Package ai.djl.nn
package ai.djl.nn
Contains classes to construct neural networks.
The primary construct used to build up the networks is the Block (see for
details). This package contains a number of implementations of blocks as well as helpers for
blocks.
The following subpackages also contain a number of standard neural network operations to use with blocks:
-
ClassDescriptionThis provides shared functionality for both the DJL-based
AbstractBlocks and the importedAbstractSymbolBlocks.AbstractBlockis an abstract implementation ofBlock.AbstractSymbolBlockis an abstract implementation ofSymbolBlock.Utility class that provides activation functions and blocks.ABlockis a composable function that forms a neural network.Block factory is a component to make standard for block creating and saving procedure.Represents a set of names and Blocks.Utility class that provides some useful blocks.ABlockFactoryclass that creates IdentityBlock.LambdaBlockis aBlockwith no parameters or children.ABlockFactoryclass that creates LambdaBlock.ParallelBlockis aBlockwhose children form a parallel branch in the network and are combined to produce a single output.Parameteris a container class that holds a learnable parameter of a model.A Builder to construct aParameter.Enumerates the types ofParameter.Represents a set of names and Parameters.SequentialBlockis aBlockwhose children form a chain of blocks with each child block feeding its output to the next.SymbolBlockis aBlockis used to load models that were exported directly from the engine in its native format.Thrown to indicate that aParameterwas not initialized.