Classes
class	ArcticClusterTree
	Cluster tree for maximization (in arctic semiring) More...

class	ArcticEvaluationAlgebra
	Maximization algebra for optimization. More...

class	ArcticFunctionAdapter
	Adapt function for maximization algebra. More...

class	ArcticOptimizer
	Maximizing optimizer (based on arctic algebra) More...

class	BoltzmannSampler
	Boltzmann sampler. More...

class	ClusterTreeBase
	Cluster tree base class. More...

class	ConsistencyError

class	ConstraintFunctionDefinitionError

class	EngineBase
	Abstract base class for samplers and optimizers. More...

class	EvaluationAlgebra
	Algebra for evaluating a constraint network. More...

class	Feature
	Feature in multi-dimensional Boltzmann sampling. More...

class	FeatureStatistics
	Keeping statistics on features. More...

class	Model

class	MultiDimensionalBoltzmannSampler
	Multi-dimensional Boltzmann sampler. More...

class	PFClusterTree
	Cluster tree for partition function calculation and sampling. More...

class	PFEvaluationAlgebra
	Partition function algebra for sampling. More...

class	PFFunctionAdapter
	Adapt function for partition function algebra. More...

class	ValueIn
	Constrain variable to have a value from a specified set. More...

class	WeightedFunction
	Function of a constraint network. More...

Functions
	seed (seed=None)
	Seed random number generator of libinfrared and treedecomp.

	def_function_class (classname, init, value, module="__main__")
	Define a function class (of type WeightedFunction)

	def_constraint_class (classname, init, value, module="__main__")
	Define a Constraint class.

	mc_optimize (model, objective, steps, temp, start=None)
	Optimize by Monte-Carlo optimization with partial resampling.

	dotfile_to_tgt (graphfile, tgt, outfile=None)
	Convert dot graph file format to png/pdf.

	dotfile_to_pdf (graphfile, outfile=None)
	Convert dot graph file to pdf.

	dotfile_to_png (graphfile, outfile=None)
	Convert dot graph file to png.

Variables
	Optimizer = ArcticOptimizer
	Alias for default optimizer ArcticOptimizer.

	Sampler = MultiDimensionalBoltzmannSampler
	Alias for MultiDimensionalBoltzmannSampler.

Detailed Description

The infrared high-level Python interface.

Function Documentation

◆ def_constraint_class()

infrared.infrared.def_constraint_class	(	classname,
		init,
		value,
		module = `"__main__"`
	)

Define a Constraint class.

Defines a new class with name classname (by default, in the main namespace) This defines a type of constraints that can be added to models.

Parameters

classname	name of the class to be defined
init	init function of generated class
value	value function of generated class
module	module where class is generated

The definitions of constraint and function classes work in the same way, only differing in the return type of the value function. In the case of constraints, it indicates satisfaction of the constraint by a boolean (at concrete values of specified variables); in the case of functions, it returns a numerical value.

The function init returns the dependency list, i.e. the list of indices of the variables on which the constraint depends.

The init function defines arguments that are passed when constructing the constraint. Typically, these include variable indices. Additional information required in the construction and/or evaluation of the constraint can be passed via arbitrary other parameters to init.

The value function computes the value of the constraint for determined values of the variables in the dependency list. These values are passed (in the order of the dependency list) as arguments of the init function; the names of these arguments must not occur in the signature of init.

The value function can then define further arguments with names that occur in the signature of the init function. These arguments are then stored at construction and due to this mechanism made available in the value function.

Examples

_bpcomp_tab = [(0, 3), (1, 2), (2, 1), (2, 3), (3, 0), (3, 2)]
def_constraint_class('BPComp', lambda i, j: [i, j],
    lambda x, y: (x, y) in _bpcomp_tab)

def_function_class('ConstrainDimer',
    # define dependency on consecutive variables X_i, X_i+1,
    # specify allowed dimers, e.g. xs=[(1,2),(2,1)]
    init  = lambda i, xs: [i, i+1],
 
    # check at assigned values X_i->x, X_i+1->y for specified xs
    value = lambda x, y, xs: (x,y) in xs)

More complex examples are provided in the accompanying Jupyter notebooks.

◆ def_function_class()

infrared.infrared.def_function_class	(	classname,
		init,
		value,
		module = `"__main__"`
	)

Define a function class (of type WeightedFunction)

Defines a new class with name classname (by default, in the main namespace) This defines a type of functions that can be added to models.

Parameters

classname	name of the class to be defined
init	init function of generated class
value	value function of generated class
module	module where class is generated

The definitions of constraint and function classes work in the same way, only differing in the return type of the value function. In the case of constraints, it indicates satisfaction of the constraint by a boolean (at concrete values of specified variables); in the case of functions, it returns a numerical value.

The function init returns the dependency list, i.e. the list of indices of the variables on which the function depends.

The init function defines arguments that are passed when constructing the constraint. Typically, these include variable indices. Additional information required in the construction and/or evaluation of the function can be passed via arbitrary other parameters to init.

The value function computes the value of the constraint for determined values of the variables in the dependency list. These values are passed (in the order of the dependency list) as arguments of the init function; the names of these arguments must not occur in the signature of init.

The value function can then define further arguments with names that occur in the signature of the init function. These arguments are then stored at construction and due to this mechanism made available in the value function.

Examples

def_function_class('GCCont', lambda i: [i], lambda x: 1 if x == 1 or x == 2 else 0)

def_function_class('CountDimer',
    # define dependency on consecutive variables,
    # specify to count pairs in xs, e.g. xs=[(1,2),(2,1)]
    init  = lambda i, xs: [i, i+1],
 
    # evaluate at assigned values X_i->x, X_i+1->y for specified xs
    value = lambda x, y, xs: (x,y) in xs)

More complex examples are provided in the accompanying Jupyter notebooks.

◆ dotfile_to_pdf()

infrared.infrared.dotfile_to_pdf	(	graphfile,
		outfile = `None`
	)

Convert dot graph file to pdf.

See also: dotfile_to_tgt

◆ dotfile_to_png()

infrared.infrared.dotfile_to_png	(	graphfile,
		outfile = `None`
	)

Convert dot graph file to png.

See also: dotfile_to_tgt

◆ dotfile_to_tgt()

infrared.infrared.dotfile_to_tgt	(	graphfile,
		tgt,
		outfile = `None`
	)

Convert dot graph file format to png/pdf.

The graph is plotted and written to a pdf or png file by calling graphviz's dot tool on the dot file.

Parameters

graphfile	file of graph in dot format
tgt	target format ('png' or 'pdf')
outfile	optional output filename

Note: Unless specified, the output filename is composed from the input file name and the target format.

◆ mc_optimize()

infrared.infrared.mc_optimize	(	model,
		objective,
		steps,
		temp,
		start = `None`
	)

Optimize by Monte-Carlo optimization with partial resampling.

Maximizes an objective function over assignments of a Infrared model using a Monte-Carlo stochastic optimization strategy with component-wise resampling and Metropolis criterion.

In each iteration, the algorithm resamples connected components of the dependency graph such. This strategy is generally applicable and allows to keep constraints satisfied. Note that this strategy can even benefit from Boltzmann sampling that allows controling the resampling distribution based on functions. Naturally, a high degree of dependencies makes problems hard for this strategy; in the extreme, the strategy degenerates to optimization by iterated sampling.

Parameters

model	Infrared model describing assignments
objective	objective function on assignments
steps	iterations
temp	temperature for the Metropolis criterion
start	optional start assignment

Returns: Pair of best assignment and its objective value

◆ seed()

infrared.infrared.seed ( seed = None )

Seed random number generator of libinfrared and treedecomp.

This seeds the RNG of lib infrared (C++ side) and as well the random number generator used by randomization in the TreeDecomposition, both with the same number. It does NOT seed Python's global RNG.

Parameters

seed	integer used as seed

Note: Without argument or seed==None, use Pythons built-in random.seed() to generate a seed.

Variable Documentation

◆ Optimizer

infrared.infrared.Optimizer = ArcticOptimizer

Alias for default optimizer ArcticOptimizer.

◆ Sampler

infrared.infrared.Sampler = MultiDimensionalBoltzmannSampler

Alias for MultiDimensionalBoltzmannSampler.

Classes

Functions

Variables

Detailed Description

Function Documentation

◆ def_constraint_class()

◆ def_function_class()

◆ dotfile_to_pdf()

◆ dotfile_to_png()

◆ dotfile_to_tgt()

◆ mc_optimize()

◆ seed()

Variable Documentation

◆ Optimizer

◆ Sampler