org.knowceans.topics.cgen
Class MixNode

java.lang.Object
  org.knowceans.topics.cgen.MixItem
      org.knowceans.topics.cgen.MixNode

Direct Known Subclasses:

MixNodeFastSerial

public class MixNode
extends MixItem

represents a mixture node.

TODO: this may be subclassed for node data types

Field Summary

(package private) java.lang.String	alpha hyperparameter name
static int	C1ASINGLE C1A single input, non-root edge (e.g., LDA: phi[k])
static int	C1BSEQADD C1B single input added sequence (e.g., PAM: theta[m,x])
static int	C1INPUT any C1 input
static int	C1ROOT single input, root edge (e.g., LDA: theta[m]), Note: this is different from SEQUENCE, where a single C1 sequence input among many leads to the classification)
static int	C2MULTI C2 multiple non-sequence input edges (w/kSel, e.g., Bi-LDA: phi[u,m])
static int	C3MERGED C3 merged component C3 (NOT using kSel, novel structure)
static int	C4SWITCH C4 switched input C4 (using kSel, e.g., Multi-grain LDA)
static int	C5COUPLED C5 coupled node C5 (using coupledNodes)
static int	C5MAIN C5 main node: all parameters and hyperparameters in coupledNodes are deferred to this node
(package private) java.lang.String	constK constant value for K
(package private) MixNode	coupledMainNode main node in a coupled group
(package private) java.util.List<MixNode>	coupledNodes nodes this shares parameters with (only one for now)
(package private) int	dimAlpha dimension of alpha: 0..2
static int	E2SPLIT E2 multiple output sequences
(package private) boolean	fixedAlpha don't estimate alpha
static int	HIDDEN summarising hidden structures
(package private) java.lang.String	j component group name
java.lang.String	J component group index range (elements of alpha)
(package private) java.lang.String	jSel component group selector (Java expression as in j = )
(package private) java.lang.String	k component name
(package private) java.lang.String	K component count = domain (= range of kSel over all inputs)
(package private) java.lang.String	kSel component selector (Java expression as in k = )
static int	N6AGGREG N6 aggregation node
static java.util.List<MixEdge>	NOEDGE represents the empty edge set
(package private) java.lang.String	ntheta count name
(package private) java.lang.String	nthetasum sum count name
static int	SEQINPUT any sequence input: state by no parameters at query-time
static int	SEQUENCE sequence node (will generate query-time structure)
(package private) boolean	sparse sparse representation of visible nodes: by a label sequence
(package private) java.lang.String	T outgoing edge count = T of first edge
(package private) java.lang.String	theta parameter name
static int	TOPIC topic node (global scope)
static int	VISIBLE is this node observed (set a priori)

Fields inherited from class org.knowceans.topics.cgen.MixItem

children, datatype, indent, linktype, name, parents, UNRECOG

Constructor Summary

MixNode(java.lang.String name, java.lang.String theta, boolean sparse)
create a visible node

MixNode(java.lang.String name, java.lang.String theta, java.lang.String alpha)
constructs a hidden node

Method Summary

java.lang.String	assignVarsInput() assign input to the variables of this edge
boolean	check(java.io.PrintStream out) check node consistency
boolean	classify() classify the data and linking types of this node.
void	coupleNode(MixNode node) couple the node with the argument.
void	coupleParams() couple the parameters of coupled nodes
java.lang.String	declareVars() declare node structures
java.lang.String	expandSelector(MixSequence seq, boolean theta, boolean alpha, java.util.List<MixEdge> ee, boolean q) expand selector code, replacing indices appropriately.
java.lang.String	getAlphaEstimator() generate a hyperparameter estimator for this node, based on DirichletEstimation methods
MixEdge	getChildEdge(int i)
MixEdge	getChildEdge(java.lang.String x)
java.util.List<MixEdge>	getChildren()
java.lang.String	getComponentCount() get the count of the components according to the kSel expression.
java.util.List<MixNode>	getCoupledNodes() get couple nodes as a shortcut to resolving a C3 structure.
MixEdge	getEdge(java.lang.String x) get the edge from its variable.
java.lang.String	getGibbsTerm(MixSequence seq, java.util.List<MixEdge> ee, boolean q)
java.lang.String	getLatexGibbsTerm(MixSequence seq, java.util.List<MixEdge> ee, boolean q, boolean mathml) latex description of this node's inference term
java.lang.String	getLikelihoodTerm(MixSequence seq, java.util.List<MixEdge> ee) get the likelihood term for this node
java.lang.String	getName() short name of this node
MixEdge	getParentEdge(int i)
MixEdge	getParentEdge(java.lang.String x)
java.util.List<MixEdge>	getParents()
java.lang.String	getThetaEstimator(boolean q) generate a parameter estimator, based on the Dirichlet posterior mean
java.lang.String	getVariable()
java.lang.String	incNode(MixSequence seq, java.util.List<MixEdge> ee, boolean inc, boolean q, boolean synced) return code to increment this node's counts
java.lang.String	initVars(boolean q) initialise this node
int	intersection(java.util.List<MixEdge> eeSamp) checks whether the set of edges intersects with the node, i.e., whether one of its elements is a parent or child edge of this node
(package private) java.lang.String	kIndex(MixSequence seq, java.util.List<MixEdge> ee) return the index that expandIndex assigns a value to, so the result of expandIndex is directly used in computations.
java.lang.String	listTypedVarsInput(java.util.Set<java.lang.String> unique) list the variables for this node comma-separated, adding to the set unique and ignoring variables already in it.
java.lang.String	listVarsInput(java.util.Set<java.lang.String> unique) list the variables for this node comma-separated, adding to the set unique and ignoring variables already in it.
void	nameEmptyVars() complete the selector and index structures after the network has been connected and classified
java.lang.String	printDetails()
java.lang.String	printSelector() print selector string (may be large)
java.lang.String	printType() print the type of this node
void	setConstants(java.util.Map<java.lang.String,java.lang.String> constants) set any constant for this node
(package private) java.lang.String[]	tIndex(MixSequence seq, java.util.List<MixEdge> ee, boolean forQuery) determine the index of the output edge, given the active edges ee (qualified by h) and the current sequence
java.lang.String	toString()

Methods inherited from class org.knowceans.topics.cgen.MixItem

typeString

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Field Detail

VISIBLE

public static final int VISIBLE

is this node observed (set a priori)

See Also:

Constant Field Values

SEQUENCE

public static final int SEQUENCE

sequence node (will generate query-time structure)

See Also:

Constant Field Values

TOPIC

public static final int TOPIC

topic node (global scope)

See Also:

Constant Field Values

HIDDEN

public static final int HIDDEN

summarising hidden structures

See Also:

Constant Field Values

N6AGGREG

public static final int N6AGGREG

N6 aggregation node

See Also:

Constant Field Values

C1ROOT

public static final int C1ROOT

single input, root edge (e.g., LDA: theta[m]), Note: this is different from SEQUENCE, where a single C1 sequence input among many leads to the classification)

See Also:

Constant Field Values

C1ASINGLE

public static final int C1ASINGLE

C1A single input, non-root edge (e.g., LDA: phi[k])

See Also:

Constant Field Values

C1BSEQADD

public static final int C1BSEQADD

C1B single input added sequence (e.g., PAM: theta[m,x])

See Also:

Constant Field Values

C2MULTI

public static final int C2MULTI

C2 multiple non-sequence input edges (w/kSel, e.g., Bi-LDA: phi[u,m])

See Also:

Constant Field Values

C3MERGED

public static final int C3MERGED

C3 merged component C3 (NOT using kSel, novel structure)

See Also:

Constant Field Values

C4SWITCH

public static final int C4SWITCH

C4 switched input C4 (using kSel, e.g., Multi-grain LDA)

See Also:

Constant Field Values

C5COUPLED

public static final int C5COUPLED

C5 coupled node C5 (using coupledNodes)

See Also:

Constant Field Values

C5MAIN

public static final int C5MAIN

C5 main node: all parameters and hyperparameters in coupledNodes are deferred to this node

See Also:

Constant Field Values

SEQINPUT

public static final int SEQINPUT

any sequence input: state by no parameters at query-time

See Also:

Constant Field Values

C1INPUT

public static final int C1INPUT

any C1 input

See Also:

Constant Field Values

E2SPLIT

public static final int E2SPLIT

E2 multiple output sequences

See Also:

Constant Field Values

theta

java.lang.String theta

parameter name

ntheta

java.lang.String ntheta

count name

nthetasum

java.lang.String nthetasum

sum count name

k

java.lang.String k

component name

K

java.lang.String K

component count = domain (= range of kSel over all inputs)

constK

java.lang.String constK

constant value for K

T

java.lang.String T

outgoing edge count = T of first edge

alpha

java.lang.String alpha

hyperparameter name

dimAlpha

int dimAlpha

dimension of alpha: 0..2

fixedAlpha

boolean fixedAlpha

don't estimate alpha

j

java.lang.String j

component group name

J

public java.lang.String J

component group index range (elements of alpha)

kSel

java.lang.String kSel

component selector (Java expression as in k = )

jSel

java.lang.String jSel

component group selector (Java expression as in j = )

NOEDGE

public static java.util.List<MixEdge> NOEDGE

represents the empty edge set

coupledNodes

java.util.List<MixNode> coupledNodes

nodes this shares parameters with (only one for now)

coupledMainNode

MixNode coupledMainNode

main node in a coupled group

sparse

boolean sparse

sparse representation of visible nodes: by a label sequence

Constructor Detail

MixNode

public MixNode(java.lang.String name,
               java.lang.String theta,
               java.lang.String alpha)

constructs a hidden node

Parameters:

name -

theta -

alpha -

MixNode

public MixNode(java.lang.String name,
               java.lang.String theta,
               boolean sparse)

create a visible node

Parameters:

name -

theta -

sparse - whether it is connected with a label sequence (like authors) that associates only the non-null subset of the domain

Method Detail

coupleNode

public void coupleNode(MixNode node)

couple the node with the argument. For a group of coupled nodes, there is one main node. If node is the main node of a group, the complete group is added to this node.

Parameters:

node -

classify

public boolean classify()

classify the data and linking types of this node. Assumes that edges have been classified already.

Specified by:

classify in class MixItem

Returns:

false if invalid

nameEmptyVars

public void nameEmptyVars()

complete the selector and index structures after the network has been connected and classified

coupleParams

public void coupleParams()

couple the parameters of coupled nodes

intersection

public int intersection(java.util.List<MixEdge> eeSamp)

checks whether the set of edges intersects with the node, i.e., whether one of its elements is a parent or child edge of this node

Parameters:

n -

eeSamp -

Returns:

0 for no intersection, 1 for parent intersection, 2 for child intersection

getCoupledNodes

public java.util.List<MixNode> getCoupledNodes()

get couple nodes as a shortcut to resolving a C3 structure.

Returns:

getChildEdge

public MixEdge getChildEdge(int i)

getChildEdge

public MixEdge getChildEdge(java.lang.String x)

getParentEdge

public MixEdge getParentEdge(int i)

getParentEdge

public MixEdge getParentEdge(java.lang.String x)

getParents

public java.util.List<MixEdge> getParents()

Overrides:

getParents in class MixItem

getChildren

public java.util.List<MixEdge> getChildren()

Overrides:

getChildren in class MixItem

getComponentCount

public java.lang.String getComponentCount()

get the count of the components according to the kSel expression. This assumes that each dimension is fully used, which (for now) excludes models like hPAM that linearise a hierarchical structure.

Alternatively, an expression in K may be used.

Returns:

getEdge

public MixEdge getEdge(java.lang.String x)

get the edge from its variable. If two such variables exist, the first one found is returned. However, variable names should be unique.

Parameters:

x -

Returns:

edge or null if it does not exist

toString

public java.lang.String toString()

Overrides:

toString in class java.lang.Object

printSelector

public java.lang.String printSelector()

print selector string (may be large)

Returns:

printType

public java.lang.String printType()

Description copied from class: MixItem

print the type of this node

Specified by:

printType in class MixItem

Returns:

printDetails

public java.lang.String printDetails()

Overrides:

printDetails in class MixItem

getVariable

public java.lang.String getVariable()

Specified by:

getVariable in class MixItem

setConstants

public void setConstants(java.util.Map<java.lang.String,java.lang.String> constants)

set any constant for this node

Overrides:

setConstants in class MixItem

check

public boolean check(java.io.PrintStream out)

check node consistency

Overrides:

check in class MixItem

Returns:

getName

public java.lang.String getName()

short name of this node

Specified by:

getName in class MixItem

Returns:

listVarsInput

public java.lang.String listVarsInput(java.util.Set<java.lang.String> unique)

list the variables for this node comma-separated, adding to the set unique and ignoring variables already in it.

Specified by:

listVarsInput in class MixItem

Returns:

listTypedVarsInput

public java.lang.String listTypedVarsInput(java.util.Set<java.lang.String> unique)

list the variables for this node comma-separated, adding to the set unique and ignoring variables already in it.

Specified by:

listTypedVarsInput in class MixItem

Returns:

assignVarsInput

public java.lang.String assignVarsInput()

assign input to the variables of this edge

Specified by:

assignVarsInput in class MixItem

Returns:

declareVars

public java.lang.String declareVars()

declare node structures

Specified by:

declareVars in class MixItem

Parameters:

u -

n -

Returns:

initVars

public java.lang.String initVars(boolean q)

initialise this node

Specified by:

initVars in class MixItem

Parameters:

q -

Returns:

expandSelector

public java.lang.String expandSelector(MixSequence seq,
                                       boolean theta,
                                       boolean alpha,
                                       java.util.List<MixEdge> ee,
                                       boolean q)

expand selector code, replacing indices appropriately. Currently, this understands comma notation and expands all parent variables as hx or x[m][n] according to the filter set ee.

TODO: allow complex statements in brackets

Parameters:

theta - expand for theta: k = kSel

alpha - expand for alpha: j = jSel

q -

Returns:

a Java expression that creates node.k or

kIndex

java.lang.String kIndex(MixSequence seq,
                        java.util.List<MixEdge> ee)

return the index that expandIndex assigns a value to, so the result of expandIndex is directly used in computations.

Parameters:

seq -

ee -

forQuery -

Returns:

tIndex

java.lang.String[] tIndex(MixSequence seq,
                          java.util.List<MixEdge> ee,
                          boolean forQuery)

determine the index of the output edge, given the active edges ee (qualified by h) and the current sequence

Parameters:

seq -

ee -

forQuery -

Returns:

getGibbsTerm

public java.lang.String getGibbsTerm(MixSequence seq,
                                     java.util.List<MixEdge> ee,
                                     boolean q)

getLatexGibbsTerm

public java.lang.String getLatexGibbsTerm(MixSequence seq,
                                          java.util.List<MixEdge> ee,
                                          boolean q,
                                          boolean mathml)

latex description of this node's inference term

Parameters:

seq -

ee -

q -

mathml - (or latex)

Returns:

incNode

public java.lang.String incNode(MixSequence seq,
                                java.util.List<MixEdge> ee,
                                boolean inc,
                                boolean q,
                                boolean synced)

return code to increment this node's counts

Parameters:

ee - active edges

inc - whether to increment or decrement

q - for querying

synced - use synchronisation

Returns:

getLikelihoodTerm

public java.lang.String getLikelihoodTerm(MixSequence seq,
                                          java.util.List<MixEdge> ee)

get the likelihood term for this node

Parameters:

seq -

ee -

Returns:

getThetaEstimator

public java.lang.String getThetaEstimator(boolean q)

generate a parameter estimator, based on the Dirichlet posterior mean

getAlphaEstimator

public java.lang.String getAlphaEstimator()

generate a hyperparameter estimator for this node, based on DirichletEstimation methods

Returns: