                glogger.debug(
                    "Created new registry entry for base model '%s'.",
                    base_model_name,
                )

            subtypes_dict = cls._registry[base_model_name]["subtypes"]
            if final_type_name in subtypes_dict:
                glogger.warning(
                    "Type '%s' already exists under '%s'; skipping duplicate.",
                    final_type_name,
                    base_model_name,
                )
                continue

            subtypes_dict[final_type_name] = subclass
            glogger.debug(
                "Registered '%s' as '%s' under '%s'.",
                subclass.__name__,
                final_type_name,
                base_model_name,
            )

        DynamicBase._recreate_models()
        return subclass

    return decorator

model_validate_toml classmethod

model_validate_toml(toml_data)

Validate a model from a TOML string.

Source code in swarmauri_base/TomlMixin.py

@classmethod
def model_validate_toml(cls, toml_data: str):
    """Validate a model from a TOML string."""
    try:
        # Parse TOML into a Python dictionary
        toml_content = tomllib.loads(toml_data)

        # Convert the dictionary to JSON and validate using Pydantic
        return cls.model_validate_json(json.dumps(toml_content))
    except tomllib.TOMLDecodeError as e:
        raise ValueError(f"Invalid TOML data: {e}")
    except ValidationError as e:
        raise ValueError(f"Validation failed: {e}")

model_dump_toml

model_dump_toml(
    fields_to_exclude=None, api_key_placeholder=None
)

Return a TOML representation of the model.

Source code in swarmauri_base/TomlMixin.py

def model_dump_toml(self, fields_to_exclude=None, api_key_placeholder=None):
    """Return a TOML representation of the model."""
    if fields_to_exclude is None:
        fields_to_exclude = []

    # Load the JSON string into a Python dictionary
    json_data = json.loads(self.model_dump_json())

    # Function to recursively remove specific keys and handle api_key placeholders
    def process_fields(data, fields_to_exclude):
        """Recursively filter fields and apply placeholders."""
        if isinstance(data, dict):
            return {
                key: (
                    api_key_placeholder
                    if key == "api_key" and api_key_placeholder is not None
                    else process_fields(value, fields_to_exclude)
                )
                for key, value in data.items()
                if key not in fields_to_exclude
            }
        elif isinstance(data, list):
            return [process_fields(item, fields_to_exclude) for item in data]
        else:
            return data

    # Filter the JSON data
    filtered_data = process_fields(json_data, fields_to_exclude)

    # Convert the filtered data into TOML
    return toml.dumps(filtered_data)

model_validate_yaml classmethod

model_validate_yaml(yaml_data)

Validate a model from a YAML string.

Source code in swarmauri_base/YamlMixin.py

@classmethod
def model_validate_yaml(cls, yaml_data: str):
    """Validate a model from a YAML string."""
    try:
        # Parse YAML into a Python dictionary
        yaml_content = yaml.safe_load(yaml_data)

        # Convert the dictionary to JSON and validate using Pydantic
        return cls.model_validate_json(json.dumps(yaml_content))
    except yaml.YAMLError as e:
        raise ValueError(f"Invalid YAML data: {e}")
    except ValidationError as e:
        raise ValueError(f"Validation failed: {e}")

model_dump_yaml

model_dump_yaml(
    fields_to_exclude=None, api_key_placeholder=None
)

Return a YAML representation of the model.

Source code in swarmauri_base/YamlMixin.py

def model_dump_yaml(self, fields_to_exclude=None, api_key_placeholder=None):
    """Return a YAML representation of the model."""
    if fields_to_exclude is None:
        fields_to_exclude = []

    # Load the JSON string into a Python dictionary
    json_data = json.loads(self.model_dump_json())

    # Function to recursively remove specific keys and handle api_key placeholders
    def process_fields(data, fields_to_exclude):
        """Recursively filter fields and apply placeholders."""
        if isinstance(data, dict):
            return {
                key: (
                    api_key_placeholder
                    if key == "api_key" and api_key_placeholder is not None
                    else process_fields(value, fields_to_exclude)
                )
                for key, value in data.items()
                if key not in fields_to_exclude
            }
        elif isinstance(data, list):
            return [process_fields(item, fields_to_exclude) for item in data]
        else:
            return data

    # Filter the JSON data
    filtered_data = process_fields(json_data, fields_to_exclude)

    # Convert the filtered data into YAML using safe mode
    return yaml.safe_dump(filtered_data, default_flow_style=False)

model_post_init

model_post_init(logger=None)

Assign a logger instance after model initialization.

Source code in swarmauri_base/LoggerMixin.py

def model_post_init(self, logger: Optional[FullUnion[LoggerBase]] = None) -> None:
    """Assign a logger instance after model initialization."""

    # Directly assign the provided FullUnion[LoggerBase] or fallback to the
    # class-level default.
    self.logger = self.logger or logger or self.default_logger

evaluate

evaluate(program, **kwargs)

Evaluate a program and return a fitness score with metadata.

This method wraps the concrete _compute_score implementation, capturing execution time and handling exceptions.

PARAMETER	DESCRIPTION
program	The program to evaluate TYPE: IProgram
**kwargs	Additional parameters for the evaluation process DEFAULT: {}

RETURNS	DESCRIPTION
Tuple[float, Dict[str, Any]]	A tuple containing: - float: A scalar fitness score (higher is better) - Dict[str, Any]: Metadata about the evaluation, including feature dimensions

RAISES	DESCRIPTION
EvaluationError	If the evaluation process fails
TypeError	If the program is not of the expected type

Source code in swarmauri_base/evaluators/EvaluatorBase.py

def evaluate(self, program: Program, **kwargs) -> Tuple[float, Dict[str, Any]]:
    """
    Evaluate a program and return a fitness score with metadata.

    This method wraps the concrete _compute_score implementation, capturing
    execution time and handling exceptions.

    Args:
        program: The program to evaluate
        **kwargs: Additional parameters for the evaluation process

    Returns:
        A tuple containing:
            - float: A scalar fitness score (higher is better)
            - Dict[str, Any]: Metadata about the evaluation, including feature dimensions

    Raises:
        EvaluationError: If the evaluation process fails
        TypeError: If the program is not of the expected type
    """
    if not isinstance(program, Program):
        raise TypeError(f"Expected Program type, got {type(program)}")

    start_time = time.time()

    try:
        # Call the concrete implementation to compute the score
        score, metadata = self._compute_score(program, **kwargs)

        # Add execution time to metadata
        execution_time = time.time() - start_time
        metadata["execution_time"] = execution_time

        logger.debug(
            f"Evaluation completed in {execution_time:.4f}s with score {score:.4f}"
        )
        return score, metadata

    except Exception as e:
        execution_time = time.time() - start_time
        logger.error(f"Evaluation failed after {execution_time:.4f}s: {str(e)}")
        raise EvaluationError(f"Evaluation failed: {str(e)}") from e

reset

reset()

Reset the evaluator to its initial state.

This method is called to clear any internal state or cached data before a new evaluation cycle begins.

Source code in swarmauri_base/evaluators/EvaluatorBase.py

def reset(self) -> None:
    """
    Reset the evaluator to its initial state.

    This method is called to clear any internal state or cached data
    before a new evaluation cycle begins.
    """
    # Reset logic can be implemented in subclasses if needed
    logger.debug("Resetting evaluator state")
    pass