Keyword reference

The keywords that are recognized by the Everest configuration system are described below. Each keyword indicates the expected data type, usually some basic type. It may also be a list, which indicates that the keyword can be provided as a (YAML) list. If a keyword is required or deprecated, this is also indicated.

EverestConfig

EverestConfig

Type: object
No Additional Properties

Controls

Type: array

Defines a list of controls.

Each control defines a group of control variables. Each group of control variables is defined by a list of variable definitions, for example like this:

controls:
  - name: point
      type: generic_control
      min: -1.0
      max: 1.0
      initial_guess: 0.1
      perturbation_magnitude: 0.001
      variables:
        - name: x
          initial_guess: 0.5
        - name: y
          min: 0.0
          max: 1.0
        - name: z

This defines a group of controls names point, with variables x, y and z. The names of variables passed to the optimizer will be composed of of the control name and the variable names: point.x, point.y, and point.z.

Note that many variable properties may be defined both at the level of the control group and at the variable level. Definitions at the level of the control group set the properties for all variables in the group, except when they are overridden by the same field at the level of the variable. For instance, in the example above, all variables have an initial guess of 0.1, except for the point.x variable which has an initial value of 0.5.

Control names must be unique. Variables names must be unique within a control group, except when they are modified with an index, like this:

controls:
  - name: point
      max: 1.0
      min: -1.0
      initial_guess: 0.25
      type: generic_control
      variables:
        - name: x
          index: 1
        - name: x
          index: 2
        - name: x
          index: 3

This defines three variables point.x.1, point.x.2 and point.x.3. When a variable has an index field, all variables in the control group must have an index, which must be a unique non-negative integer.

There is a shortcut for the case that the index fields form a sequence starting at 1. In this case, the indexed variables can be defined by providing a list of initial guesses. This is equivalent to the example above:

controls:
  - name: point
      max: 1.0
      min: -1.0
      type: generic_control
      variables:
        - name: x
          initial_guess: [0.25, 0.25, 0.25]

Must contain a minimum of 1 items

No Additional Items

Each item of this array must be:

ControlConfig

Type: object
No Additional Properties

Name

Type: string

The control group name.

Controls name must be unique.

Type

Type: enum (of string)

Control group type.

Only two allowed control types are accepted:

  • "well_control": Standard built-in Everest control type designed for field optimization
  • "generic_control": Enables the user to define controls types to be employed for customized optimization jobs.

Must be one of:

  • "well_control"
  • "generic_control"

Variables


List of control variables.

Type: array
No Additional Items

Each item of this array must be:

variable control

Type: object
No Additional Properties

Name

Type: string

Variable name.

The name of the variable must be unique within a control group, unless an index field is defined.

Control Type

Default: null

Variable data type.

If set, overrides the value of control_type of the control group.

Set to "integer" for discrete optimization, or "real" for continuous optimization (default). This may be ignored if the optimization algorithm that is used does not support this.

Type: enum (of string)

Must be one of:

  • "real"
  • "integer"
Type: null

Enabled

Default: null

Enable/disable the variable.

If set to False the variable will be kept kept constant at the initial guess value during optimization.

If set, overrides the value of enabled in the control group.

Type: boolean
Type: null

Scaled Range

Default: null

The target range of the variable scaling.

Internally variables are scaled between their minimal and maximum values (min/max settings) to the range given by target_range (default = [0, 1]).

Overrides the value of scaled_range in the control group.

This option has no effect on discrete controls.

Type: array

Must contain a minimum of 2 items

Must contain a maximum of 2 items

No Additional Items

Tuple Validation

Item at 1 must be:
Type: number
Item at 2 must be:
Type: number
Type: null

Min

Default: null

The minimum value of the variable kept during optimization.

Overrides the value of min in the control group.

The initial guess for this variable must respect this minimum value.

Type: number
Type: null

Max

Default: null

The maximum value of the variable kept during optimization.

Overrides the value of max in the control group.

The initial guess for this variable must respect this maximum value.

Type: number
Type: null

Perturbation Type

Type: enum (of string) Default: "absolute"

The perturbation type for the control group.

The perturbation_type keyword defines whether the perturbation magnitude (perturbation_magnitude) should be treated as an absolute value or as relative to the dynamic range of the controls.

Overrides the value of perturbation_type in the control group.

Must be one of:

  • "absolute"
  • "relative"

Perturbation Magnitude

Default: null

The perturbation magnitude for this variable.

This controls the magnitude of perturbations (e.g. the standard deviation in case of a normal distribution) of controls, used to approximate the gradient.

The interpretation of this field depends on the value of the perturbation_type field:

  • absolute: The given value is used as-is.
  • relative: The perturbation magnitude is calculated by multiplying the given by value by the difference between the max and min fields.

Overrides the value of perturbation_magnitude in the control group.

Type: number

Value must be strictly greater than 0

Type: null

Default: null

Sampler configuration for this variable.

A sampler specifications is not required, if not provided, a normal distribution is used.

If at least one control group or variable has a sampler specification, only the groups or variables with a sampler specification are perturbed. Controls/variables that do not have a sampler section will not be perturbed at all. If that is not desired, make sure to specify a sampler for each control group and/or variable (or none at all to use a normal distribution for each control).

Within the sampler section, the shared keyword can be used to direct the sampler to use the same perturbations for each realization.

Overrides the value of sampler in the control group.

SamplerConfig

Type: object
No Additional Properties

Backend

Default: null

[Deprecated]

The correct backend will be inferred by the method. If several backends have a method named A, pick a specific backend B by putting B/A in the method field.

Type: string
Type: null

Method

Type: string Default: "norm"

The sampling method or distribution used by the sampler backend.

The set of available methods depends on the sampler backend used. By default a plugin based on scipy.stats is used, implementing the following methods:

  • From Probability Distributions:

    • norm: Samples from a standard normal distribution (mean 0, standard deviation 1).
    • truncnorm: Samples from a truncated normal distribution (mean 0, std. dev. 1), truncated to the range [-1, 1].
    • uniform: Samples from a uniform distribution in the range [-1, 1].

  • From Quasi-Monte Carlo Sequences:

    • sobol: Uses Sobol' sequences.
    • halton: Uses Halton sequences.
    • lhs: Uses Latin Hypercube Sampling.

    Note: QMC samples are generated in the unit hypercube [0, 1]^d and then scaled to the hypercube [-1, 1]^d.

Options

Default: null

Specifies a dict of optional parameters for the sampler backend.

This dict of values is passed unchanged to the selected method in the backend.

Type: object

Additional Properties of any type are allowed.

Type: object
Type: null

Shared

Default: null

Whether to share perturbations between realizations.

Type: boolean
Type: null
Type: null

Initial Guess

Default: null

Initial guess for the variable.

Type: number
Type: null

Index

Default: null

A Non-negative integer value to distinguish variables with the same name within a control group.

index should be given either for all of the variables, or for none of them.

Type: integer

Value must be greater or equal to 0

Type: null
Type: array
No Additional Items

Each item of this array must be:

ControlVariableGuessListConfig

Type: object
No Additional Properties

Name

Type: string

Variable name.

The name of the variable must be unique within a control group, unless an index field is defined.

Control Type

Default: null

Variable data type.

If set, overrides the value of control_type of the control group.

Set to "integer" for discrete optimization, or "real" for continuous optimization (default). This may be ignored if the optimization algorithm that is used does not support this.

Type: enum (of string)

Must be one of:

  • "real"
  • "integer"
Type: null

Enabled

Default: null

Enable/disable the variable.

If set to False the variable will be kept kept constant at the initial guess value during optimization.

If set, overrides the value of enabled in the control group.

Type: boolean
Type: null

Scaled Range

Default: null

The target range of the variable scaling.

Internally variables are scaled between their minimal and maximum values (min/max settings) to the range given by target_range (default = [0, 1]).

Overrides the value of scaled_range in the control group.

This option has no effect on discrete controls.

Type: array

Must contain a minimum of 2 items

Must contain a maximum of 2 items

No Additional Items

Tuple Validation

Item at 1 must be:
Type: number
Item at 2 must be:
Type: number
Type: null

Min

Default: null

The minimum value of the variable kept during optimization.

Overrides the value of min in the control group.

The initial guess for this variable must respect this minimum value.

Type: number
Type: null

Max

Default: null

The maximum value of the variable kept during optimization.

Overrides the value of max in the control group.

The initial guess for this variable must respect this maximum value.

Type: number
Type: null

Perturbation Type

Type: enum (of string) Default: "absolute"

The perturbation type for the control group.

The perturbation_type keyword defines whether the perturbation magnitude (perturbation_magnitude) should be treated as an absolute value or as relative to the dynamic range of the controls.

Overrides the value of perturbation_type in the control group.

Must be one of:

  • "absolute"
  • "relative"

Perturbation Magnitude

Default: null

The perturbation magnitude for this variable.

This controls the magnitude of perturbations (e.g. the standard deviation in case of a normal distribution) of controls, used to approximate the gradient.

The interpretation of this field depends on the value of the perturbation_type field:

  • absolute: The given value is used as-is.
  • relative: The perturbation magnitude is calculated by multiplying the given by value by the difference between the max and min fields.

Overrides the value of perturbation_magnitude in the control group.

Type: number

Value must be strictly greater than 0

Type: null

Default: null

Sampler configuration for this variable.

A sampler specifications is not required, if not provided, a normal distribution is used.

If at least one control group or variable has a sampler specification, only the groups or variables with a sampler specification are perturbed. Controls/variables that do not have a sampler section will not be perturbed at all. If that is not desired, make sure to specify a sampler for each control group and/or variable (or none at all to use a normal distribution for each control).

Within the sampler section, the shared keyword can be used to direct the sampler to use the same perturbations for each realization.

Overrides the value of sampler in the control group.

SamplerConfig

Type: object
No Additional Properties

Backend

Default: null

[Deprecated]

The correct backend will be inferred by the method. If several backends have a method named A, pick a specific backend B by putting B/A in the method field.

Type: string
Type: null

Method

Type: string Default: "norm"

The sampling method or distribution used by the sampler backend.

The set of available methods depends on the sampler backend used. By default a plugin based on scipy.stats is used, implementing the following methods:

  • From Probability Distributions:

    • norm: Samples from a standard normal distribution (mean 0, standard deviation 1).
    • truncnorm: Samples from a truncated normal distribution (mean 0, std. dev. 1), truncated to the range [-1, 1].
    • uniform: Samples from a uniform distribution in the range [-1, 1].

  • From Quasi-Monte Carlo Sequences:

    • sobol: Uses Sobol' sequences.
    • halton: Uses Halton sequences.
    • lhs: Uses Latin Hypercube Sampling.

    Note: QMC samples are generated in the unit hypercube [0, 1]^d and then scaled to the hypercube [-1, 1]^d.

Options

Default: null

Specifies a dict of optional parameters for the sampler backend.

This dict of values is passed unchanged to the selected method in the backend.

Type: object

Additional Properties of any type are allowed.

Type: object
Type: null

Shared

Default: null

Whether to share perturbations between realizations.

Type: boolean
Type: null
Type: null

Initial Guess

Type: array of number Default: null

A List of initial values for the variables.

If given, a series of variables with the same name are generated, with their index field set by enumeration, staring from 1. The initial value of these variables is taken from this list, whereas all other fields keep the same given value.

 No Additional Items

Each item of this array must be:

Type: number

Initial Guess

Default: null

Initial guess for the control group.

The initial guess value that is assigned to all control variables in the control group.

This default value can be overridden at the variable level.

Type: number
Type: null

Control Type

Type: enum (of string) Default: "real"

Control data type for the control group.

The data type value that is assigned to all control variables in the control group.

Set to "integer" for discrete optimization, or "real" for continuous optimization (default). This may be ignored if the optimization algorithm that is used does not support this.

This default value can be overridden at the variable level.

Must be one of:

  • "real"
  • "integer"

Enabled

Type: boolean Default: true

Enable/disable control groups.

Whether all control variables in the control group are enabled or not. When not enabled, variables are kept constant at the initial guess value during optimization.

This default value can be overridden at the variable level.

Min

Default: null

The minimum values of all control variables in the control group.

The initial_guess field of this control group and of all of its variables must respect this minimum value.

This default value can be overridden at the variable level.

Type: number
Type: null

Max

Default: null

The maximum values of all control variables in the control group.

The initial_guess field of this control group and of all of its variables must respect this maximum value.

This default value can be overridden at the variable level.

Type: number
Type: null

Perturbation Type

Type: enum (of string) Default: "absolute"

The perturbation type for the control group.

The perturbation_type keyword defines whether the perturbation magnitude (perturbation_magnitude) should be treated as an absolute value or as relative to the dynamic range of the controls.

Must be one of:

  • "absolute"
  • "relative"

Perturbation Magnitude

Type: number

The magnitude of the perturbation of all control variables in the control group.

This controls the magnitude of perturbations (e.g. the standard deviation in case of a normal distribution) of controls, used to approximate the gradient.

The interpretation of this field depends on the value of the perturbation_type field:

  • absolute: The given value is used as-is.
  • relative: The perturbation magnitude is calculated by multiplying the given by value by the difference between the max and min fields.

This default value can be overridden at the variable level.

Scaled Range

Type: array Default: [0.0, 1.0]

The target range of the control variable scaling for the control group.

Internally control variables are scaled from their minimal and maximum values (min/max settings) to the range given by target_range (default = [0, 1]).

This default value can be overridden at the variable level.

This option has no effect on discrete controls.

Must contain a minimum of 2 items

Must contain a maximum of 2 items

No Additional Items

Tuple Validation

Item at 1 must be:
Type: number
Item at 2 must be:
Type: number

Default: null

Sampler configuration for a control group.

A sampler specification section applies to a group of controls, or to an individual control. Sampler specifications are not required, if no sampler sections are provided, a normal distribution is used.

This can be overridden at the variable level.

If at least one control group or variable has a sampler specification, only the groups or variables with a sampler specification are perturbed. Controls/variables that do not have a sampler section will not be perturbed at all. If that is not desired, make sure to specify a sampler for each control group and/or variable (or none at all to use a normal distribution for each control).

Within the sampler section, the shared keyword can be used to direct the sampler to use the same perturbations for each realization.

SamplerConfig

Type: object
No Additional Properties

Backend

Default: null

[Deprecated]

The correct backend will be inferred by the method. If several backends have a method named A, pick a specific backend B by putting B/A in the method field.

Type: string
Type: null

Method

Type: string Default: "norm"

The sampling method or distribution used by the sampler backend.

The set of available methods depends on the sampler backend used. By default a plugin based on scipy.stats is used, implementing the following methods:

  • From Probability Distributions:

    • norm: Samples from a standard normal distribution (mean 0, standard deviation 1).
    • truncnorm: Samples from a truncated normal distribution (mean 0, std. dev. 1), truncated to the range [-1, 1].
    • uniform: Samples from a uniform distribution in the range [-1, 1].

  • From Quasi-Monte Carlo Sequences:

    • sobol: Uses Sobol' sequences.
    • halton: Uses Halton sequences.
    • lhs: Uses Latin Hypercube Sampling.

    Note: QMC samples are generated in the unit hypercube [0, 1]^d and then scaled to the hypercube [-1, 1]^d.

Options

Default: null

Specifies a dict of optional parameters for the sampler backend.

This dict of values is passed unchanged to the selected method in the backend.

Type: object

Additional Properties of any type are allowed.

Type: object
Type: null

Shared

Default: null

Whether to share perturbations between realizations.

Type: boolean
Type: null
Type: null

Objective Functions

Type: array

List of objective function specifications.

Must contain a minimum of 1 items

No Additional Items

Each item of this array must be:

ObjectiveFunctionConfig

Type: object
No Additional Properties

Name

Type: string

The name of the objective function.

Weight

Default: null

The weight determines the importance of an objective function relative to the other objective functions.

Everest optimizes the weighted sum of all objectives. The weights of all objectives are normalized to a total value of one by dividing them by their sum. Weights may be zero or negative, but should add up to a positive value.

Type: number
Type: null

Scale

Default: null

Optional scaling of the objective function value.

Each objective function will be divided by this scale value. This can be used to change the overall range of the objective values. For instance, this may be needed to properly balance the relative scales of objectives and output constraints.

Note: This option should be used with care in case of multi-objective optimization, since scaling each objective differently will change their relative weights.

This option will be disabled if auto_scale is set in the optimization section.

Type: number

Value must be strictly greater than 0

Type: null

Type

Type: enum (of string) Default: "mean"

How to calculate the objective from realizations.

The objective function value is aggregated from their individual realization values, usually by averaging them. The calculation used for this operation can be modified by setting this field. Currently, the two values are supported:

  • "mean" (default): calculates the mean over the realizations.
  • "stddev": calculates the negative of the standard deviation over the realizations. The negative is used since in general the aim is to minimize the standard deviation (as opposed to the mean, which is preferred to be maximized).

Must be one of:

  • "mean"
  • "stddev"

OptimizationConfig

Type: object

Optimizer options.

Ensemble-based optimization in Everest is handled internally by the ropt library for robust optimization, which in turn uses a plugin mechanism to provide multiple low-level optimization backends. By default Everest includes backends based on the Dakota optimization package and on the SciPy package.

Many of the optimization options in this section are handled by ropt in an uniform way, i.e. they will be handled identically by the backend optimization algorithm. However, some options may be handled differently, or not be supported, depending on the optimization backend used. If this is the case, this will be indicated in the description of the option.

 No Additional Properties

Algorithm

Type: string Default: "optpp_q_newton"

The optimization algorithm used by Everest. Defaults to optpp_q_newton, a quasi-Newton algorithm in Dakota's OPT++ library.

Convergence Tolerance

Default: null

Defines the threshold value to test for convergence.

In most cases, this is used to set a relative convergence tolerance for the objective function: if the change in the objective function between one or more successive iterations divided by the previous objective function is less than convergence_tolerance, the optimization is terminated.

Note: this option is passed to the optimization backend unchanged, and therefore its actual interpretation depends on the algorithm used.

Type: number
Type: null

Backend

Default: null

[Deprecated]

The correct backend will be inferred by the method. If several backends have a method named A, pick a specific backend B by putting B/A in the method field.

Type: string
Type: null

Backend Options

Default: null

[Deprecated]

This field is deprecated. Please use options instead, it will accept both objects and lists of strings.

Type: object

Additional Properties of any type are allowed.

Type: object
Type: null

Options

Default: null

Specify options that are passed unchanged to the optimization algorithm.

This field accepts dictionaries with key/values pair or lists of strings. The type required depends on the optimization backend that is used. The Dakota backend requires a list of a strings that are added to the Dakota configuration. The SciPy backend requires a dictionary with values, which will be passed as keyword arguments to the optimizer.

Consult the documentation of the optimization backend for valid options.

Type: array of string
No Additional Items

Each item of this array must be:

Type: string
Type: object

Additional Properties of any type are allowed.

Type: object
Type: null

Constraint Tolerance

Default: null

Output constraint tolerance for Dakota.

Determines the maximum allowable value of infeasibility that any constraint in an optimization problem may possess and still be considered to be satisfied.

It is specified as a positive real value. If a constraint function is greater than this value then it is considered to be violated by the optimization algorithm. This specification gives some control over how tightly the constraints will be satisfied at convergence of the algorithm. However, if the value is set too small the algorithm may terminate with one or more constraints being violated.

This option is only used by the Dakota backend.

Type: number
Type: null

Default: null

Directs the optimizer to use CVaR estimation.

When this section is present, Conditional Value at Risk (CVaR) will be used to minimize risk. Effectively this means that at each iteration the objective and constraint functions will be calculated as the mean over the sub-set of the realizations that perform worst. The size of this set is specified as an absolute number or as a percentile value. These options are selected by setting either the number_of_realizations option, or the percentile option, which are mutually exclusive.

CVaRConfig

Type: object
No Additional Properties

Number Of Realizations

Default: null

The number of realizations used for CVaR estimation.

Sets the number of realizations that is used to calculate the total objective.

This option is mutually exclusive with the percentile option.

Type: integer
Type: null

Percentile

Default: null

The percentile used for CVaR estimation.

Sets the percentile of distribution of the objective over the realizations that is used to calculate the total objective.

This option is mutually exclusive with the number_of_realizations option.

Type: number

Value must be greater or equal to 0.0 and lesser or equal to 1.0

Type: null
Type: null

Max Batch Num

Default: null

Limit the number of batches of simulations.

The optimization will be terminated if the given number of batches has been reached.

Type: integer

Value must be strictly greater than 0

Type: null

Max Function Evaluations

Default: null

Limits the maximum number of function evaluations.

If the given number of function evaluations is reached, the optimization will be terminated. This differs from the max_batch_num and max_iterations options, because these may require multiple (or no) function evaluations.

Note: Evaluations of perturbed values are not included in this limit.

Type: integer

Value must be strictly greater than 0

Type: null

Max Iterations

Default: null

Limits the maximum number of iterations.

This limits the number of iterations that the optimization algorithm will perform.

This differs from max_batch_num and max_functions options, since an iteration may require multiple batches and/or function evaluations.

This option is passed through unchanged, and its exact interpretation depends on the backend optimization algorithm used.

Type: integer

Value must be strictly greater than 0

Type: null

Min Pert Success

Default: null

The number of perturbations that must succeed.

To calculate gradients, the optimizer requests a number of forward model evaluations for perturbed variables. There are two possible ways to calculate the gradient of the objective function:

  1. For each realization, perturbation_num objective functions are evaluated with perturbed controls, which will be used to calculate a gradient for that realization. The gradient calculation for a single realization is considered successful if at least min_pert_success forward model runs for that realization are successful. The overall gradient can be calculated from the set of realization gradients if a sufficient number succeeded (see the min_realizations_success option). If not, the optimization terminates with an error.
  2. If the perturbation_num field is equal to 1, the overall gradient is directly calculated from a single perturbation for each realization. In this case min_pert_success is internally set to 1, and the number of successful realizations is equal to the number of successful perturbed runs. The gradient calculation succeeds if this is at least equal to min_realizations_success.

Note: The value of min_pert_success is internally adjusted to be capped by perturbation_num.

Type: integer

Value must be strictly greater than 0

Type: null

Min Realizations Success

Default: null

Minimum number of successful realizations.

The overall objective functions and gradients are calculated from a set of forward model runs for a number of realizations of an ensemble. This is done by aggregating the functions and gradients, usually by averaging. If one or more of the forward model evaluations fail, this may lead to a failure of either the function or the gradient evaluation. However, the number of realizations that are successfully evaluated may still be sufficient to calculate robust aggregated functions and gradients.

To calculate the aggregated objective function and gradient, the number of successful forward models that calculate the objective function for each realization must be at least equal to min_realizations_success. The number of realizations that are successful in terms of the gradient calculation is additionally determined by the min_pert_success field, see its documentation for more details.

If these requirements are not met, the optimizer will stop and report an error.

Note: The value of min_realizations_success is internally adjusted to be capped by the number of realizations. It is possible to set the minimum number of successful realizations equal to zero. Some optimization algorithms are able to handle this and will proceed even if all realizations failed. Most algorithms are not capable of this and will internally adjust the value to be equal to one.

Type: integer

Value must be greater or equal to 0

Type: null

Perturbation Num

Default: null

The number of perturbed control vectors per realization.

The number of simulation runs used for estimating the gradient is equal to the the product of perturbation_num and model.realizations.

Type: integer

Value must be strictly greater than 0

Type: null

Speculative

Type: boolean Default: false

Calculate gradients, even if not strictly needed.

When running forward models in parallel, e.g. on a computing cluster, it may be advantageous to calculate a gradient along with each function evaluation. The reason is that the optimizer may be able to use the gradient in a next iteration. Although this wastes resources if the gradient is not used, it may save clock time, since the gradient does not need to be calculated again.

Notes:

  • This is mostly useful if a full set of functions and perturbations (for the gradient) can run in parallel. Additional batches may be needed anyway if this is not the case.
  • Running an optimization with or without this disabled will yield slightly different results, since the perturbed variables will differ because different (longer) sequences of random numbers will be generated.
  • This does potentially waste a lot of computational resources!

Parallel

Type: boolean Default: true

Enable parallel function evaluation.

If set to False, a single function evaluation is performed in each batch. In case of gradient-free optimizers this can be highly inefficient, since these tend to need many independent function evaluations at each iteration. By setting parallel to True, multiple functions may be evaluated in parallel, if supported by the optimization algorithm.

The default is to use parallel evaluation if supported.

Auto Scale

Type: boolean Default: false

Enable auto-scaling of objectives and input/output constraints.

If set to True, objectives, input constraints and output constraints are automatically scaled.

In the case of objectives and output constraints this is done by scaling by the values obtained in the first batch. Input constraints are scaled by dividing by the maximum values of their coefficients or right-hand-sides.

Note: If enabled, the individual scale options for objectives and constraints are not allowed.

ModelConfig

Type: object

Configuration of the ensemble model that is optimized.

 No Additional Properties

Realizations

Type: array of integer

Realizations to use from the ensemble.

This should be a list of realization ID's, indicating which realizations of the ensemble should be used.

The list can be specified directly as a (YAML) list, or as a string consisting of comma separated values and ranges.

For example 1, 2, 5-7, 10 is equivalent to [1, 2, 5, 6, 7, 10].

Must contain a minimum of 1 items

No Additional Items

Each item of this array must be:

Type: integer

Value must be greater or equal to 0

Realizations Weights

Type: array of number

A list of realization weights.

If provided, the list must consist of one weight for each realization (as determined by the realizations field). If not provided, the weight of each realization is equal to one divided by the number or realizations.

 No Additional Items

Each item of this array must be:

Type: number

EnvironmentConfig

Type: object

The environment of the optimization run.

This specifies which folders are used for simulation and output, as well as the level of detail in the logs

 No Additional Properties

Simulation Folder

Type: string Default: "simulation_folder"

Folder where the forward model outputs are stored.

It the provided path is not absolute, it is assumed to be relative with respect to the path given by output_folder.

Output Folder

Type: string Default: "everest_output"

Folder for Everest output.

Log Level

Type: enum (of string) Default: "info"

Defines the verbosity of logs output by Everest.

The default log level is info. The supported log levels are:

  • debug: Detailed information, typically of interest only when diagnosing problems.
  • info: Confirmation that things are working as expected.
  • warning: An indication that something unexpected happened, or indicative of some problem in the near future (e.g. disk space low). The software is still working as expected.
  • error: Due to a more serious problem, the software has not been able to perform some function.
  • critical: A serious error, indicating that the program will not be able to continue running.

Must be one of:

  • "debug"
  • "info"
  • "warning"
  • "error"
  • "critical"

Random Seed

Type: integer Default: null

Random seed.

A positive integer used to seed the random number generator that is used to generate perturbed controls for gradient estimation.

Value must be strictly greater than 0

Wells

Default: null

An optional list of well configurations.

Each well configuration consists of a name field, and an optional drill_time field. All variables in control groups with control.type == "well_control" must also be listed in this section.

If not present, a minimal well configuration containing only the names of wells will be derived from controls that have control.type == "well_control".

Type: array
No Additional Items

Each item of this array must be:

WellConfig

Type: object
No Additional Properties

Name

Type: string

The unique name of the well.

Drill Time

Type: integer Default: 0

Specifies the time it takes to drill the well.

Value must be strictly greater than 0

Type: null

Definitions

Type: object

Section for specifying variables.

Used to specify variables that will be replaced in the file when encountered as r{{ NAME }}.

Additional Properties of any type are allowed.

Type: object

Input Constraints

Type: array

A list of input constraint definitions.

Input constraints (linear constraints) are defined by a set of linear equations that must be equal to (or less- or greater-than) a given right-hand-side value. The optimizer will be directed to limit the control variables, during optimization, to the set of of values that obey all of these equations (i.e. find feasible control values).

The values of the linear constraints are calculated directly from the control vector values and do not require the results of the forward model that is being optimized.

How linear constraints are handled depends on the backend optimization algorithm: some may keep the control variables feasible during the entire optimization, some may allow constraint violations at intermediate trial points.

 No Additional Items

Each item of this array must be:

InputConstraintConfig

Type: object
No Additional Properties

Weights

Type: object

The coefficients of the linear equation that defines the input constraint.

Note: Although the term "weights" suggests that these coefficients must be normalized, or are internally normalized, this is not the case: they can take any value.
Example:

upper_bound: 0.2
weights:
- point_3D.x.0: 1
- point_3D.y.1: 1
- point_3D.z.2: 1

Each additional property must conform to the following schema

Type: number

Target

Default: null

Sets the right-hand-side value for a linear equality constraint.

This will direct the optimizer to limit the control values during optimization to the set of values where the sum of the controls multiplied by their coefficients (weights) is equal to target.

Type: number
Type: null

Lower Bound

Type: number Default: -Infinity

Sets the right-hand-side value for a linear equality constraint.

This will direct the optimizer to limit the control values during optimization to the set of values where the sum of the controls multiplied by their coefficients (weights) is equal to, or greater than lower_bound.

Upper Bound

Type: number Default: Infinity

Sets the right-hand-side value for a linear equality constraint.

This will direct the optimizer to limit the control values during optimization to the set of values where the sum of the controls multiplied by their coefficients (weights) is equal to, or less than upper_bound.

Scale

Default: null

Scaling of input constraints.

scale is a normalization factor which can be used to scale the input constraint. The bounds or target, and the weights will be scaled with this number.

This option will be disabled if auto_scale is set in the optimization section.

Type: number

Value must be strictly greater than 0

Type: null

Output Constraints

Type: array

A list of output constraint definitions.

Output constraints (non-linear constraints) are defined by a set of arbitrary functions that must be equal to (or less- or greater-than) a given right-hand-side value. The optimizer will attempt to find a solution that maximizes the objective function and does not violate these constraints.

The values of the non-linear constraints must be provided by the forward model that is being optimized.

How non-linear constraints are handled depends on the backend optimization algorithm: in most cases some intermediate trial points will violate the constraint.

 No Additional Items

Each item of this array must be:

OutputConstraintConfig

Type: object
No Additional Properties

Name

Type: string

The unique name of the output constraint.

Target

Default: null

Defines the equality constraint

f(x) = b,

where f is a function of the control vector x and b is the target.

Type: number
Type: null

Lower Bound

Type: number Default: -Infinity

Defines the equality constraint

f(x) >= b,

where f is a function of the control vector x and b is the lower bound.

Upper Bound

Type: number Default: Infinity

Defines the equality constraint

f(x) <= b,

where f is a function of the control vector x and b is the upper bound.

Scale

Default: null

Scaling of constraints.

scale is a normalization factor which can be used to scale the constraint to control its importance relative to the objective.

Both the bounds or target and the function evaluation value will be scaled with this number. For example, if the upper bound is 0.5 and the scaling is 10, then the function evaluation value will be divided by 10 and bounded from above by 0.05.

This option will be disabled if auto_scale is set in the optimization section.

Type: number

Value must be strictly greater than 0

Type: null

Install Jobs

Type: array

A list of jobs to install.

This defines the set of executables that can be referenced in the forward_model section.

 No Additional Items

Each item of this array must be:

InstallForwardModelStepConfig

Type: object
No Additional Properties

Name

Type: string

The name of the installed job.

This name is used to identify the job in the list of jobs to be executed, as defined by the forward_model field.

Source

Default: null

The source file of the ert job.

source is deprecated, please use executable instead.

Type: string
Type: null

Executable

Default: null

The executable linked to the job name.

Type: string
Type: null

Install Workflow Jobs

Type: array

A list of workflow jobs to install.

This defines the set of executables that can be referenced in the workflows section.

 No Additional Items

Each item of this array must be:

InstallWorkflowJobConfig

Type: object
No Additional Properties

Name

Type: string

The name of the installed job.

This name is used to identify the job in the list of jobs to be executed, as defined by the forward_model field.

Source

Default: null

The source file of the ert job.

source is deprecated, please use executable instead.

Type: string
Type: null

Executable

Default: null

The executable linked to the job name.

Type: string
Type: null

Install Data

Type: array

A list of directories and/or files to copy or link to the directory where the evaluation jobs will run.

 No Additional Items

Each item of this array must be:

InstallDataConfig

Type: object
No Additional Properties

Source

Default: null

Path to file or directory that needs to be copied or linked into the directory where the forward model executes.

Type: string
Type: null

Target

Type: string

Relative destination path to copy or link the given source to.

Data

Default: null

Data to be exported to the target file.

If provided, the data is exported to a JSON file. If target has no .json extension, it is added.

The data and source fields are mutually exclusive.

If data is provided, link will be ignored.

Type: object

Additional Properties of any type are allowed.

Type: object
Type: null

Install Templates

Type: array

Generate files using a jinja2 template.

A list of template files to be rendered to an output file. Extra JSON or YAML files can be provided with data that will be exposed as variables to the jinja2 renderer.

Example:

Given two files my_input.json and tmpl.jinja, stored in a files directory:

files/my_input.json:

{
    "my_variable": "my_value"
}

and files/tmpl.jinja:

This is written in my file: {{my_input.my_variable}}

Add this to the config file:

install_data:
  - source: files
    target: files

install_templates:
  - template: files/tmpl.jinja
    output_file: my_output.txt
    extra_data: files/my_input.json

This will produce an output file with the content:

This is written in my file: my_value
No Additional Items

Each item of this array must be:

InstallTemplateConfig

Type: object
No Additional Properties

Template

Type: string

The jinja2 template file.

Output File

Type: string

The name of the output file.

Extra Data

Default: null

Extra input files.

The content of each extra JSON or YAML file is exposed to the jinja2 renderer, using the name of the file as a variable name.

Type: string
Type: null

ServerConfig

Type: object

Defines Everest server settings, i.e., which queue system, queue name and queue options are used for the everest server. This makes it possible to reduce the resource requirements for the server that tend to be low compared with the forward models. This may prevent situations where everest times out because it can not add the server to the queue.

Queue system and queue name defaults to the same as simulator, and the server should not need to be configured by most users. This is also true for the --include-host and --exclude-host options that are used by the SLURM driver.

Note that changing values in this section has no impact on the resource requirements of the forward models.

 No Additional Properties

Queue System

Default: null

Defines which queue system the everest server is submitted to.

For example, to run the server locally even if the forward model may run on another system:

server:
  queue_system:
    name: local


LocalQueueOptions

Type: object
No Additional Properties

Name

Type: const Default: "local"
Specific value: "local"

Max Running

Type: integer Default: 0

Value must be greater or equal to 0

Submit Sleep

Type: number Default: 0.0

Value must be greater or equal to 0

Num Cpu

Type: integer Default: 1

Value must be greater or equal to 0

Realization Memory

Type: integer Default: 0

Amount of memory to set aside for a forward model.

This information is propagated to the queue system as the amount of memory to reserve/book for a realization to complete. It is up to the configuration of the queuing system how to treat this information, but usually it will stop more realizations being assigned to a compute node if the compute nodes memory is already fully booked.

Setting this number lower than the peak memory consumption of each realization puts the realization at risk of being killed in an out-of-memory situation. Setting this number higher than needed will give longer wait times in the queue.

For the local queue system, this keyword has no effect. In that scenario, you can use max_running to choke the memory consumption. scheduling of compute jobs.

realization_memory may be an integer value, indicating the number of bytes, or a string consisting of a number followed by a unit. The unit indicates the multiplier that is applied, and must start with one of these characters:

  • b, B: bytes
  • k, K: kilobytes (1024 bytes)
  • m, M: megabytes (1024**2 bytes)
  • g, G: gigabytes (1024**3 bytes)
  • t, T: terabytes (1024**4 bytes)
  • p, P: petabytes (1024**5 bytes)

Spaces between the number and the unit are ignored, and so are any characters after the first. For example: 2g, 2G, and 2 GB all resolve to the same value: 2 gigabytes, equaling 2 * 1024**3 bytes.

Value must be greater or equal to 0

Project Code

Default: null

Type: string
Type: null

Activate Script

Default: null

Type: string
Type: null

LsfQueueOptions

Type: object
No Additional Properties

Name

Type: const Default: "lsf"
Specific value: "lsf"

Max Running

Type: integer Default: 0

Value must be greater or equal to 0

Submit Sleep

Type: number Default: 0.0

Value must be greater or equal to 0

Num Cpu

Type: integer Default: 1

Value must be greater or equal to 0

Realization Memory

Type: integer Default: 0

Amount of memory to set aside for a forward model.

This information is propagated to the queue system as the amount of memory to reserve/book for a realization to complete. It is up to the configuration of the queuing system how to treat this information, but usually it will stop more realizations being assigned to a compute node if the compute nodes memory is already fully booked.

Setting this number lower than the peak memory consumption of each realization puts the realization at risk of being killed in an out-of-memory situation. Setting this number higher than needed will give longer wait times in the queue.

For the local queue system, this keyword has no effect. In that scenario, you can use max_running to choke the memory consumption. scheduling of compute jobs.

realization_memory may be an integer value, indicating the number of bytes, or a string consisting of a number followed by a unit. The unit indicates the multiplier that is applied, and must start with one of these characters:

  • b, B: bytes
  • k, K: kilobytes (1024 bytes)
  • m, M: megabytes (1024**2 bytes)
  • g, G: gigabytes (1024**3 bytes)
  • t, T: terabytes (1024**4 bytes)
  • p, P: petabytes (1024**5 bytes)

Spaces between the number and the unit are ignored, and so are any characters after the first. For example: 2g, 2G, and 2 GB all resolve to the same value: 2 gigabytes, equaling 2 * 1024**3 bytes.

Value must be greater or equal to 0

Project Code

Default: null

Type: string
Type: null

Activate Script

Default: null

Type: string
Type: null

Bhist Cmd

Default: null

Type: string

Must be at least 1 characters long

Type: null

Bjobs Cmd

Default: null

Type: string

Must be at least 1 characters long

Type: null

Bkill Cmd

Default: null

Type: string

Must be at least 1 characters long

Type: null

Bsub Cmd

Default: null

Type: string

Must be at least 1 characters long

Type: null

Exclude Host

Default: null

Type: string
Type: null

Lsf Queue

Default: null

Type: string

Must be at least 1 characters long

Type: null

Lsf Resource

Default: null

Type: string
Type: null

SlurmQueueOptions

Type: object
No Additional Properties

Name

Type: const Default: "slurm"
Specific value: "slurm"

Max Running

Type: integer Default: 0

Value must be greater or equal to 0

Submit Sleep

Type: number Default: 0.0

Value must be greater or equal to 0

Num Cpu

Type: integer Default: 1

Value must be greater or equal to 0

Realization Memory

Type: integer Default: 0

Amount of memory to set aside for a forward model.

This information is propagated to the queue system as the amount of memory to reserve/book for a realization to complete. It is up to the configuration of the queuing system how to treat this information, but usually it will stop more realizations being assigned to a compute node if the compute nodes memory is already fully booked.

Setting this number lower than the peak memory consumption of each realization puts the realization at risk of being killed in an out-of-memory situation. Setting this number higher than needed will give longer wait times in the queue.

For the local queue system, this keyword has no effect. In that scenario, you can use max_running to choke the memory consumption. scheduling of compute jobs.

realization_memory may be an integer value, indicating the number of bytes, or a string consisting of a number followed by a unit. The unit indicates the multiplier that is applied, and must start with one of these characters:

  • b, B: bytes
  • k, K: kilobytes (1024 bytes)
  • m, M: megabytes (1024**2 bytes)
  • g, G: gigabytes (1024**3 bytes)
  • t, T: terabytes (1024**4 bytes)
  • p, P: petabytes (1024**5 bytes)

Spaces between the number and the unit are ignored, and so are any characters after the first. For example: 2g, 2G, and 2 GB all resolve to the same value: 2 gigabytes, equaling 2 * 1024**3 bytes.

Value must be greater or equal to 0

Project Code

Default: null

Type: string
Type: null

Activate Script

Default: null

Type: string
Type: null

Sbatch

Type: string Default: "sbatch"

Must be at least 1 characters long

Scancel

Type: string Default: "scancel"

Must be at least 1 characters long

Scontrol

Type: string Default: "scontrol"

Must be at least 1 characters long

Sacct

Type: string Default: "sacct"

Must be at least 1 characters long

Squeue

Type: string Default: "squeue"

Must be at least 1 characters long

Exclude Host

Type: string Default: ""

Include Host

Type: string Default: ""

Partition

Default: null

Type: string

Must be at least 1 characters long

Type: null

Squeue Timeout

Type: number Default: 2

Value must be strictly greater than 0

Max Runtime

Default: null

Type: number

Value must be greater or equal to 0

Type: null

TorqueQueueOptions

Type: object
No Additional Properties

Name

Type: const Default: "torque"
Specific value: "torque"

Max Running

Type: integer Default: 0

Value must be greater or equal to 0

Submit Sleep

Type: number Default: 0.0

Value must be greater or equal to 0

Num Cpu

Type: integer Default: 1

Value must be greater or equal to 0

Realization Memory

Type: integer Default: 0

Amount of memory to set aside for a forward model.

This information is propagated to the queue system as the amount of memory to reserve/book for a realization to complete. It is up to the configuration of the queuing system how to treat this information, but usually it will stop more realizations being assigned to a compute node if the compute nodes memory is already fully booked.

Setting this number lower than the peak memory consumption of each realization puts the realization at risk of being killed in an out-of-memory situation. Setting this number higher than needed will give longer wait times in the queue.

For the local queue system, this keyword has no effect. In that scenario, you can use max_running to choke the memory consumption. scheduling of compute jobs.

realization_memory may be an integer value, indicating the number of bytes, or a string consisting of a number followed by a unit. The unit indicates the multiplier that is applied, and must start with one of these characters:

  • b, B: bytes
  • k, K: kilobytes (1024 bytes)
  • m, M: megabytes (1024**2 bytes)
  • g, G: gigabytes (1024**3 bytes)
  • t, T: terabytes (1024**4 bytes)
  • p, P: petabytes (1024**5 bytes)

Spaces between the number and the unit are ignored, and so are any characters after the first. For example: 2g, 2G, and 2 GB all resolve to the same value: 2 gigabytes, equaling 2 * 1024**3 bytes.

Value must be greater or equal to 0

Project Code

Default: null

Type: string
Type: null

Activate Script

Default: null

Type: string
Type: null

Qsub Cmd

Default: null

Type: string

Must be at least 1 characters long

Type: null

Qstat Cmd

Default: null

Type: string

Must be at least 1 characters long

Type: null

Qdel Cmd

Default: null

Type: string

Must be at least 1 characters long

Type: null

Queue

Default: null

Type: string

Must be at least 1 characters long

Type: null

Cluster Label

Default: null

Type: string

Must be at least 1 characters long

Type: null

Job Prefix

Default: null

Type: string

Must be at least 1 characters long

Type: null

Keep Qsub Output

Type: boolean Default: false
Type: null

SimulatorConfig

Type: object

Defines simulator settings, i.e., which queue system, queue name and queue options are used for running the forward models.

 No Additional Properties
Example:

queue_system:
  max_running: 3
  name: local

Cores Per Node

Default: null

Defines the number of CPUs when running the forward models. This can for example be used in conjunction with the Eclipse parallel keyword for multiple CPU simulation runs.

This keyword has no effect when running with the local queue.

Type: integer

Value must be strictly greater than 0

Type: null

Delete Run Path

Type: boolean Default: false

If True, delete the output folder of a successful forward model run.

Max Runtime

Default: null

Maximum allowed running time for each individual forward model job specified in seconds.

A value of 0 means unlimited runtime.

Type: integer

Value must be greater or equal to 0

Type: null

Max Memory

Default: null

Amount of memory to set aside for a forward model.

This information is propagated to the queue system as the amount of memory to reserve/book for a realization to complete. It is up to the configuration of the queuing system how to treat this information, but usually it will stop more realizations being assigned to a compute node if the compute nodes memory is already fully booked.

Setting this number lower than the peak memory consumption of each realization puts the realization at risk of being killed in an out-of-memory situation. Setting this number higher than needed will give longer wait times in the queue.

For the local queue system, this keyword has no effect. In that scenario, you can use max_running to choke the memory consumption. scheduling of compute jobs.

max_memory may be an integer value, indicating the number of bytes, or a string consisting of a number followed by a unit. The unit indicates the multiplier that is applied, and must start with one of these characters:

  • b, B: bytes
  • k, K: kilobytes (1024 bytes)
  • m, M: megabytes (1024**2 bytes)
  • g, G: gigabytes (1024**3 bytes)
  • t, T: terabytes (1024**4 bytes)
  • p, P: petabytes (1024**5 bytes)

Spaces between the number and the unit are ignored, and so are any characters after the first. For example: 2g, 2G, and 2 GB all resolve to the same value: 2 gigabytes, equaling 2 * 1024**3 bytes.

Type: integer
Type: string
Type: null

Queue System

Default: null

Defines which queue system the everest submits jobs to.


LocalQueueOptions

Type: object
No Additional Properties

Name

Type: const Default: "local"
Specific value: "local"

Max Running

Type: integer Default: 0

Value must be greater or equal to 0

Submit Sleep

Type: number Default: 0.0

Value must be greater or equal to 0

Num Cpu

Type: integer Default: 1

Value must be greater or equal to 0

Realization Memory

Type: integer Default: 0

Amount of memory to set aside for a forward model.

This information is propagated to the queue system as the amount of memory to reserve/book for a realization to complete. It is up to the configuration of the queuing system how to treat this information, but usually it will stop more realizations being assigned to a compute node if the compute nodes memory is already fully booked.

Setting this number lower than the peak memory consumption of each realization puts the realization at risk of being killed in an out-of-memory situation. Setting this number higher than needed will give longer wait times in the queue.

For the local queue system, this keyword has no effect. In that scenario, you can use max_running to choke the memory consumption. scheduling of compute jobs.

realization_memory may be an integer value, indicating the number of bytes, or a string consisting of a number followed by a unit. The unit indicates the multiplier that is applied, and must start with one of these characters:

  • b, B: bytes
  • k, K: kilobytes (1024 bytes)
  • m, M: megabytes (1024**2 bytes)
  • g, G: gigabytes (1024**3 bytes)
  • t, T: terabytes (1024**4 bytes)
  • p, P: petabytes (1024**5 bytes)

Spaces between the number and the unit are ignored, and so are any characters after the first. For example: 2g, 2G, and 2 GB all resolve to the same value: 2 gigabytes, equaling 2 * 1024**3 bytes.

Value must be greater or equal to 0

Project Code

Default: null

Type: string
Type: null

Activate Script

Default: null

Type: string
Type: null

LsfQueueOptions

Type: object
No Additional Properties

Name

Type: const Default: "lsf"
Specific value: "lsf"

Max Running

Type: integer Default: 0

Value must be greater or equal to 0

Submit Sleep

Type: number Default: 0.0

Value must be greater or equal to 0

Num Cpu

Type: integer Default: 1

Value must be greater or equal to 0

Realization Memory

Type: integer Default: 0

Amount of memory to set aside for a forward model.

This information is propagated to the queue system as the amount of memory to reserve/book for a realization to complete. It is up to the configuration of the queuing system how to treat this information, but usually it will stop more realizations being assigned to a compute node if the compute nodes memory is already fully booked.

Setting this number lower than the peak memory consumption of each realization puts the realization at risk of being killed in an out-of-memory situation. Setting this number higher than needed will give longer wait times in the queue.

For the local queue system, this keyword has no effect. In that scenario, you can use max_running to choke the memory consumption. scheduling of compute jobs.

realization_memory may be an integer value, indicating the number of bytes, or a string consisting of a number followed by a unit. The unit indicates the multiplier that is applied, and must start with one of these characters:

  • b, B: bytes
  • k, K: kilobytes (1024 bytes)
  • m, M: megabytes (1024**2 bytes)
  • g, G: gigabytes (1024**3 bytes)
  • t, T: terabytes (1024**4 bytes)
  • p, P: petabytes (1024**5 bytes)

Spaces between the number and the unit are ignored, and so are any characters after the first. For example: 2g, 2G, and 2 GB all resolve to the same value: 2 gigabytes, equaling 2 * 1024**3 bytes.

Value must be greater or equal to 0

Project Code

Default: null

Type: string
Type: null

Activate Script

Default: null

Type: string
Type: null

Bhist Cmd

Default: null

Type: string

Must be at least 1 characters long

Type: null

Bjobs Cmd

Default: null

Type: string

Must be at least 1 characters long

Type: null

Bkill Cmd

Default: null

Type: string

Must be at least 1 characters long

Type: null

Bsub Cmd

Default: null

Type: string

Must be at least 1 characters long

Type: null

Exclude Host

Default: null

Type: string
Type: null

Lsf Queue

Default: null

Type: string

Must be at least 1 characters long

Type: null

Lsf Resource

Default: null

Type: string
Type: null

SlurmQueueOptions

Type: object
No Additional Properties

Name

Type: const Default: "slurm"
Specific value: "slurm"

Max Running

Type: integer Default: 0

Value must be greater or equal to 0

Submit Sleep

Type: number Default: 0.0

Value must be greater or equal to 0

Num Cpu

Type: integer Default: 1

Value must be greater or equal to 0

Realization Memory

Type: integer Default: 0

Amount of memory to set aside for a forward model.

This information is propagated to the queue system as the amount of memory to reserve/book for a realization to complete. It is up to the configuration of the queuing system how to treat this information, but usually it will stop more realizations being assigned to a compute node if the compute nodes memory is already fully booked.

Setting this number lower than the peak memory consumption of each realization puts the realization at risk of being killed in an out-of-memory situation. Setting this number higher than needed will give longer wait times in the queue.

For the local queue system, this keyword has no effect. In that scenario, you can use max_running to choke the memory consumption. scheduling of compute jobs.

realization_memory may be an integer value, indicating the number of bytes, or a string consisting of a number followed by a unit. The unit indicates the multiplier that is applied, and must start with one of these characters:

  • b, B: bytes
  • k, K: kilobytes (1024 bytes)
  • m, M: megabytes (1024**2 bytes)
  • g, G: gigabytes (1024**3 bytes)
  • t, T: terabytes (1024**4 bytes)
  • p, P: petabytes (1024**5 bytes)

Spaces between the number and the unit are ignored, and so are any characters after the first. For example: 2g, 2G, and 2 GB all resolve to the same value: 2 gigabytes, equaling 2 * 1024**3 bytes.

Value must be greater or equal to 0

Project Code

Default: null

Type: string
Type: null

Activate Script

Default: null

Type: string
Type: null

Sbatch

Type: string Default: "sbatch"

Must be at least 1 characters long

Scancel

Type: string Default: "scancel"

Must be at least 1 characters long

Scontrol

Type: string Default: "scontrol"

Must be at least 1 characters long

Sacct

Type: string Default: "sacct"

Must be at least 1 characters long

Squeue

Type: string Default: "squeue"

Must be at least 1 characters long

Exclude Host

Type: string Default: ""

Include Host

Type: string Default: ""

Partition

Default: null

Type: string

Must be at least 1 characters long

Type: null

Squeue Timeout

Type: number Default: 2

Value must be strictly greater than 0

Max Runtime

Default: null

Type: number

Value must be greater or equal to 0

Type: null

TorqueQueueOptions

Type: object
No Additional Properties

Name

Type: const Default: "torque"
Specific value: "torque"

Max Running

Type: integer Default: 0

Value must be greater or equal to 0

Submit Sleep

Type: number Default: 0.0

Value must be greater or equal to 0

Num Cpu

Type: integer Default: 1

Value must be greater or equal to 0

Realization Memory

Type: integer Default: 0

Amount of memory to set aside for a forward model.

This information is propagated to the queue system as the amount of memory to reserve/book for a realization to complete. It is up to the configuration of the queuing system how to treat this information, but usually it will stop more realizations being assigned to a compute node if the compute nodes memory is already fully booked.

Setting this number lower than the peak memory consumption of each realization puts the realization at risk of being killed in an out-of-memory situation. Setting this number higher than needed will give longer wait times in the queue.

For the local queue system, this keyword has no effect. In that scenario, you can use max_running to choke the memory consumption. scheduling of compute jobs.

realization_memory may be an integer value, indicating the number of bytes, or a string consisting of a number followed by a unit. The unit indicates the multiplier that is applied, and must start with one of these characters:

  • b, B: bytes
  • k, K: kilobytes (1024 bytes)
  • m, M: megabytes (1024**2 bytes)
  • g, G: gigabytes (1024**3 bytes)
  • t, T: terabytes (1024**4 bytes)
  • p, P: petabytes (1024**5 bytes)

Spaces between the number and the unit are ignored, and so are any characters after the first. For example: 2g, 2G, and 2 GB all resolve to the same value: 2 gigabytes, equaling 2 * 1024**3 bytes.

Value must be greater or equal to 0

Project Code

Default: null

Type: string
Type: null

Activate Script

Default: null

Type: string
Type: null

Qsub Cmd

Default: null

Type: string

Must be at least 1 characters long

Type: null

Qstat Cmd

Default: null

Type: string

Must be at least 1 characters long

Type: null

Qdel Cmd

Default: null

Type: string

Must be at least 1 characters long

Type: null

Queue

Default: null

Type: string

Must be at least 1 characters long

Type: null

Cluster Label

Default: null

Type: string

Must be at least 1 characters long

Type: null

Job Prefix

Default: null

Type: string

Must be at least 1 characters long

Type: null

Keep Qsub Output

Type: boolean Default: false
Type: null

Resubmit Limit

Type: integer Default: 1

Specifies how many times a forward model may be submitted to the queue system.

A forward model may fail for external reasons, examples include node in the cluster crashing, network issues, etc. Therefore, it might make sense to resubmit a forward model in case it fails. resubmit_limit defines the number of times we will resubmit a failing forward model. If not specified, a default value of 1 will be used.

Value must be greater or equal to 0

Forward Model

Type: array

The list of jobs to run.

 No Additional Items

Each item of this array must be:

ForwardModelStepConfig

Type: object

Job

Type: string

The forward model step to execute.

A string that consists of the name of the job, followed by zero or more job options.

Results

Default: null


SummaryResults

Type: object

File Name

Type: string

The output file produced by the forward model.

Type

Type: const
Specific value: "summary"

Keys

Default: "*"

The list of keys to include in the result.

Defaults to '*' indicating all keys.

Type: const
Specific value: "*"
Type: array of string
No Additional Items

Each item of this array must be:

Type: string

GenDataResults

Type: object

File Name

Type: string

The output file produced by the forward model.

Type

Type: const
Specific value: "gen_data"
Type: null

WorkflowConfig

Type: object

Optional workflow jobs to run during optimization.

 No Additional Properties

Pre Simulation

Type: array of string

The list of workflow jobs triggered pre-simulation.

Each entry consists of the name of the job, followed by zero or more job options.

 No Additional Items

Each item of this array must be:

Type: string

Post Simulation

Type: array of string

The list of workflow jobs triggered post-simulation.

Each entry consists of the name of the job, followed by zero or more job options.

 No Additional Items

Each item of this array must be:

Type: string

ExportConfig

Type: object

Settings to control the exports of an optimization run by everest.

 No Additional Properties

Keywords

Type: array of string

A list of Eclipse keywords to export.

 No Additional Items

Each item of this array must be:

Type: string