gymbo::GDOptimizer Struct Reference

Gradient Descent Optimizer for Symbolic Path Constraints. More...

#include <gd.h>

Public Member Functions
	GDOptimizer (int num_epochs=100, float lr=1.0, float eps=1.0, float param_low=-10.0, float param_high=10.0, bool sign_grad=true, bool init_param_uniform_int=true, int seed=42)
	Constructor for GDOptimizer. More...
bool	eval (std::vector< Sym > &path_constraints, std::unordered_map< int, float > params)
	Evaluate if path constraints are satisfied for given parameters. More...
bool	solve (std::vector< Sym > &path_constraints, std::unordered_map< int, float > &params, bool is_init_params_const=true)
	Solve path constraints using gradient descent optimization. More...

Public Attributes
int	num_epochs
	Maximum number of optimization epochs. More...
float	lr
	Learning rate for gradient descent. More...
float	eps
	The smallest positive value. More...
float	param_low
	Lower bound for parameter initialization. More...
float	param_high
	Upper bound for parameter initialization. More...
bool	sign_grad
bool	init_param_uniform_int
bool	contain_randomized_vars
int	seed
	Random seed for initializing parameter values. More...
int	num_used_itr
	Number of used iterations during optimization. More...

Detailed Description

Gradient Descent Optimizer for Symbolic Path Constraints.

The GDOptimizer class provides functionality to optimize symbolic path constraints by using gradient descent. It aims to find parameter values that satisfy the given path constraints, making them true or non-positive.

Note

This class assumes that symbolic expressions are represented by the Sym type, and path constraints are a vector of Sym objects.

The optimization process relies on the gradient information of the path constraints. The optimizer iteratively updates the parameters until the constraints are satisfied or a maximum number of epochs is reached.

Constructor & Destructor Documentation

GDOptimizer()

gymbo::GDOptimizer::GDOptimizer ( int  num_epochs = 100, float  lr = 1.0, float  eps = 1.0, float  param_low = -10.0, float  param_high = 10.0, bool  sign_grad = true, bool  init_param_uniform_int = true, int  seed = 42  )

inline

Constructor for GDOptimizer.

Parameters

num_epochs	Maximum number of optimization epochs (default: 100).
lr	Learning rate for gradient descent (default: 1).
eps	The smallest positive value of the target type (default: 1.0).
param_low	Lower bound for parameter initialization (default: -10).
param_high	Upper bound for parameter initialization (default: 10).
sign_grad	If true, use sign gradient descent. Otherwise, use standard gradient descent. (default true).
init_param_uniform_int	Flag indicating whether initial parameter values are drawn from the uniform int distribution or uniform real distribution (default true).
seed	Random seed for initializing parameter values (default: 42).

Member Function Documentation

eval()

bool gymbo::GDOptimizer::eval ( std::vector< Sym > &  path_constraints, std::unordered_map< int, float >  params  )

inline

Evaluate if path constraints are satisfied for given parameters.

This function checks if the given path constraints are satisfied (non-positive) for the provided parameter values.

Parameters

path_constraints	Vector of symbolic path constraints.
params	Map of parameter values.

Returns

true if all constraints are satisfied; otherwise, false.

solve()

bool gymbo::GDOptimizer::solve ( std::vector< Sym > &  path_constraints, std::unordered_map< int, float > &  params, bool  is_init_params_const = true  )

inline

Solve path constraints using gradient descent optimization.

This function attempts to find parameter values that satisfy the given path constraints by using gradient descent optimization. It iteratively updates the parameters until the constraints are satisfied or the maximum number of epochs is reached.

Parameters

path_constraints	Vector of symbolic path constraints.
params	Map of parameter values (will be modified during optimization).
is_init_params_const	Flag indicating whether initial parameter values are constant.

Returns

true if the constraints are satisfied after optimization; otherwise, false.

Member Data Documentation

contain_randomized_vars

bool gymbo::GDOptimizer::contain_randomized_vars

If true, use aeval and agrad. Otherwise, use eval and grad.

eps

float gymbo::GDOptimizer::eps

The smallest positive value.

init_param_uniform_int

bool gymbo::GDOptimizer::init_param_uniform_int

Flag indicating whether initial parameter values are drawn from the uniform int distribution or uniform real distribution (default true).

lr

float gymbo::GDOptimizer::lr

Learning rate for gradient descent.

num_epochs

int gymbo::GDOptimizer::num_epochs

Maximum number of optimization epochs.

num_used_itr

int gymbo::GDOptimizer::num_used_itr

Number of used iterations during optimization.

param_high

float gymbo::GDOptimizer::param_high

Upper bound for parameter initialization.

param_low

float gymbo::GDOptimizer::param_low

Lower bound for parameter initialization.

seed

int gymbo::GDOptimizer::seed

Random seed for initializing parameter values.

sign_grad

bool gymbo::GDOptimizer::sign_grad

If true, use sign gradient descent. Otherwise, use standard gradient descent. (default true).

---

The documentation for this struct was generated from the following file:

• libgymbo/gd.h