Package io.github.kirstenali.deepj.optimisers

Class ParameterFactory

java.lang.Object

io.github.kirstenali.deepj.optimisers.ParameterFactory

public final class ParameterFactory
extends Object

Explicit parameter storage policy: a requested dtype applies to multi-row parameters; one-row parameters remain FLOAT32. Descriptor-based loading preserves each stored dtype. Names do not depend on allocation order.

Constructor Summary

Constructors

Constructor	Description
ParameterFactory(DType dtype)
ParameterFactory(Map<String,TensorDescriptor> descriptors)

Method Summary

Modifier and Type	Method	Description
static ParameterFactory	float32()
ParameterFactory	forLoading()	Zero-filled random-weight slots for a complete checkpoint replacement, without device initialization.
Parameter	ones(String name, int rows, int columns)
Parameter	random(String name, int rows, int columns, Random random)
ParameterFactory	scope(String name)
Parameter	zeros(String name, int rows, int columns)

Methods inherited from class java.lang.Object

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

Constructor Details

ParameterFactory

public ParameterFactory(DType dtype)

ParameterFactory

public ParameterFactory(Map<String,TensorDescriptor> descriptors)

Method Details

float32

public static ParameterFactory float32()

forLoading

public ParameterFactory forLoading()

Zero-filled random-weight slots for a complete checkpoint replacement, without device initialization.

scope

public ParameterFactory scope(String name)

random

public Parameter random(String name,
 int rows,
 int columns,
 Random random)

zeros

public Parameter zeros(String name,
 int rows,
 int columns)

ones

public Parameter ones(String name,
 int rows,
 int columns)