Inheritance diagram for src.optimizer.Optimizer:

Collaboration diagram for src.optimizer.Optimizer:

Public Member Functions
def	__init__ (self, options, Objective, FF)
	Create an Optimizer object. More...

def	recover (self)

def	set_goodstep (self, val)
	Mark in each target that the previous optimization step was good or bad. More...

def	save_mvals_to_input (self, vals, priors=None, jobtype=None)
	Write a new input file (s_save.in) containing the current mathematical parameters. More...

def	Run (self)
	Call the appropriate optimizer. More...

def	adjh (self, trust)

def	MainOptimizer (self, b_BFGS=0)
	The main ForceBalance adaptive trust-radius pseudo-Newton optimizer. More...

def	step (self, xk, data, trust)
	Computes the next step in the parameter space. More...

def	NewtonRaphson (self)
	Optimize the force field parameters using the Newton-Raphson method (. More...

def	BFGS (self)
	Optimize the force field parameters using the BFGS method; currently the recommended choice (. More...

def	ScipyOptimizer (self, Algorithm="None")
	Driver for SciPy optimizations. More...

def	GeneticAlgorithm (self)
	Genetic algorithm, under development. More...

def	Simplex (self)
	Use SciPy's built-in simplex algorithm to optimize the parameters. More...

def	Powell (self)
	Use SciPy's built-in Powell direction-set algorithm to optimize the parameters. More...

def	Anneal (self)
	Use SciPy's built-in simulated annealing algorithm to optimize the parameters. More...

def	ConjugateGradient (self)
	Use SciPy's built-in conjugate gradient algorithm to optimize the parameters. More...

def	Scipy_BFGS (self)
	Use SciPy's built-in BFGS algorithm to optimize the parameters. More...

def	BasinHopping (self)
	Use SciPy's built-in basin hopping algorithm to optimize the parameters. More...

def	TruncatedNewton (self)
	Use SciPy's built-in truncated Newton (fmin_tnc) algorithm to optimize the parameters. More...

def	NewtonCG (self)
	Use SciPy's built-in Newton-CG (fmin_ncg) algorithm to optimize the parameters. More...

def	Scan_Values (self, MathPhys=1)
	Scan through parameter values. More...

def	ScanMVals (self)
	Scan through the mathematical parameter space. More...

def	ScanPVals (self)
	Scan through the physical parameter space. More...

def	SinglePoint (self)
	A single-point objective function computation. More...

def	Gradient (self)
	A single-point gradient computation. More...

def	Hessian (self)
	A single-point Hessian computation. More...

def	Precondition (self)
	An experimental method to determine the parameter scale factors that results in the best conditioned Hessian. More...

def	FDCheckG (self)
	Finite-difference checker for the objective function gradient. More...

def	FDCheckH (self)
	Finite-difference checker for the objective function Hessian. More...

def	readchk (self)
	Read the checkpoint file for the main optimizer. More...

def	writechk (self)
	Write the checkpoint file for the main optimizer. More...

Public Attributes
	OptTab
	A list of all the things we can ask the optimizer to do. More...

	mvals_bak
	The root directory. More...

	failmsg
	Print a special message on failure. More...

	goodstep
	Specify whether the previous optimization step was good or bad. More...

	iterinit
	The initial iteration number (nonzero if we restart a previous run.) More...

	iteration
	The current iteration number. More...

	Objective
	reset the global variable More...

	bhyp
	Whether the penalty function is hyperbolic. More...

	FF
	The force field itself. More...

	uncert
	Target types which introduce uncertainty into the objective function. More...

	bakdir

	resdir

	excision
	The indices to be excluded from the Hessian update. More...

	np
	Number of parameters. More...

	mvals0
	The original parameter values. More...

	read_mvals

	h

	chk

	H

	dx

	Val

	Grad

	Hess

	Penalty

	iter

	prev_bad

	xk_prev

	x_prev

	x_best

Detailed Description

Optimizer class.

Contains several methods for numerical optimization.

For various reasons, the optimizer depends on the force field and fitting targets (i.e. we cannot treat it as a fully independent numerical optimizer). The dependency is rather weak which suggests that I can remove it someday.

Definition at line 46 of file optimizer.py.

Constructor & Destructor Documentation

◆ init()

def src.optimizer.Optimizer.__init__	(	self,
		options,
		Objective,
		FF
	)

Create an Optimizer object.

The optimizer depends on both the FF and the fitting targets so there is a chain of dependencies: FF –> FitSim –> Optimizer, and FF –> Optimizer

Here's what we do:

Take options from the parser
Pass in the objective function, force field, all fitting targets

Definition at line 59 of file optimizer.py.

Member Function Documentation

◆ adjh()

def src.optimizer.Optimizer.adjh	(	self,
		trust
	)

Definition at line 382 of file optimizer.py.

◆ Anneal()

def src.optimizer.Optimizer.Anneal ( self )

Use SciPy's built-in simulated annealing algorithm to optimize the parameters.

See also: Optimizer::ScipyOptimizer

Definition at line 1208 of file optimizer.py.

Here is the call graph for this function:

◆ BasinHopping()

def src.optimizer.Optimizer.BasinHopping ( self )

Use SciPy's built-in basin hopping algorithm to optimize the parameters.

See also: Optimizer::ScipyOptimizer

Definition at line 1223 of file optimizer.py.

Here is the call graph for this function:

◆ BFGS()

def src.optimizer.Optimizer.BFGS ( self )

Optimize the force field parameters using the BFGS method; currently the recommended choice (.

See also: MainOptimizer)

Definition at line 956 of file optimizer.py.

Here is the call graph for this function:

◆ ConjugateGradient()

def src.optimizer.Optimizer.ConjugateGradient ( self )

Use SciPy's built-in conjugate gradient algorithm to optimize the parameters.

See also: Optimizer::ScipyOptimizer

Definition at line 1213 of file optimizer.py.

Here is the call graph for this function:

◆ FDCheckG()

def src.optimizer.Optimizer.FDCheckG ( self )

Finite-difference checker for the objective function gradient.

For each element in the gradient, use a five-point finite difference stencil to compute a finite-difference derivative, and compare it to the analytic result.

Definition at line 1530 of file optimizer.py.

Here is the call graph for this function:

◆ FDCheckH()

def src.optimizer.Optimizer.FDCheckH ( self )

Finite-difference checker for the objective function Hessian.

For each element in the Hessian, use a five-point stencil in both parameter indices to compute a finite-difference derivative, and compare it to the analytic result.

This is meant to be a foolproof checker, so it is pretty slow. We could write a faster checker if we assumed we had accurate first derivatives, but it's better to not make that assumption.

The second derivative is computed by double-wrapping the objective function via the 'wrap2' function.

Definition at line 1562 of file optimizer.py.

Here is the call graph for this function:

◆ GeneticAlgorithm()

def src.optimizer.Optimizer.GeneticAlgorithm ( self )

Genetic algorithm, under development.

It currently works but a genetic algorithm is more like a concept; i.e. there is no single way to implement it.

Todo:: Massive parallelization hasn't been implemented yet

Definition at line 1105 of file optimizer.py.

◆ Gradient()

def src.optimizer.Optimizer.Gradient ( self )

A single-point gradient computation.

Definition at line 1322 of file optimizer.py.

Here is the call graph for this function:

◆ Hessian()

def src.optimizer.Optimizer.Hessian ( self )

A single-point Hessian computation.

Definition at line 1330 of file optimizer.py.

Here is the call graph for this function:

◆ MainOptimizer()

def src.optimizer.Optimizer.MainOptimizer	(	self,
		b_BFGS = `0`
	)

The main ForceBalance adaptive trust-radius pseudo-Newton optimizer.

Tried and true in many situations. :)

   Usually this function is called with the BFGS or NewtonRaphson
   method.  The NewtonRaphson method is consistently the best
   method I have, because I always provide at least an
   approximate Hessian to the objective function.  The BFGS
   method works well, but if gradients are cheap the SciPy_BFGS
   method also works nicely.

   The method adaptively changes the step size.  If the step is
   sufficiently good (i.e. the objective function goes down by a
   large fraction of the predicted decrease), then the step size
   is increased; if the step is bad, then it rejects the step and
   tries again.

   The optimization is terminated after either a function value or
   step size tolerance is reached.

   @param[in] b_BFGS Switch to use BFGS (True) or Newton-Raphson (False)

Definition at line 416 of file optimizer.py.

Here is the call graph for this function:

◆ NewtonCG()

def src.optimizer.Optimizer.NewtonCG ( self )

Use SciPy's built-in Newton-CG (fmin_ncg) algorithm to optimize the parameters.

See also: Optimizer::ScipyOptimizer

Definition at line 1233 of file optimizer.py.

Here is the call graph for this function:

◆ NewtonRaphson()

def src.optimizer.Optimizer.NewtonRaphson ( self )

Optimize the force field parameters using the Newton-Raphson method (.

See also: MainOptimizer)

Definition at line 951 of file optimizer.py.

Here is the call graph for this function:

◆ Powell()

def src.optimizer.Optimizer.Powell ( self )

Use SciPy's built-in Powell direction-set algorithm to optimize the parameters.

See also: Optimizer::ScipyOptimizer

Definition at line 1203 of file optimizer.py.

Here is the call graph for this function:

◆ Precondition()

def src.optimizer.Optimizer.Precondition ( self )

An experimental method to determine the parameter scale factors that results in the best conditioned Hessian.

Condition number function to be optimized.

       Parameters
       ----------
       logrskeys : np.ndarray
           Logarithms of the rescaling factor of each parameter type.
           The optimization is done in the log space.
       multiply : bool
           If set to True, then the exponentiated logrskeys will
           multiply the existing rescaling factors defined in the force field.

       Returns
       -------
       float
           Condition number of the Hessian matrix.

Definition at line 1342 of file optimizer.py.

Here is the call graph for this function:

◆ readchk()

def src.optimizer.Optimizer.readchk ( self )

Read the checkpoint file for the main optimizer.

Definition at line 1590 of file optimizer.py.

◆ recover()

def src.optimizer.Optimizer.recover ( self )

Definition at line 210 of file optimizer.py.

Here is the call graph for this function:

◆ Run()

def src.optimizer.Optimizer.Run ( self )

Call the appropriate optimizer.

This is the method we might want to call from an executable.

Definition at line 321 of file optimizer.py.

Here is the call graph for this function:

◆ save_mvals_to_input()

def src.optimizer.Optimizer.save_mvals_to_input	(	self,
		vals,
		priors = `None`,
		jobtype = `None`
	)

Write a new input file (s_save.in) containing the current mathematical parameters.

Definition at line 254 of file optimizer.py.

Here is the call graph for this function:

◆ Scan_Values()

def src.optimizer.Optimizer.Scan_Values	(	self,
		MathPhys = `1`
	)

Scan through parameter values.

This option is activated using the inputs:

scan[mp]vals
scan_vals low:hi:nsteps
scan_idxnum (number) -or-
scan_idxname (name)

This method goes to the specified parameter indices and scans through the supplied values, evaluating the objective function at every step.

I hope this method will be useful for people who just want to look at changing one or two parameters and seeing how it affects the force field performance.

Todo:: Maybe a multidimensional grid can be done.

Parameters

[in] MathPhys Switch to use mathematical (True) or physical (False) parameters.

Definition at line 1259 of file optimizer.py.

Here is the call graph for this function:

◆ ScanMVals()

def src.optimizer.Optimizer.ScanMVals ( self )

Scan through the mathematical parameter space.

See also: Optimizer::ScanValues

Definition at line 1306 of file optimizer.py.

Here is the call graph for this function:

◆ ScanPVals()

def src.optimizer.Optimizer.ScanPVals ( self )

Scan through the physical parameter space.

See also: Optimizer::ScanValues

Definition at line 1311 of file optimizer.py.

Here is the call graph for this function:

◆ Scipy_BFGS()

def src.optimizer.Optimizer.Scipy_BFGS ( self )

Use SciPy's built-in BFGS algorithm to optimize the parameters.

See also: Optimizer::ScipyOptimizer

Definition at line 1218 of file optimizer.py.

Here is the call graph for this function:

◆ ScipyOptimizer()

def src.optimizer.Optimizer.ScipyOptimizer	(	self,
		Algorithm = `"None"`
	)

Driver for SciPy optimizations.

Using any of the SciPy optimizers requires that SciPy is installed. This method first defines several wrappers around the objective function that the SciPy optimizers can use. Then it calls the algorith mitself.

Parameters

[in] Algorithm The optimization algorithm to use, for example 'powell', 'simplex' or 'anneal'

Definition at line 969 of file optimizer.py.

◆ set_goodstep()

def src.optimizer.Optimizer.set_goodstep	(	self,
		val
	)

Mark in each target that the previous optimization step was good or bad.

Definition at line 247 of file optimizer.py.

◆ Simplex()

def src.optimizer.Optimizer.Simplex ( self )

Use SciPy's built-in simplex algorithm to optimize the parameters.

See also: Optimizer::ScipyOptimizer

Definition at line 1198 of file optimizer.py.

Here is the call graph for this function:

◆ SinglePoint()

def src.optimizer.Optimizer.SinglePoint ( self )

A single-point objective function computation.

Definition at line 1316 of file optimizer.py.

Here is the call graph for this function:

◆ step()

def src.optimizer.Optimizer.step	(	self,
		xk,
		data,
		trust
	)

Computes the next step in the parameter space.

There are lots of tricks here that I will document later.

   @param[in] G The gradient
   @param[in] H The Hessian
   @param[in] trust The trust radius

Definition at line 745 of file optimizer.py.

Here is the call graph for this function:

◆ TruncatedNewton()

def src.optimizer.Optimizer.TruncatedNewton ( self )

Use SciPy's built-in truncated Newton (fmin_tnc) algorithm to optimize the parameters.

See also: Optimizer::ScipyOptimizer

Definition at line 1228 of file optimizer.py.

Here is the call graph for this function:

◆ writechk()

def src.optimizer.Optimizer.writechk ( self )

Write the checkpoint file for the main optimizer.

Definition at line 1602 of file optimizer.py.

Here is the call graph for this function:

Member Data Documentation

◆ bakdir

src.optimizer.Optimizer.bakdir

Definition at line 182 of file optimizer.py.

◆ bhyp

src.optimizer.Optimizer.bhyp

Whether the penalty function is hyperbolic.

Definition at line 176 of file optimizer.py.

◆ chk

src.optimizer.Optimizer.chk

Definition at line 541 of file optimizer.py.

◆ dx

src.optimizer.Optimizer.dx

Definition at line 774 of file optimizer.py.

◆ excision

src.optimizer.Optimizer.excision

The indices to be excluded from the Hessian update.

Definition at line 189 of file optimizer.py.

◆ failmsg

src.optimizer.Optimizer.failmsg

Print a special message on failure.

Definition at line 160 of file optimizer.py.

◆ FF

src.optimizer.Optimizer.FF

The force field itself.

Definition at line 178 of file optimizer.py.

◆ goodstep

src.optimizer.Optimizer.goodstep

Specify whether the previous optimization step was good or bad.

Definition at line 162 of file optimizer.py.

◆ Grad

src.optimizer.Optimizer.Grad

Definition at line 776 of file optimizer.py.

◆ h

src.optimizer.Optimizer.h

Definition at line 387 of file optimizer.py.

◆ H

src.optimizer.Optimizer.H

Definition at line 773 of file optimizer.py.

◆ Hess

src.optimizer.Optimizer.Hess

Definition at line 777 of file optimizer.py.

◆ iter

src.optimizer.Optimizer.iter

Definition at line 779 of file optimizer.py.

◆ iteration

src.optimizer.Optimizer.iteration

The current iteration number.

Definition at line 166 of file optimizer.py.

◆ iterinit

src.optimizer.Optimizer.iterinit

The initial iteration number (nonzero if we restart a previous run.)

This will be invoked if we quit RIGHT at the start of an iteration (i.e.

jobs were launched but none were finished) If data exists in the temp-dir corresponding to the highest iteration number, read the data.

Definition at line 164 of file optimizer.py.

◆ mvals0

src.optimizer.Optimizer.mvals0

The original parameter values.

Check derivatives by finite difference after the optimization is over (for good measure)

Don't print a "result" force field if it's the same as the input.

Determine the save file name.

Parse the save file for mvals, if exist.

The "precondition" job type takes care of its own output files.

Definition at line 196 of file optimizer.py.

◆ mvals_bak

src.optimizer.Optimizer.mvals_bak

The root directory.

Determine the output file name.

The job type Initial step size trust radius Minimum trust radius (for noisy objective functions) Lower bound on Hessian eigenvalue (below this, we add in steepest descent) Lower bound on step size (will fail below this) Guess value for Brent Step size for numerical finite difference When the trust radius get smaller, the finite difference step might need to get smaller. Number of steps to average over Function value convergence threshold Step size convergence threshold Gradient convergence threshold Allow convergence on low quality steps Maximum number of optimization steps For scan[mp]vals: The parameter index to scan over For scan[mp]vals: The parameter name to scan over, it just looks up an index For scan[mp]vals: The values that are fed into the scanner Name of the checkpoint file that we're reading in Name of the checkpoint file that we're writing out Whether to write the checkpoint file at every step Adaptive trust radius adjustment factor Adaptive trust radius adjustment damping Whether to print gradient during each step of the optimization Whether to print Hessian during each step of the optimization Whether to print parameters during each step of the optimization Error tolerance (if objective function rises by less than this, then the optimizer will forge ahead!) Search tolerance (The Hessian diagonal search will stop if the change is below this threshold) Whether to make backup files Name of the original input file Number of convergence criteria that must be met Only backup the "mvals" input file once per calculation.

Clone the input file to the output,

Definition at line 158 of file optimizer.py.

◆ np

src.optimizer.Optimizer.np

Number of parameters.

Definition at line 192 of file optimizer.py.

◆ Objective

src.optimizer.Optimizer.Objective

reset the global variable

The objective function (needs to pass in when I instantiate)

Definition at line 174 of file optimizer.py.

◆ OptTab

src.optimizer.Optimizer.OptTab

A list of all the things we can ask the optimizer to do.

Definition at line 63 of file optimizer.py.

◆ Penalty

src.optimizer.Optimizer.Penalty

Definition at line 778 of file optimizer.py.

◆ prev_bad

src.optimizer.Optimizer.prev_bad

Definition at line 972 of file optimizer.py.

◆ read_mvals

src.optimizer.Optimizer.read_mvals

Definition at line 223 of file optimizer.py.

◆ resdir

src.optimizer.Optimizer.resdir

Definition at line 183 of file optimizer.py.

◆ uncert

src.optimizer.Optimizer.uncert

Target types which introduce uncertainty into the objective function.

Will re-evaluate the objective function when an optimization step is rejected

Definition at line 181 of file optimizer.py.

◆ Val

src.optimizer.Optimizer.Val

Definition at line 775 of file optimizer.py.

◆ x_best

src.optimizer.Optimizer.x_best

Definition at line 975 of file optimizer.py.

◆ x_prev

src.optimizer.Optimizer.x_prev

Definition at line 974 of file optimizer.py.

◆ xk_prev

src.optimizer.Optimizer.xk_prev

Definition at line 973 of file optimizer.py.

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

optimizer.py

Public Member Functions

Public Attributes

Detailed Description

Constructor & Destructor Documentation

◆ __init__()

Member Function Documentation

◆ adjh()

◆ Anneal()

◆ BasinHopping()

◆ BFGS()

◆ ConjugateGradient()

◆ FDCheckG()

◆ FDCheckH()

◆ GeneticAlgorithm()

◆ Gradient()

◆ Hessian()

◆ MainOptimizer()

◆ NewtonCG()

◆ NewtonRaphson()

◆ Powell()

◆ Precondition()

◆ readchk()

◆ recover()

◆ Run()

◆ save_mvals_to_input()

◆ Scan_Values()

◆ ScanMVals()

◆ ScanPVals()

◆ Scipy_BFGS()

◆ ScipyOptimizer()

◆ set_goodstep()

◆ Simplex()

◆ SinglePoint()

◆ step()

◆ TruncatedNewton()

◆ writechk()

Member Data Documentation

◆ bakdir

◆ bhyp

◆ chk

◆ dx

◆ excision

◆ failmsg

◆ FF

◆ goodstep

◆ Grad

◆ h

◆ H

◆ Hess

◆ iter

◆ iteration

◆ iterinit

◆ mvals0

◆ mvals_bak

◆ np

◆ Objective

◆ OptTab

◆ Penalty

◆ prev_bad

◆ read_mvals

◆ resdir

◆ uncert

◆ Val

◆ x_best

◆ x_prev

◆ xk_prev

◆ init()