JSON SCHEMA ReLeSO

The input file must be written to comply to the following json-schema.

BaseParser

Class parses the experiment definition and conducts the training. This class can be used to initialize the ReLeSO Framework from the command line by reading in a json representation of the RL based Shape Optimization problem which is to be solved via Reinforcement Learning.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
verbosity	Verbosity
agent	Agent
	anyOf	PPOAgent
		DDPGAgent
		SACAgent
		PretrainedAgent
		DQNAgent
		A2CAgent
environment	Environment
number_of_timesteps	Number Of Timesteps
	type	integer
	minimum	1
number_of_episodes	Number Of Episodes
	type	integer
	minimum	1
validation	Validation
n_environments	N Environments
	type	integer
	minimum	1
	default	1
normalize_training_values	Normalize Training Values
	type	boolean
	default	False
multi_env_sequential	Multi Env Sequential
	type	boolean
	default	False
episode_log_update	Episode Log Update
	type	integer
	minimum	1
	default	100
export_step_log	Export Step Log
	type	boolean
	default	False
step_log_update	Step Log Update
	type	integer
	minimum	0
	default	0
step_log_info	Step Log Info
	type	boolean
	default	False
additionalProperties	False

Verbosity

Verbosity class. Defines the settings for the different loggers used in the current experiment. This class is the only class which is copied to all children. (this happens outside of the the standard channels and will hopefully not break with multiprocessing) Please note, the parser logger only ever can have the following name ReLeSO_parser
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
parser	Parser
	type	string
	enum	ERROR, WARNING, DEBUG, INFO
	default	INFO
environment	Environment
	type	string
	enum	ERROR, WARNING, DEBUG, INFO
	default	INFO
logfile_location	Logfile Location
	type	string
	default	logging/
console_logging	Console Logging
	type	boolean
	default	False
base_logger_name	Base Logger Name
	type	string
	default	ReLeSO
environment_extension	Environment Extension
	type	string
	default	environment
additionalProperties	False

PPOAgent

PPO agent definition. PPO definition for the stable_baselines3 implementation for this algorithm. Variable comments are taken from the stable_baselines3 documentation.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
tensorboard_log	Tensorboard Log
	type	string
policy	Policy
	type	string
	enum	MlpPolicy, CnnPolicy, MultiInputPolicy
use_custom_feature_extractor	Use Custom Feature Extractor
	type	string
	enum	resnet18, mobilenet_v2
cfe_without_linear	Cfe Without Linear
	type	boolean
	default	False
policy_kwargs	Policy Kwargs
	type	object
type	Type
	type	string
	enum	PPO
learning_rate	Learning Rate
	type	number
	default	0.0003
n_steps	N Steps
	type	integer
	default	2048
batch_size	Batch Size
	type	integer
	default	64
n_epochs	N Epochs
	type	integer
	default	10
gamma	Gamma
	type	number
	default	0.99
gae_lambda	Gae Lambda
	type	number
	default	0.95
clip_range	Clip Range
	type	number
	default	0.2
ent_coef	Ent Coef
	type	number
	default	0.0
vf_coef	Vf Coef
	type	number
	default	0.5
seed	Seed
	type	integer
device	Device
	type	string
	default	auto
additionalProperties	False

DDPGAgent

DDPG agent definition. DDPG definition for the stable_baselines3 implementation for this algorithm. Variable comments are taken from the stable_baselines3 documentation.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
tensorboard_log	Tensorboard Log
	type	string
policy	Policy
	type	string
	enum	MlpPolicy, CnnPolicy, MultiInputPolicy
use_custom_feature_extractor	Use Custom Feature Extractor
	type	string
	enum	resnet18, mobilenet_v2
cfe_without_linear	Cfe Without Linear
	type	boolean
	default	False
policy_kwargs	Policy Kwargs
	type	object
type	Type
	type	string
	enum	DDPG
learning_rate	Learning Rate
	type	number
	default	0.001
buffer_size	Buffer Size
	type	integer
	default	100000
learning_starts	Learning Starts
	type	integer
	default	100
batch_size	Batch Size
	type	integer
	default	64
tau	Tau
	type	number
	default	0.005
gamma	Gamma
	type	number
	default	0.99
optimize_memory_usage	Optimize Memory Usage
	type	boolean
	default	False
seed	Seed
	type	integer
device	Device
	type	string
	default	auto
additionalProperties	False

SACAgent

SAC Agent definition. SAC definition for the stable_baselines3 implementation for this algorithm. Variable comments are taken from the stable_baselines3 documentation.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
tensorboard_log	Tensorboard Log
	type	string
policy	Policy
	type	string
	enum	MlpPolicy, CnnPolicy, MultiInputPolicy
use_custom_feature_extractor	Use Custom Feature Extractor
	type	string
	enum	resnet18, mobilenet_v2
cfe_without_linear	Cfe Without Linear
	type	boolean
	default	False
policy_kwargs	Policy Kwargs
	type	object
type	Type
	type	string
	enum	SAC
learning_rate	Learning Rate
	type	number
	default	0.001
buffer_size	Buffer Size
	type	integer
	default	1000000
learning_starts	Learning Starts
	type	integer
	default	100
batch_size	Batch Size
	type	integer
	default	64
tau	Tau
	type	number
	default	0.005
gamma	Gamma
	type	number
	default	0.99
optimize_memory_usage	Optimize Memory Usage
	type	boolean
	default	False
ent_coef	Ent Coef
	default	auto
	anyOf	type	string
		type	number
target_update_interval	Target Update Interval
	type	integer
	default	1
target_entropy	Target Entropy
	default	auto
	anyOf	type	string
		type	number
use_sde	Use Sde
	type	boolean
	default	False
sde_sample_freq	Sde Sample Freq
	type	integer
	default	-1
use_sde_at_warmup	Use Sde At Warmup
	type	boolean
	default	False
seed	Seed
	type	integer
device	Device
	type	string
	default	auto
additionalProperties	False

PretrainedAgent

Pretrained agent definition. This class can be used to load pretrained agents, instead of using untrained agents. Can also be used to only validate this agent without training it further. Please see validation section for this use-case.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
tensorboard_log	Tensorboard Log
	type	string
type	Type
	type	string
	enum	PPO, SAC, DDPG, A2C, DQN
path	Path
	anyOf	type	string
		format	file-path
		type	string
		format	path
tesorboard_run_directory	Tesorboard Run Directory
	anyOf	type	string
		type	string
		format	path
additionalProperties	False

DQNAgent

DQN Agent definition. DQN definition for the stable_baselines3 implementation for this algorithm. Variable comments are taken from the stable_baselines3 documentation.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
tensorboard_log	Tensorboard Log
	type	string
policy	Policy
	type	string
	enum	MlpPolicy, CnnPolicy, MultiInputPolicy
use_custom_feature_extractor	Use Custom Feature Extractor
	type	string
	enum	resnet18, mobilenet_v2
cfe_without_linear	Cfe Without Linear
	type	boolean
	default	False
policy_kwargs	Policy Kwargs
	type	object
type	Type
	type	string
	enum	DQN
learning_rate	Learning Rate
	type	number
	default	0.0001
buffer_size	Buffer Size
	type	integer
	default	1000000
learning_starts	Learning Starts
	type	integer
	default	256
batch_size	Batch Size
	type	integer
	default	32
tau	Tau
	type	number
	default	1.0
gamma	Gamma
	type	number
	default	0.99
train_freq	Train Freq
	type	integer
	default	4
gradient_steps	Gradient Steps
	type	integer
	default	1
optimize_memory_usage	Optimize Memory Usage
	type	boolean
	default	False
target_update_interval	Target Update Interval
	type	integer
	default	256
exploration_fraction	Exploration Fraction
	type	number
	default	0.1
exploration_initial_eps	Exploration Initial Eps
	type	number
	default	1.0
exploration_final_eps	Exploration Final Eps
	type	number
	default	0.05
max_grad_norm	Max Grad Norm
	type	number
	default	10
seed	Seed
	type	integer
device	Device
	type	string
	default	auto
additionalProperties	False

A2CAgent

A2c agent definition. A2C definition for the stable_baselines3 implementation for this algorithm. Variable comments are taken from the stable_baselines3 documentation.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
tensorboard_log	Tensorboard Log
	type	string
policy	Policy
	type	string
	enum	MlpPolicy, CnnPolicy, MultiInputPolicy
use_custom_feature_extractor	Use Custom Feature Extractor
	type	string
	enum	resnet18, mobilenet_v2
cfe_without_linear	Cfe Without Linear
	type	boolean
	default	False
policy_kwargs	Policy Kwargs
	type	object
type	Type
	type	string
	enum	A2C
learning_rate	Learning Rate
	type	number
	default	0.0007
n_steps	N Steps
	type	integer
	default	5
gamma	Gamma
	type	number
	default	0.99
gae_lambda	Gae Lambda
	type	number
	default	1.0
ent_coef	Ent Coef
	type	number
	default	0.0
vf_coef	Vf Coef
	type	number
	default	0.5
seed	Seed
	type	integer
device	Device
	type	string
	default	auto
max_grad_norm	Max Grad Norm
	type	number
	default	0.5
rms_prop_eps	Rms Prop Eps
	type	number
	default	1e-05
use_rms_prop	Use Rms Prop
	type	boolean
	default	True
use_sde	Use Sde
	type	boolean
	default	False
sde_sample_freq	Sde Sample Freq
	type	integer
	default	-1
normalize_advantage	Normalize Advantage
	type	boolean
	default	False
additionalProperties	False

MultiProcessing

Define multiprocessing capability of environment. Defines if the Problem should use Multiprocessing and with how many cores the solver can work. Does not force Multiprocessing for example if the solver does not support it.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
number_of_cores	Number Of Cores
	type	integer
	minimum	1
	default	1
additionalProperties	False

VariableLocation

Variable location class. Object of this class defines the position and movement possibilities for a single dimensions of a single control point of the geometry.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
current_position	Current Position
	type	number
min_value	Min Value
	type	number
max_value	Max Value
	type	number
n_steps	N Steps
	type	integer
	minimum	1
	default	10
step	Step
	type	number
additionalProperties	False

ShapeDefinition

Base of shape parameterization, also represents a simple point cloud.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
control_points	Control Points
	type	array
	items	type	array
		items	VariableLocation
additionalProperties	False

SplineSpaceDimension

Defines a single spline space dimension of the current spline. The dimension is a dimension of the parametric spline dimension.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
name	Name
	type	string
number_of_points	Number Of Points
	type	integer
	minimum	1
degree	Degree
	type	integer
	minimum	1
knot_vector	Knot Vector
	type	array
	items	type	number
additionalProperties	False

BSplineDefinition

Definition of the BSpline implementation of the ReLeSO Toolbox.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
control_points	Control Points
	type	array
	items	type	array
		items	VariableLocation
space_dimensions	Space Dimensions
	type	array
	items	SplineSpaceDimension
additionalProperties	False

NURBSDefinition

Definition of a NURBS spline. Definition of the NURBS implementation of the ReLeSO Toolbox, in comparison to the B-Spline implementation only an additional weights vector is added.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
control_points	Control Points
	type	array
	items	type	array
		items	VariableLocation
space_dimensions	Space Dimensions
	type	array
	items	SplineSpaceDimension
weights	Weights
	type	array
	items	anyOf	type	number
			VariableLocation
additionalProperties	False

Geometry

Definition of the Geometry.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
shape_definition	Shape Definition
	anyOf	ShapeDefinition
		BSplineDefinition
		NURBSDefinition
		type	array
		items	anyOf	ShapeDefinition
				BSplineDefinition
				NURBSDefinition
action_based_observation	Action Based Observation
	type	boolean
	default	True
discrete_actions	Discrete Actions
	type	boolean
	default	True
reset_with_random_action_values	Reset With Random Action Values
	type	boolean
	default	False
additionalProperties	False

MeshExporter

Class which defines in which format and where the mesh is exported.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
format	Format
	type	string
	enum	mixd
	default	mixd
export_path	Export Path
	anyOf	type	string
		type	string
		format	path
additionalProperties	False

MeshIOMesh

Provides an interface to load meshio meshes. Since gustaf currently only really supports .msh files, this also only supports files with this ending. If and when gustaf implements support for additional files types already supported by meshio. This will be updated.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
path	Path
	anyOf	type	string
		type	string
		format	path
export	MeshExporter
dimensions	Dimensions
	type	integer
	minimum	1
additionalProperties	False

MixdMesh

Class used to read in the correct mixd mesh file and load it.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
path	Path
	anyOf	type	string
		type	string
		format	path
export	MeshExporter
dimensions	Dimensions
	type	integer
	minimum	1
mxyz_path	Mxyz Path
	Please use either the path variable xor the mxyz variable, since if used both the used mxyz path might not be the one you think.
	type	string
	format	file-path
mien_path	Mien Path
	Please use either the path variable xor the mien variable, since if used both the used mien path might not be the one you think.
	type	string
	format	file-path
hypercube	Hypercube
	If True Mesh is made of hypercubes. If False Mesh is made of simplexes (triangles).
	type	boolean
	default	True
additionalProperties	False

FFDGeometry

FFD based variable shape.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
shape_definition	Shape Definition
	anyOf	ShapeDefinition
		BSplineDefinition
		NURBSDefinition
		type	array
		items	anyOf	ShapeDefinition
				BSplineDefinition
				NURBSDefinition
action_based_observation	Action Based Observation
	type	boolean
	default	True
discrete_actions	Discrete Actions
	type	boolean
	default	True
reset_with_random_action_values	Reset With Random Action Values
	type	boolean
	default	False
mesh	Mesh
	anyOf	MeshIOMesh
		MixdMesh
export_mesh	MeshExporter
additionalProperties	False

ObservationDefinition

Definition of an Observation. Definition of a single Observations by providing the name of the value and the range in which the value is limited to. The range is necessary due to normalization of the input of the agent networks.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
name	Name
	type	string
value_min	Value Min
	type	number
value_max	Value Max
	type	number
additionalProperties	False

ObservationDefinitionMulti

Define a multidimensional Observations space. Definition of a single Observations by providing the name of the value and the range in which the value is limited to. The range is necessary due to normalization of the input of the agent networks.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
name	Name
	type	string
value_min	Value Min
	type	number
value_max	Value Max
	type	number
observation_shape	Observation Shape
	type	array
	items	type	integer
value_type	Value Type
	type	string
	enum	float, CNN
additionalProperties	False

MultiProcessor

Definition of the utilized multiprocessor. Default MPIEXEC. The MultiProcessor gives access to command prefixes which can enable parallelization. Currently only tested for MPI. Note: The cluster uses PATH variables to correctly call mpi. IF you want to use MPI multiprocessing on the cluster please use ReLeSO.spor.MPIClusterMultiProcessor.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
command	Command
	type	string
	default	mpiexec -n
max_core_count	Max Core Count
	type	integer
	minimum	1
always_use	Always Use
	type	boolean
	default	False
additionalProperties	False

MPIClusterMultiProcessor

Multi-processor extension for the cluster MPI call.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
command	Command
	type	string
	default	$MPIEXEC
max_core_count	Max Core Count
	type	integer
	minimum	1
always_use	Always Use
	type	boolean
	default	False
location	Location
	type	string
	enum	cluster
mpi_flags_variable	Mpi Flags Variable
	type	string
	default	$FLAGS_MPI_BATCH
additionalProperties	False

SPORObjectCommandLine

Base definition of a SPORCommandLineObject. Can be instantiated. Base class for all possible SPOR object classes which use the command line. This class is meant to add some default functionality, so that users do not need to add them themselves. These are: access to the (MPI) Multi-processors via ReLeSO.spor.MultiProcessor command line command additional flags working directory whether or not to use the SPOR Communication Interface
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
name	Name
	type	string
stop_after_error	Stop After Error
	type	boolean
	default	True
reward_on_error	Reward On Error
	type	number
reward_on_completion	Reward On Completion
	type	number
run_on_reset	Run On Reset
	type	boolean
	default	True
additional_observations	Additional Observations
	anyOf	ObservationDefinition
		ObservationDefinitionMulti
		type	array
		items	anyOf	ObservationDefinition
				ObservationDefinitionMulti
multi_processor	Multi Processor
	anyOf	MultiProcessor
		MPIClusterMultiProcessor
use_communication_interface	Use Communication Interface
	type	boolean
	default	False
add_step_information	Add Step Information
	type	boolean
	default	False
working_directory	Working Directory
	type	string
execution_command	Execution Command
	type	string
command_options	Command Options
	type	array
	items	type	string
additionalProperties	False

SPORObjectExternalPythonFunction

Load a python function from an external file. This class is meant to load a python function from an external file and execute it. The function will be imported in to the package and run in the same python environment as the rest of the package is. If you want to call the function you can use the SPORCommandLineObject. The function needs to be called main and needs to have the following signature: args, logger, func_data Where args are the SPOR COMM arguments, logger is a logger provided by the SPORObject and, func_data is a persistent data variable which is not touched by the SPORObject.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
name	Name
	type	string
stop_after_error	Stop After Error
	type	boolean
	default	True
reward_on_error	Reward On Error
	type	number
reward_on_completion	Reward On Completion
	type	number
run_on_reset	Run On Reset
	type	boolean
	default	True
additional_observations	Additional Observations
	anyOf	ObservationDefinition
		ObservationDefinitionMulti
		type	array
		items	anyOf	ObservationDefinition
				ObservationDefinitionMulti
multi_processor	Multi Processor
	anyOf	MultiProcessor
		MPIClusterMultiProcessor
use_communication_interface	Use Communication Interface
	type	boolean
	default	False
add_step_information	Add Step Information
	type	boolean
	default	False
working_directory	Working Directory
	type	string
python_file_path	Python File Path
	anyOf	type	string
		type	string
		format	path
additionalProperties	False

SPORObjectInternalPythonFunction

This class will load SPOR functions provided by this package. Currently available functions are the following: xns_cnn: This function will use the result of the xns solver and create an image of the solution. The image is created via contourf.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
name	Name
	type	string
stop_after_error	Stop After Error
	type	boolean
	default	True
reward_on_error	Reward On Error
	type	number
reward_on_completion	Reward On Completion
	type	number
run_on_reset	Run On Reset
	type	boolean
	default	True
additional_observations	Additional Observations
	anyOf	ObservationDefinition
		ObservationDefinitionMulti
		type	array
		items	anyOf	ObservationDefinition
				ObservationDefinitionMulti
multi_processor	Multi Processor
	anyOf	MultiProcessor
		MPIClusterMultiProcessor
use_communication_interface	Use Communication Interface
	type	boolean
	default	False
add_step_information	Add Step Information
	type	boolean
	default	False
working_directory	Working Directory
	type	string
function_name	Function Name
	type	string
	enum	xns_cnn
additionalProperties	False

SPORList

The SPORList defines the custom defined steps. In the current version these steps take place after the FFD is done. These can include the fluid solver, post-processing steps, reward generation, additional logging, etc. In a future version the FFD might be incorporated into a SPOR step to stream line the internal functionality. If a step sends a stop signal all consecutive steps will be ignored and the Each step can generate observations, rewards and info for the current episode step. These are aggregated as follows.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
steps	Steps
	type	array
	items	anyOf	SPORObjectCommandLine
			SPORObjectExternalPythonFunction
			SPORObjectInternalPythonFunction
reward_aggregation	Reward Aggregation
	type	string
	enum	sum, min, max, mean, absmax
additionalProperties	False

Environment

Parser class of which the environment is based. Parser Environment object is created by pydantic during the parsing of the json object defining the RL based Shape optimization. Each object can create a gym environment that represents the given problem.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
multi_processing	MultiProcessing
geometry	Geometry
	anyOf	Geometry
		FFDGeometry
spor	SPORList
max_timesteps_in_episode	Max Timesteps In Episode
	type	integer
	minimum	1
end_episode_on_geometry_not_changed	End Episode On Geometry Not Changed
	type	boolean
	default	False
reward_on_geometry_not_changed	Reward On Geometry Not Changed
	type	number
reward_on_episode_exceeds_max_timesteps	Reward On Episode Exceeds Max Timesteps
	type	number
additionalProperties	False

Validation

Parser class to define the validation to be performed during training. This class is used for the configuration of how validation is to be performed during training.
type	object
properties
save_location	Save Location
	type	string
	format	path
logger_name	Logger Name
	type	string
validation_freq	Validation Freq
	type	integer
	minimum	1
validation_values	Validation Values
	type	array
	items	type	number
	minItems	1
save_best_agent	Save Best Agent
	type	boolean
validate_on_training_end	Validate On Training End
	type	boolean
max_timesteps_in_episode	Max Timesteps In Episode
	type	integer
	minimum	1
end_episode_on_geometry_not_changed	End Episode On Geometry Not Changed
	type	boolean
	default	False
reward_on_geometry_not_changed	Reward On Geometry Not Changed
	type	number
reward_on_episode_exceeds_max_timesteps	Reward On Episode Exceeds Max Timesteps
	type	number
additionalProperties	False