ReplicableValue#

class impacthpc.src.core.ReplicableValue(name: str | None, value: Quantity | str, min: Quantity | str | None, max: Quantity | str | None, standard_deviation: Quantity | str | None = None, warnings: List[str] = [], explaination: str | None = None, ontology: Ontology | None = None, important: bool = False)#

Bases: ABC

ReplicableValue is a core class of ImpactHPC for representing numerical values with associated uncertainties and provenance. This class enables the construction of operation trees, where each leaf represents a value (with its minimum, maximum, and standard deviation) and each node represents an operation that produces a result. Arithmetic operations between ReplicableValue instances automatically propagate uncertainties according to standard rules (see https://en.wikipedia.org/wiki/Propagation_of_uncertainty#Example_formulae). See ReplicableValue.

Schema of an operation represented as a tree, where leaves are variables and each node is an operator.
correlation_mode#

(static) Class-level setting for how uncertainties are combined during operations.

Type:

CorrelationMode

name#

Name of the value, if any.

Type:

str | None

value#

The central value, with units.

Type:

Quantity

min#

Minimum plausible value.

Type:

Quantity

max#

Maximum plausible value.

Type:

Quantity

standard_deviation#

Standard deviation (uncertainty) of the value.

Type:

Quantity

warnings#

List of warnings or notes about the value.

Type:

List[str]

explanation#

Optional explanation or description.

Type:

str | None

ontology#

Optional ontology tag for semantic annotation.

Type:

str | None

important#

True if the value is important. When true, this value is extracted during the explaination process in explain()

Type:

bool

Note

This class is abstract and should not be instantiated directly. Subclasses must implement _as_formula() and explain(). The two subclasses are SourcedValue and Operation.

Attributes Summary

correlation_mode

Methods Summary

average(values, unit, name, source)

Computes the average of the list as a SourcedValue.

explain([formatter])

Provides an explanation of the current object using the specified formatter.

ln(value)

Computes the natural logarithm of a value.

make_intermediate_result(name[, ...])

Makes this value an intermediate result.

set_correlation_mode(mode)

Context manager to temporarily set the correlation mode used for arithmetic operations involving ReplicableValue instances.

set_unit(unit)

Sets the unit of the value.

sum(values)

Computes the sum of a list of ReplicableValue using __add__(), handling duplicates.

Attributes Documentation

correlation_mode: CorrelationMode = 'INDEPENDENT'#

Methods Documentation

static average(values: List[float | int], unit: str, name: str, source: str) → SourcedValue#

Computes the average of the list as a SourcedValue.

  • value is the average of values.

  • min and max are the minimum and maximum of values.

  • standard_deviation is the standard deviation of values.

Parameters:

  • values (List[float | int]) – A list of numeric values to average.

  • unit (str) – The unit of the values as a string (e.g., “kg”, “m”). See impacthpc.src.core.config.ureg

  • name (str) – The name to associate with the resulting SourcedValue.

  • source (str) – The source or provenance of the values.

Returns:

An object containing the average, minimum, maximum, and standard deviation of the input values, all formatted with the specified unit, along with the provided name and source.

Return type:

SourcedValue

Raises:

ValueError – If the input list of values is empty.

explain(formatter: ~impacthpc.src.core.formatters.Formatter[~typing.Any, T] = <impacthpc.src.core.formatters.TextFormatter object>) → T#

Provides an explanation of the current object using the specified formatter.

Internally calls _explain_rec() and format the result with Formatter.format_result(). It extracts the important values using extract_important_values() and pass them to Formatter.format_result().

Parameters:
Returns:

The formatted explanation of the object.

Return type:

T

static ln(value: ReplicableValue) → ReplicableValue#

Computes the natural logarithm of a value.

Parameters:

value (ReplicableValue) – The value for which to compute the natural logarithm.

Returns:

The result of the natural logarithm as a new Operation object.

Return type:

ReplicableValue

Raises:

NotImplementedError – If the operand type is not supported.

make_intermediate_result(name: str, explaination: str | None = None, ontology: Ontology | None = None, override_unit: Unit | None = None)#

Makes this value an intermediate result. This is used in the explain() method. For example, if you have the computation A + B + C that isn’t easy to understand, you can decompose it with intermediate results by doing:

AandB = A + B
AandB.make_intermediate_result("AandB")
result = AandB + C
result.make_intermediate_result("result")

This will print something like:

result is [some_value]
Formula = AandB + C
where
    - AandB is [other value]
      Formula = A + B
      where
        - A is [value]
        - B is [value]
    - C is [value]

You can also add an explanation and an ontology to the results, and override its unit.

Parameters:

  • name (str) – The name of the intermediate result that will be displayed by explain().

  • explanation (str | None, optional) – If not None, sets the explanation attribute. Defaults to None.

  • ontology (str | None, optional) – If not None, sets the ontology attribute. Defaults to None.

  • override_unit (Unit | None, optional) – If not None, overrides the unit of the result by calling set_unit(). Defaults to None.

static set_correlation_mode(mode: CorrelationMode)#

Context manager to temporarily set the correlation mode used for arithmetic operations involving ReplicableValue instances. ReplicableValue instances represent random variables, where the value attribute is the expectation and the standard_deviation attribute quantifies uncertainty. When performing arithmetic operations on these values, the standard deviation of the result depends on the correlation between operands. In practice, determining the exact correlation between arbitrary values is often infeasible.

To address this, ImpactHPC restricts correlation handling to two modes: - Independent (correlation = 0): Standard deviations are combined assuming no correlation. - Dependent (correlation = 1): Standard deviations are combined assuming full positive correlation.

This context manager allows you to specify which correlation mode should be used within a given block of code. For example, when subtracting a value from itself (e.g., C - C), the operands are fully correlated, and the standard deviation of the result should be zero. In contrast, when adding unrelated values (e.g., A + B), the operands are considered independent.

Example usage: .. code-block:: python

with ReplicableValue.set_correlation_mode(CorrelationMode.DEPENDENT):

# Computations here assume full correlation (correlation = 1) result = C - C

# Outside the context, computations assume independence (correlation = 0) result2 = A + B

This approach simplifies the handling of correlations in complex systems, such as server impact computations, by allowing explicit control over correlation assumptions at the group or component level.

Parameters:

mode (CorrelationMode) – The correlation mode to use within the context.

set_unit(unit: Unit)#

Sets the unit of the value. ImpactHPC uses the library Pint (https://pint.readthedocs.io/en/stable/) for handling units.

Parameters:

unit (Unit) – The new unit. You should use a Unit from impacthpc.src.core.config.ureg.

Raises:

ValueError – Raised if the value should be dimensionless to be set but is not.

static sum(values: List[ReplicableValue]) → ReplicableValue#

Computes the sum of a list of ReplicableValue using __add__(), handling duplicates. Duplicates are removed: same_value + other_value + same_value + same_value becomes other_value + 3 * same_value.

Parameters:

values (List[ReplicableValue]) – The list to sum.

Returns:

The sum of the ReplicableValue instances in values.

Return type:

ReplicableValue

__add__(other: ReplicableValue | int | float) → ReplicableValue#
__mul__(other: ReplicableValue | int | float) → ReplicableValue#
__repr__() → str#

Return repr(self).

__sub__(other: ReplicableValue | int | float) → ReplicableValue#
__truediv__(other: ReplicableValue | int | float) → ReplicableValue#
static _add(a: ReplicableValue | int | float, b: ReplicableValue | int | float) → ReplicableValue#

Implements the addition operator for ReplicableValue objects. Supports addition with another ReplicableValue, int, or float. If other is an int or float, the value is added to the current value, considering a standard deviation of 0 and the same unit as self.

Parameters:
Returns:

The result of the addition as a new Operation object.

Return type:

ReplicableValue

Raises:

NotImplementedError – If the operand type is not supported.

abstractmethod _as_formula() → str#

Returns a string representing the formula that this ReplicableValue represents.

See also

SourcedValue._as_formula() Operation._as_formula()

static _div(a: ReplicableValue | int | float, b: ReplicableValue | int | float) → ReplicableValue#

Implements the division operator for ReplicableValue objects. Supports division with another ReplicableValue, int, or float. If other is an int or float, the value is added to the current value, considering a standard deviation of 0 and the same unit as self.

Parameters:
Returns:

The result of the division as a new Operation object.

Return type:

ReplicableValue

Raises:

NotImplementedError – If the operand type is not supported.

abstractmethod _explain_rec(formatter: ~impacthpc.src.core.formatters.Formatter[~typing.Any, T] = <impacthpc.src.core.formatters.TextFormatter object>, recursion_level: int = 0, already_seen_values: set[int] = {}) → T | list[T]#

Recusive method called by explain() to provides an explanation of the current object using the specified formatter.

See implementations in:
Args:static_usage_attributed_to_job

formatter (impacthpc.src.core.formatters.Formatter[T]): The formatter used to format the explanation. Defaults to TextFormatter. recursion_level (int, optional): The current recursion depth, used to control nested explanations. Defaults to 0. already_seen_values (set[int]): A set of hashs of the ReplicableValue that have already been traversed when a value is duplicated and present in several branches.

Returns:

The formatted explanation of the object.

Return type:

T

static _mul(a: ReplicableValue | int | float, b: ReplicableValue | int | float) → ReplicableValue#

Implements the multiplication operator for ReplicableValue objects. Supports multiplication with another ReplicableValue, int, or float. If other is an int or float, the value is added to the current value, considering a standard deviation of 0 and the same unit as self.

Parameters:
Returns:

The result of the multiplication as a new Operation object.

Return type:

ReplicableValue

Raises:

NotImplementedError – If the operand type is not supported.

static _sub(a: ReplicableValue | int | float, b: ReplicableValue | int | float) → ReplicableValue#

Implements the substraction operator for ReplicableValue objects. Supports substraction with another ReplicableValue, int, or float. If other is an int or float, the value is added to the current value, considering a standard deviation of 0 and the same unit as self.

Parameters:
Returns:

The result of the substraction as a new Operation object.

Return type:

ReplicableValue

Raises:

NotImplementedError – If the operand type is not supported.

static average(values: List[float | int], unit: str, name: str, source: str) → SourcedValue#

Computes the average of the list as a SourcedValue.

  • value is the average of values.

  • min and max are the minimum and maximum of values.

  • standard_deviation is the standard deviation of values.

Parameters:

  • values (List[float | int]) – A list of numeric values to average.

  • unit (str) – The unit of the values as a string (e.g., “kg”, “m”). See impacthpc.src.core.config.ureg

  • name (str) – The name to associate with the resulting SourcedValue.

  • source (str) – The source or provenance of the values.

Returns:

An object containing the average, minimum, maximum, and standard deviation of the input values, all formatted with the specified unit, along with the provided name and source.

Return type:

SourcedValue

Raises:

ValueError – If the input list of values is empty.

correlation_mode: CorrelationMode = 'INDEPENDENT'#
explain(formatter: ~impacthpc.src.core.formatters.Formatter[~typing.Any, T] = <impacthpc.src.core.formatters.TextFormatter object>) → T#

Provides an explanation of the current object using the specified formatter.

Internally calls _explain_rec() and format the result with Formatter.format_result(). It extracts the important values using extract_important_values() and pass them to Formatter.format_result().

Parameters:
Returns:

The formatted explanation of the object.

Return type:

T

static ln(value: ReplicableValue) → ReplicableValue#

Computes the natural logarithm of a value.

Parameters:

value (ReplicableValue) – The value for which to compute the natural logarithm.

Returns:

The result of the natural logarithm as a new Operation object.

Return type:

ReplicableValue

Raises:

NotImplementedError – If the operand type is not supported.

make_intermediate_result(name: str, explaination: str | None = None, ontology: Ontology | None = None, override_unit: Unit | None = None)#

Makes this value an intermediate result. This is used in the explain() method. For example, if you have the computation A + B + C that isn’t easy to understand, you can decompose it with intermediate results by doing:

AandB = A + B
AandB.make_intermediate_result("AandB")
result = AandB + C
result.make_intermediate_result("result")

This will print something like:

result is [some_value]
Formula = AandB + C
where
    - AandB is [other value]
      Formula = A + B
      where
        - A is [value]
        - B is [value]
    - C is [value]

You can also add an explanation and an ontology to the results, and override its unit.

Parameters:

  • name (str) – The name of the intermediate result that will be displayed by explain().

  • explanation (str | None, optional) – If not None, sets the explanation attribute. Defaults to None.

  • ontology (str | None, optional) – If not None, sets the ontology attribute. Defaults to None.

  • override_unit (Unit | None, optional) – If not None, overrides the unit of the result by calling set_unit(). Defaults to None.

static set_correlation_mode(mode: CorrelationMode)#

Context manager to temporarily set the correlation mode used for arithmetic operations involving ReplicableValue instances. ReplicableValue instances represent random variables, where the value attribute is the expectation and the standard_deviation attribute quantifies uncertainty. When performing arithmetic operations on these values, the standard deviation of the result depends on the correlation between operands. In practice, determining the exact correlation between arbitrary values is often infeasible.

To address this, ImpactHPC restricts correlation handling to two modes: - Independent (correlation = 0): Standard deviations are combined assuming no correlation. - Dependent (correlation = 1): Standard deviations are combined assuming full positive correlation.

This context manager allows you to specify which correlation mode should be used within a given block of code. For example, when subtracting a value from itself (e.g., C - C), the operands are fully correlated, and the standard deviation of the result should be zero. In contrast, when adding unrelated values (e.g., A + B), the operands are considered independent.

Example usage: .. code-block:: python

with ReplicableValue.set_correlation_mode(CorrelationMode.DEPENDENT):

# Computations here assume full correlation (correlation = 1) result = C - C

# Outside the context, computations assume independence (correlation = 0) result2 = A + B

This approach simplifies the handling of correlations in complex systems, such as server impact computations, by allowing explicit control over correlation assumptions at the group or component level.

Parameters:

mode (CorrelationMode) – The correlation mode to use within the context.

set_unit(unit: Unit)#

Sets the unit of the value. ImpactHPC uses the library Pint (https://pint.readthedocs.io/en/stable/) for handling units.

Parameters:

unit (Unit) – The new unit. You should use a Unit from impacthpc.src.core.config.ureg.

Raises:

ValueError – Raised if the value should be dimensionless to be set but is not.

static sum(values: List[ReplicableValue]) → ReplicableValue#

Computes the sum of a list of ReplicableValue using __add__(), handling duplicates. Duplicates are removed: same_value + other_value + same_value + same_value becomes other_value + 3 * same_value.

Parameters:

values (List[ReplicableValue]) – The list to sum.

Returns:

The sum of the ReplicableValue instances in values.

Return type:

ReplicableValue