All Downloads are FREE. Search and download functionalities are using the official Maven repository.

ai.djl.nn.AbstractBlock Maven / Gradle / Ivy

The newest version!
/*
 * Copyright 2019 Amazon.com, Inc. or its affiliates. All Rights Reserved.
 *
 * Licensed under the Apache License, Version 2.0 (the "License"). You may not use this file except in compliance
 * with the License. A copy of the License is located at
 *
 * http://aws.amazon.com/apache2.0/
 *
 * or in the "license" file accompanying this file. This file is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES
 * OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions
 * and limitations under the License.
 */
package ai.djl.nn;

import ai.djl.ndarray.NDArray;
import ai.djl.ndarray.NDList;
import ai.djl.ndarray.NDManager;
import ai.djl.ndarray.types.DataType;
import ai.djl.ndarray.types.Shape;
import ai.djl.training.ParameterStore;
import ai.djl.util.Pair;
import ai.djl.util.PairList;

import java.io.DataOutputStream;
import java.util.LinkedHashMap;
import java.util.Locale;
import java.util.function.Function;

/**
 * {@code AbstractBlock} is an abstract implementation of {@link Block}.
 *
 * 

It is recommended that all {@code Block} classes that have children extend the {@code * AbstractBlock}. * *

To create your own blocks, you need to do the following: * *

    *
  • Define a version for serializing parameter and metadata and pass it to the parent * constructor *
  • Use {@link AbstractBlock#addParameter(Parameter)} to add parameters to your block in the * constructor if necessary. *
  • Use {@link AbstractBlock#addChildBlock(String, Block)} to add child blocks if necessary. *
  • Override {@link Block#getOutputShapes(Shape[])} to determine the shape of your custom * block's output based on the input it will receive. *
  • Override {@link AbstractBlock#initializeChildBlocks(NDManager, DataType, Shape...)} if you * added child blocks to initialize them based on the input shape your block will receive. You * can skip this if your block does not contain child blocks *
  • Override {@link AbstractBlock#forward(ParameterStore, NDList, boolean, PairList)} to * implement the computation of your block *
  • IFF you need to save data apart from the parameter values of your block, you need to * override {@link AbstractBlock#saveMetadata(DataOutputStream)} and {@link * AbstractBlock#loadMetadata(byte, java.io.DataInputStream)}. If you do not need to save or * load any state other than parameters in your block, you can skip this. *
* *

If you use {@link AbstractBlock#addParameter(Parameter)} to add parameters, you have to take * care of parameter initialization yourself. In this case, you need to setShape to your parameters * if you know the shape of Parameter or you can implement prepare to setShape when you see the * input shape. */ // Using LinkedHashMap instead of Map is intentional: we want to make sure that consumers // of this API know the children and parameters are always iterated over in insertion order. // LinkedHashMap provides this guarantee, Map does not. @SuppressWarnings("PMD.LooseCoupling") public abstract class AbstractBlock extends AbstractBaseBlock { /** * All direct children of this Block. Keys are names of the blocks. * *

Use the {@link AbstractBlock#addChildBlock(String, Block)} method to add children. All * children in this map are automagically loaded / saved. */ protected BlockList children = new BlockList(); /** * All direct parameters of this Block. Keys are name of the parameters. * *

Use the {@link AbstractBlock#addParameter(Parameter)} method to add children. All * parameters in this map are automatically loaded / saved. */ protected LinkedHashMap parameters = new LinkedHashMap<>(); /** Constructs a new {@code AbstractBlock} instance. */ public AbstractBlock() {} /** * Builds an empty block with the given version for parameter serialization. * * @param version the version to use for parameter serialization. */ public AbstractBlock(byte version) { super(version); } /** * Use this to add a child block to this block. * * @param name Name of the block, must not be null. * @param block The block, must not be null. * @param The type of block * @return the block given as a parameter - that way the block can be created and reassigned to * a member variable more easily. */ protected final B addChildBlock(String name, B block) { int childNumber = children.size() + 1; children.add(String.format(Locale.ROOT, "%02d%s", childNumber, name), block); return block; } /** * Adds a {@link LambdaBlock} as a child block to this block. * * @param name Name of the block, must not be null. * @param f the function forms the {@link LambdaBlock} * @return the child block */ protected LambdaBlock addChildBlock(String name, Function f) { return addChildBlock(name, new LambdaBlock(f, name)); } /** * Adds a {@link LambdaBlock#singleton(Function)} as a child block to this block. * * @param name Name of the block, must not be null. * @param f the function forms the {@link LambdaBlock} * @return the child block * @see LambdaBlock#singleton(Function) */ protected final LambdaBlock addChildBlockSingleton(String name, Function f) { return addChildBlock(name, LambdaBlock.singleton(f, name)); } /** * Adds a parameter to this block. If parameters are added with this method, initialization of * the parameter works out of the box * * @param

the specific parameter subclass * @param parameter the parameter to add, not null * @return the parameter passed as arguments to make it easier to create and assign parameters * in one line */ protected final

P addParameter(P parameter) { parameters.put(parameter.getName(), parameter); return parameter; } /** {@inheritDoc} */ @Override public BlockList getChildren() { BlockList defensiveCopy = new BlockList(children.size()); for (Pair entry : children) { defensiveCopy.add(entry); } return defensiveCopy; } /** {@inheritDoc} */ @Override public ParameterList getDirectParameters() { return new ParameterList(parameters); } }