Package edu.cmu.tetrad.search

Class LvLite

java.lang.Object

edu.cmu.tetrad.search.LvLite

All Implemented Interfaces:

IGraphSearch

public final class LvLite
extends Object
implements IGraphSearch

The LV-Lite algorithm (Latent Variable "Lite") algorithm implements a search algorithm for learning the structure of a graphical model from observational data with latent variables. The algorithm uses the BOSS or GRaSP algorithm to get an initial CPDAG. Then it uses scoring steps to infer some unshielded colliders in the graph, then finishes with a testing step to remove extra edges and orient more unshielded colliders. Finally, the final FCI orientation is applied to the graph.

Author:

josephramsey

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
static enum	LvLite.ExtraEdgeRemovalStyle	The ExtraEdgeRemovalStyle enum specifies the styles for removing extra edges.
static enum	LvLite.START_WITH	Enumeration representing different start options.

Constructor Summary

Constructors

Constructor	Description
LvLite(Graph cpdag, IndependenceTest test)	Alternative LV-Lite constructor.
LvLite(IndependenceTest test, Score score)	LV-Lite constructor.

Method Summary

Modifier and Type	Method	Description
Graph	search()	Run the search and return s a PAG.
void	setCompleteRuleSetUsed(boolean completeRuleSetUsed)	Sets whether the complete rule set should be used during the search algorithm.
void	setDepth(int depth)	Sets the maximum size of the separating set used in the graph search algorithm.
void	setDoDdpEdgeRemovalStep(boolean doDdpEdgeRemovalStep)	Sets whether to perform DDP edge removal step.
void	setEnsureMarkov(boolean ensureMarkov)	Sets the value indicating whether the process should ensure Markov property.
void	setExtraEdgeRemovalStyle(LvLite.ExtraEdgeRemovalStyle extraEdgeRemovalStyle)	Sets the style for removing extra edges.
void	setGuaranteePag(boolean guaranteePag)	Sets whether to guarantee a PAG output by repairing a faulty PAG.
void	setKnowledge(Knowledge knowledge)	Sets the knowledge used in search.
void	setMaxBlockingPathLength(int maxBlockingPathLength)	Sets the maximum length of any discriminating path.
void	setMaxDdpPathLength(int maxDdpPathLength)	Sets the maximum DDP path length.
void	setNumStarts(int numStarts)	Sets the number of starts for BOSS.
void	setRecursionDepth(int recursionDepth)	Sets the depth of the GRaSP if it is used.
void	setStartWith(LvLite.START_WITH startWith)	Sets the algorithm to use to obtain the initial CPDAG.
void	setTestTimeout(long testTimeout)	Sets the timeout for the testing steps, for the extra edge removal steps and the discriminating path steps.
void	setUseBes(boolean useBes)	Sets whether to use the BES (Backward Elimination Search) algorithm during the search.
void	setUseDataOrder(boolean useDataOrder)	Sets the flag indicating whether to use data order.
void	setVerbose(boolean verbose)	Sets the verbosity level of the search algorithm.

Methods inherited from class java.lang.Object

equals, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Details

LvLite

public LvLite(IndependenceTest test,
 Score score)

LV-Lite constructor. Initializes a new object of LV-Lite search algorithm with the given IndependenceTest and Score object.

In this constructor, we will use BOSS or GRaSP internally to infer an initial CPDAG and valid order of the variables. This is the default behavior of the LV-Lite algorithm.

Parameters:

test - The IndependenceTest object to be used for testing independence between variables.

score - The Score object to be used for scoring DAGs.

Throws:

NullPointerException - if the score is null.

LvLite

public LvLite(Graph cpdag,
 IndependenceTest test)

Alternative LV-Lite constructor. Initializes a new object of LV-Lite search algorithm with the given initial CPDAG, along with the IndependenceTest. These should all be over variables with the same names as the variables in the supplied test.

This constructor allows the user to employ an external algorithm to find this initial CPDAG (and an implied order of the variables). This is useful when the user has a preferred algorithm for this task. In this case, the algorithm will perform only the steps after the CPDAG initialization step.

It is important here that the CPDAG be obtained from an algorithm like BOSS or GRaSP that yields a high-quality DAG or CPDAG over the variables, with a valid order, under the (possibly false) assumption that the data model is causally sufficient. The CPDAG is used to establish the initial structure of the estimated PAG and to copy unshielded colliders from the CPDAG into the estimated PAG. This step is justified in the GFCI algorithm. Ogarrio, J. M., Spirtes, P., & Ramsey, J. (2016, August). A hybrid causal search algorithm for latent variable models. In Conference on probabilistic graphical models (pp. 368-379). PMLR. The idea is we start with this initial estimated of the PAG, with edges reoriented as o-o edges. Then we use the scorer to copy unshielded colliders from the CPDAG into the estimated PAG, as an initial step of converting the initial CPDAG into a PAG.

Parameters:

cpdag - The initial CPDAG.

test - The independence test.

Method Details

search

public Graph search()
             throws InterruptedException

Run the search and return s a PAG.

Specified by:

search in interface IGraphSearch

Returns:

The PAG.

Throws:

InterruptedException - if any.

setMaxBlockingPathLength

public void setMaxBlockingPathLength(int maxBlockingPathLength)

Sets the maximum length of any discriminating path.

Parameters:

maxBlockingPathLength - the maximum length of any discriminating path, or -1 if unlimited.

setRecursionDepth

public void setRecursionDepth(int recursionDepth)

Sets the depth of the GRaSP if it is used.

Parameters:

recursionDepth - The depth of the GRaSP.

setGuaranteePag

public void setGuaranteePag(boolean guaranteePag)

Sets whether to guarantee a PAG output by repairing a faulty PAG.

Parameters:

guaranteePag - true if a faulty PAGs should be repaired, false otherwise

setStartWith

public void setStartWith(LvLite.START_WITH startWith)

Sets the algorithm to use to obtain the initial CPDAG.

Parameters:

startWith - the algorithm to use to obtain the initial CPDAG.

setKnowledge

public void setKnowledge(Knowledge knowledge)

Sets the knowledge used in search.

Parameters:

knowledge - This knowledge.

setCompleteRuleSetUsed

public void setCompleteRuleSetUsed(boolean completeRuleSetUsed)

Sets whether the complete rule set should be used during the search algorithm. By default, the complete rule set is not used.

Parameters:

completeRuleSetUsed - true if the complete rule set should be used, false otherwise

setVerbose

public void setVerbose(boolean verbose)

Sets the verbosity level of the search algorithm.

Parameters:

verbose - true to enable verbose mode, false to disable it

setNumStarts

public void setNumStarts(int numStarts)

Sets the number of starts for BOSS.

Parameters:

numStarts - The number of starts.

setUseBes

public void setUseBes(boolean useBes)

Sets whether to use the BES (Backward Elimination Search) algorithm during the search.

Parameters:

useBes - true to use the BES algorithm, false otherwise

setUseDataOrder

public void setUseDataOrder(boolean useDataOrder)

Sets the flag indicating whether to use data order.

Parameters:

useDataOrder - true if the data order should be used, false otherwise.

setDepth

public void setDepth(int depth)

Sets the maximum size of the separating set used in the graph search algorithm.

Parameters:

depth - the maximum size of the separating set

setMaxDdpPathLength

public void setMaxDdpPathLength(int maxDdpPathLength)

Sets the maximum DDP path length.

Parameters:

maxDdpPathLength - the maximum DDP path length to set

setExtraEdgeRemovalStyle

public void setExtraEdgeRemovalStyle(LvLite.ExtraEdgeRemovalStyle extraEdgeRemovalStyle)

Sets the style for removing extra edges.

Parameters:

extraEdgeRemovalStyle - the style for removing extra edges

setTestTimeout

public void setTestTimeout(long testTimeout)

Sets the timeout for the testing steps, for the extra edge removal steps and the discriminating path steps.

Parameters:

testTimeout - the timeout for the testing steps, for the extra edge removal steps and the discriminating path steps.

setDoDdpEdgeRemovalStep

public void setDoDdpEdgeRemovalStep(boolean doDdpEdgeRemovalStep)

Sets whether to perform DDP edge removal step. This step may be computationally expensive for large models and addresses a relatively rare condition in the search space. By default the step is done, since it is needed for correctness.

Parameters:

doDdpEdgeRemovalStep - a boolean indicating if the DDP edge removal step should be executed

setEnsureMarkov

public void setEnsureMarkov(boolean ensureMarkov)

Sets the value indicating whether the process should ensure Markov property.

Parameters:

ensureMarkov - a boolean value, true to ensure Markov property, false otherwise.