org.knowceans.topics.cgen
Class MixNetParser

java.lang.Object
  org.knowceans.topics.cgen.MixNetParser

public class MixNetParser
extends java.lang.Object

parse a mixnet script. Example:

 mixnet = LDA
 description:
        test for the mixnet model parser and generator
        
 preferences:
 # produce serial acceleration code (bounds-based sampling)
 fastSerial = false
 # produce parallel acceleration code (shared memory)
 fastParallel = false
 # produce independent samplers for each edge
 indepSamplers = false
 
 # declare nodes (parameters)
 nodes:
 # descriptions and "=" optional
 doc-topic =    theta : M, K | alpha : 1
 # observed nodes, "in" operator means sparse indexing
 doc-author =   am : M, am[m].length in A
 # single label per document
 doc-label =    cm : M, 1 in C
 # option: hyperparameter not estimated
 topic-word =   phi : K, V | beta : 1 : fixed
 topic-word2 =  psi : K, V | gamma : 1
 # constant value C
 topic-char =   xi : K, C = 3 | delta : 1
 # coupled parameters (all statistics joined, hyperparam of first node used)
 connect nodes = xi == phi
 
 # declare sequences with data variable, indexes, range and count
 # sequences. Each edge has a sequence
 words =                w : m, n : M, w[m].length : W
 # labels may have their own sequence (if used as output)
 labels                 c : m, j : M, c[m].length : Wc
 # defines nesting between sequences
 # eg paragraphs nesting words
 pos =          q : m, t in n : M, w[m].length, q[m][n].length : Wq
 
 # declare edges (variables) for sequences
 edges:
 words = w ::
                document =              m : M
                author =                x : A
                label =                 c : C
                topic =                 z : K
                word =                  w : V
 paragraphs      = s ::
        stopic                  y : Y 
 
 # declare network (dependencies)
 # prior alpha and selectors [] optional
 # selectors default to [x,y] of input edges 
 network:
        m >> theta >> z,x
        m,z >> phi[m,z] | alpha[j] >> w
        x >> psi[k] >> w2
  k : { 
        // some Java code, with indices [m,x] and IDLE
        if (x == 0) { k = IDLE; } else { k = [m,x]; } 
  }.
  # k : { ... }. may be on one line, as well.
 
 # special expressions: Java expressions, with { k = [x,y]; } being
 # parent values, linearly indexed. Switching is possible via 
 # { k = IDLE; }, so node is not used for this sample (IDLE = -1). 
 # All "common" Java expressions are permitted (doc for details).
 

Author:

gregor

Field Summary

java.lang.String	comment comments start with # on an otherwise empty line
java.lang.String	edge edges start with a name and have a variable and dimension, they belong to the last sequence declared
(package private) boolean	fastSerial
java.lang.String	file
(package private) boolean	formatCode
java.lang.String	network a network line links one or more incoming and outgoing edges with a node a node has optional alpha and selectors theta[kSel] | alpha[jSel]
java.lang.String	node nodes optional name and dimensions (name =) theta : dim | alpha : dim
java.lang.String	nodecouple C5 parameter coupling structures are specified by theta1 == theta2
java.lang.String	preference preferences are key = value, also used for the title
java.util.HashMap<java.lang.String,java.lang.String>	preferences
java.lang.String	section sections start with a section title and a colon
java.lang.String	selectend finish a selector
java.lang.String	selector Selectors may be set with a variable in kSel, which is defined in a free Java expression starting with a variable and ":$sp{" and ending with "}.", at the end of the first line or on an empty line.
java.lang.String	seqlabel sequence label
java.lang.String	sequence sequences start with a \@name = variable : indexes : dimensions : number of tokens
java.lang.String[]	terminals terminals
java.lang.String	title title follows the word mixnet =
static java.lang.Object	TRUESTRING
(package private) boolean	verbose

Constructor Summary

MixNetParser(java.lang.String file)
create the mixnet parser from the file

Method Summary

boolean	getBoolean(java.lang.String key, boolean standard)
double	getNumber(java.lang.String key, double standard)
java.util.HashMap<java.lang.String,java.lang.String>	getPreferences()
java.lang.String	getString(java.lang.String key, java.lang.String standard)
static void	main(java.lang.String[] args)
MixNet	parse() parse the file and return the mixture network

Methods inherited from class java.lang.Object

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

Field Detail

file

public java.lang.String file

preferences

public java.util.HashMap<java.lang.String,java.lang.String> preferences

verbose

boolean verbose

fastSerial

boolean fastSerial

formatCode

boolean formatCode

TRUESTRING

public static final java.lang.Object TRUESTRING

terminals

public final java.lang.String[] terminals

terminals

title

public final java.lang.String title

title follows the word mixnet =

See Also:

Constant Field Values

comment

public final java.lang.String comment

comments start with # on an otherwise empty line

See Also:

Constant Field Values

section

public final java.lang.String section

sections start with a section title and a colon

See Also:

Constant Field Values

preference

public final java.lang.String preference

preferences are key = value, also used for the title

See Also:

Constant Field Values

node

public final java.lang.String node

nodes optional name and dimensions (name =) theta : dim | alpha : dim

See Also:

Constant Field Values

nodecouple

public final java.lang.String nodecouple

C5 parameter coupling structures are specified by theta1 == theta2

See Also:

Constant Field Values

sequence

public final java.lang.String sequence

sequences start with a \@name = variable : indexes : dimensions : number of tokens

See Also:

Constant Field Values

seqlabel

public final java.lang.String seqlabel

sequence label

See Also:

Constant Field Values

edge

public final java.lang.String edge

edges start with a name and have a variable and dimension, they belong to the last sequence declared

See Also:

Constant Field Values

network

public final java.lang.String network

a network line links one or more incoming and outgoing edges with a node a node has optional alpha and selectors theta[kSel] | alpha[jSel]

See Also:

Constant Field Values

selector

public final java.lang.String selector

Selectors may be set with a variable in kSel, which is defined in a free Java expression starting with a variable and ":$sp{" and ending with "}.", at the end of the first line or on an empty line. The expression must evaluate to the variable name in the node specification. More allowed expressions:

Switch: k : {if (j==1) {k = u;} else if (j==2) {k = v;} else {k = IDLE;}}. selects value of u or v or does not use the input (no counts iterated).

Aggregation: nzz : { nzz = new int[w[m].length]; for (int i = 0; i < w[m].length; i++) { nzz[z[i]]++; }}.

Any other Java expression that evaluates to a value of k or the name of the index variable for the node given in theta[var]. Check for naming collisions. Use in braces with period at the end. In assignments, comma-separated indexes may be used, like k = u,v; as long as u and v are parent edges. k is not an edge but a placeholder for the joint index. It is resolved to a linear component index.

NB: merging is default for an input edge that has several parent nodes.

See Also:

Constant Field Values

selectend

public final java.lang.String selectend

finish a selector

See Also:

Constant Field Values

Constructor Detail

MixNetParser

public MixNetParser(java.lang.String file)

create the mixnet parser from the file

Method Detail

main

public static void main(java.lang.String[] args)
                 throws java.lang.Exception

Throws:

java.lang.Exception

getPreferences

public java.util.HashMap<java.lang.String,java.lang.String> getPreferences()

getString

public java.lang.String getString(java.lang.String key,
                                  java.lang.String standard)

getNumber

public double getNumber(java.lang.String key,
                        double standard)

getBoolean

public boolean getBoolean(java.lang.String key,
                          boolean standard)

parse

public MixNet parse()
             throws java.io.IOException

parse the file and return the mixture network

Returns:

Throws:

java.io.IOException