edu.stanford.nlp.loglinear.model

Class ConcatVectorTable

java.lang.Object

edu.stanford.nlp.loglinear.model.NDArray<java.util.function.Supplier<ConcatVector>>

edu.stanford.nlp.loglinear.model.ConcatVectorTable

All Implemented Interfaces:

java.lang.Iterable<int[]>

public class ConcatVectorTable
extends NDArray<java.util.function.Supplier<ConcatVector>>

Created on 8/9/15.

Author:

keenon

This is basically a type specific wrapper over NDArray

Constructor Summary

Constructors

Constructor and Description
ConcatVectorTable(int[] dimensions) Constructor takes a list of neighbor variables to use for this factor.

Method Summary

Modifier and Type	Method and Description
void	cacheVectors() This is an optimization that will fault all the ConcatVectors into memory, and future .get() on the Supplier objs will result in a very fast return by reference.
ConcatVectorTable	cloneTable() Clones the table, but keeps the values by reference.
ConcatVectorTableProto.ConcatVectorTable.Builder	getProtoBuilder() Returns the proto builder object for this feature factor.
static ConcatVectorTable	readFromProto(ConcatVectorTableProto.ConcatVectorTable proto) Creates a new in-memory feature factor from a proto serialization,
static ConcatVectorTable	readFromStream(java.io.InputStream stream) Convenience function to read a factor (assumed serialized with proto) directly from a stream.
void	releaseCache() This will release references to the cached ConcatVectors created by cacheVectors(), so that they can be cleaned up by the GC.
boolean	valueEquals(ConcatVectorTable other, double tolerance) Deep comparison for equality of value, plus tolerance, for every concatvector in the table, plus dimensional arrangement.
void	writeToStream(java.io.OutputStream stream) Convenience function to write this factor directly to a stream, encoded as proto.

Methods inherited from class edu.stanford.nlp.loglinear.model.NDArray

cloneArray, combinatorialNeighborStatesCount, fastPassByReferenceIterator, getAssignmentValue, getDimensions, iterator, setAssignmentValue

Methods inherited from class java.lang.Object

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

Methods inherited from interface java.lang.Iterable

forEach, spliterator

Constructor Detail

ConcatVectorTable

public ConcatVectorTable(int[] dimensions)

Constructor takes a list of neighbor variables to use for this factor. This must not change after construction, and the number of states of those variables must also not change.

Parameters:

dimensions - list of neighbor variables assignment range sizes

Method Detail

writeToStream

public void writeToStream(java.io.OutputStream stream)
                   throws java.io.IOException

Convenience function to write this factor directly to a stream, encoded as proto. Reversible with readFromStream.

Parameters:

stream - the stream to write to. does not flush automatically

Throws:

java.io.IOException - passed through from the stream

readFromStream

public static ConcatVectorTable readFromStream(java.io.InputStream stream)
                                        throws java.io.IOException

Convenience function to read a factor (assumed serialized with proto) directly from a stream.

Parameters:

stream - the stream to be read from

Returns:

a new in-memory feature factor

Throws:

java.io.IOException - passed through from the stream

getProtoBuilder

public ConcatVectorTableProto.ConcatVectorTable.Builder getProtoBuilder()

Returns the proto builder object for this feature factor. Recursively constructs protos for all the concat vectors in factorTable.

Returns:

proto Builder object

readFromProto

public static ConcatVectorTable readFromProto(ConcatVectorTableProto.ConcatVectorTable proto)

Creates a new in-memory feature factor from a proto serialization,

Parameters:

proto - the proto object to be turned into an in-memory feature factor

Returns:

an in-memory feature factor, complete with in-memory concat vectors

valueEquals

public boolean valueEquals(ConcatVectorTable other,
                           double tolerance)

Deep comparison for equality of value, plus tolerance, for every concatvector in the table, plus dimensional arrangement. This is mostly useful for testing.

Parameters:

other - the vector table to compare against

tolerance - the tolerance to use in value comparisons

Returns:

whether the two tables are equivalent by value

cacheVectors

public void cacheVectors()

This is an optimization that will fault all the ConcatVectors into memory, and future .get() on the Supplier objs will result in a very fast return by reference. Basically this works by wrapping the output of the old thunks inside new, thinner closures that carry around the answer in memory. This is a no-op if vectors were already cached.

releaseCache

public void releaseCache()

This will release references to the cached ConcatVectors created by cacheVectors(), so that they can be cleaned up by the GC. If no cache was constructed, this is a no-op.

cloneTable

public ConcatVectorTable cloneTable()

Clones the table, but keeps the values by reference.

Returns:

a new NDArray, a perfect replica of this one