自己打造一个深度学习框架 for java

前言

从16年开始利用空余时间研究深度学习的方面，由于工作的原因，最熟悉的编程语言就是java，所以框架的编程语言自然而然就使用了java。自己打造框架的初衷就是为了更加深入了解各个算法、模型、实现的原理和思路。

框架介绍

Omega-AI：基于java打造的深度学习框架，帮助你快速搭建神经网络，实现训练或测试模型，支持多线程运算，框架目前支持BP神经网络和卷积神经网络的构建。

源码地址：

https://gitee.com/iangellove/omega-ai

https://github.com/iangellove/Omega-AI

依赖

由于omega-engine-1.0.3加入了jcuda支持，所以1.0.3需要安装与jcuda版本对应的cuda，我在该项目中使用的是jcuda-11.2.0版本的包，那么我cuda需要安装11.2.x版本

系统参数

由于训练vgg16模型的参数比较庞大，所以在部署项目的时候需要对jvm内存进行调整. 调整事例如：-Xmx20480m -Xms20480m -Xmn10240m

Demo展示

基于卷积神经网络mnist手写数字识别在这里插入图片描述

功能介绍

支持的网络层类型：

Fullylayer 全连接层

ConvolutionLayer 卷积层

PoolingLayer 池化层

激活函数层

SoftmaxLayer (softmax激活函)

ReluLayer

LeakyReluLayer

TanhLayer

SigmodLayer

归一化层

BNLayer (Batch Normalization)

DropoutLayer

优化器

Momentum

Adam

训练器

BGDOptimizer (批量梯度下降法)

MBSGDOptimizer (小批量随机梯度下降)

SGDOptimizer（随机梯度下降算法）

损失函数(loss function)

SquareLoss (平方差损失函数)

CrossEntropyLoss (交叉熵损失函数)

学习率更新器（LearnRateUpdate）

NONE (固定学习率)

LR_DECAY (decay)

GD_GECAY (gd_decay)

CONSTANT(gd_decay)

RANDOM [Math.pow(RandomUtils.getInstance().nextFloat(), power) * this.lr]

POLY [this.lr * Math.pow((1.0f - (batchIndex * 1.0f / trainTime / dataSize * batchSize)), power)]

STEP [this.lr * Math.pow(this.scale, batchIndex / step)]

EXP [this.lr * Math.pow(this.gama, batchIndex)]

SIG [this.lr / (1 + Math.pow(Math.E, this.gama * (batchIndex - step)))]

数据加载器

.bin (二进制数据文件)

.idx3-ubyte

.txt

使用说明

自带的数据集

iris（鸢尾花数据集）

mnist（手写数字数据集）

cifat_10 （cifat_10数据集）

数据集成绩

iris 训练次数8 bp神经网络[3层全连接层] 测试数据集准确率100%

mnist 训练次数10 alexnet 测试数据集准确率98.6%

cifat_10 训练次数10 alexnet 测试数据集准确率76.6%

cifat_10 训练次数10 vgg16 测试数据集准确率86.45%

cifat_10 训练次数10 resnet18 [batchSize:128,learningRate:0.001] 测试数据集准确率93.22%

事例代码

bp iris demo

public void bpNetwork_iris() {
		// TODO Auto-generated method stub

		/**
		 * 读取训练数据集
		 */
		String iris_train = "/dataset/iris/iris.txt";
		
		String iris_test = "/dataset/iris/iris_test.txt";
		
		String[] labelSet = new String[] {"1","-1"};
		
		DataSet trainData = DataLoader.loalDataByTxt(iris_train, ",", 1, 1, 4, 2,labelSet);
		DataSet testData = DataLoader.loalDataByTxt(iris_test, ",", 1, 1, 4, 2,labelSet);
		
		System.out.println("train_data:"+JsonUtils.toJson(trainData));
	
		BPNetwork netWork = new BPNetwork(new SoftmaxWithCrossEntropyLoss());
		
		InputLayer inputLayer = new InputLayer(1,1,4);
		
		FullyLayer hidden1 = new FullyLayer(4, 40);
		
		ReluLayer active1 = new ReluLayer();
		
		FullyLayer hidden2 = new FullyLayer(40, 20);
		
		ReluLayer active2 = new ReluLayer();
		
		FullyLayer hidden3 = new FullyLayer(20, 2);

		SoftmaxWithCrossEntropyLayer hidden4 = new SoftmaxWithCrossEntropyLayer(2);
		
		netWork.addLayer(inputLayer);
		netWork.addLayer(hidden1);
		netWork.addLayer(active1);
		netWork.addLayer(hidden2);
		netWork.addLayer(active2);
		netWork.addLayer(hidden3);
		netWork.addLayer(hidden4);

		try {
			
			MBSGDOptimizer optimizer = new MBSGDOptimizer(netWork, 8, 0.00001d, 10, LearnRateUpdate.NONE);
		
			optimizer.train(trainData);
			
			optimizer.test(testData);
			
		} catch (Exception e) {
			// TODO Auto-generated catch block
			e.printStackTrace();
		}

	}

cnn mnist demo

public void cnnNetwork_mnist() {
		// TODO Auto-generated method stub
		
		try {

			/**
			 * 读取训练数据集
			 */
			String mnist_train_data = "/dataset/mnist/train-images.idx3-ubyte";
			
			String mnist_train_label = "/dataset/mnist/train-labels.idx1-ubyte";
			
			String mnist_test_data = "/dataset/mnist/t10k-images.idx3-ubyte";
			
			String mnist_test_label = "/dataset/mnist/t10k-labels.idx1-ubyte";
			
			String[] labelSet = new String[] {"0","1","2","3","4","5","6","7","8","9"};
			
			Resource trainDataRes = new ClassPathResource(mnist_train_data);

			Resource trainLabelRes = new ClassPathResource(mnist_train_label);
			
			Resource testDataRes = new ClassPathResource(mnist_test_data);
			
			Resource testLabelRes = new ClassPathResource(mnist_test_label);
			
			DataSet trainData = DataLoader.loadDataByUByte(trainDataRes.getFile(), trainLabelRes.getFile(), labelSet, 1, 1 , 784, true);
			
			DataSet testData = DataLoader.loadDataByUByte(testDataRes.getFile(), testLabelRes.getFile(), labelSet, 1, 1 , 784, true);

			int channel = 1;
			
			int height = 28;
			
			int width = 28;
			
			CNN netWork = new CNN(new SoftmaxWithCrossEntropyLoss(), UpdaterType.momentum);
			
			netWork.learnRate = 0.001d;
			
			InputLayer inputLayer = new InputLayer(channel, 1, 784);
			
			ConvolutionLayer conv1 = new ConvolutionLayer(channel, 6, width, height, 5, 5, 2, 1, false);
			
			BNLayer bn1 = new BNLayer();
			
			LeakyReluLayer active1 = new LeakyReluLayer();
			
			PoolingLayer pool1 = new PoolingLayer(conv1.oChannel, conv1.oWidth, conv1.oHeight, 2, 2, 2, PoolingType.MAX_POOLING);
			
			ConvolutionLayer conv2 = new ConvolutionLayer(pool1.oChannel, 12, pool1.oWidth, pool1.oHeight, 5, 5, 0, 1, false);
			
			BNLayer bn2 = new BNLayer();
			
			LeakyReluLayer active2 = new LeakyReluLayer();
			
			DropoutLayer drop1 = new DropoutLayer(0.5d);
			
			
			PoolingLayer pool2 = new PoolingLayer(conv2.oChannel, conv2.oWidth, conv2.oHeight, 2, 2, 2, PoolingType.MAX_POOLING);

			int fInputCount = pool2.oChannel * pool2.oWidth * pool2.oHeight;
			
			int inputCount = (int) (Math.sqrt((fInputCount) + 10) + 10);
			
			FullyLayer full1 = new FullyLayer(fInputCount, inputCount, false);

			BNLayer bn3 = new BNLayer();
			
			LeakyReluLayer active3 = new LeakyReluLayer();
			
			FullyLayer full2 = new FullyLayer(inputCount, 10);
			
			SoftmaxWithCrossEntropyLayer softmax = new SoftmaxWithCrossEntropyLayer(10);

			netWork.addLayer(inputLayer);
			netWork.addLayer(conv1);
			netWork.addLayer(bn1);
			netWork.addLayer(active1);
			netWork.addLayer(pool1);
			netWork.addLayer(conv2);
			netWork.addLayer(bn2);
			netWork.addLayer(active2);
			netWork.addLayer(drop1);
			netWork.addLayer(pool2);
			netWork.addLayer(full1);
			netWork.addLayer(bn3);
			netWork.addLayer(active3);
			netWork.addLayer(full2);
			netWork.addLayer(softmax);

			MBSGDOptimizer optimizer = new MBSGDOptimizer(netWork, 10, 0.0001d, 96, LearnRateUpdate.NONE);

			long start = System.currentTimeMillis();
			
			optimizer.train(trainData);
			
			optimizer.test(testData);
			
			System.out.println(((System.currentTimeMillis() - start) / 1000) + "s.");

			
		} catch (Exception e) {
			// TODO: handle exception
			e.printStackTrace();
		}
		
	}

cnn cifar10 demo

public void cnnNetwork_cifar10() {
		// TODO Auto-generated method stub
		
		try {

	    	String[] labelSet = new String[] {"airplane","automobile","bird","cat","deer","dog","frog","horse","ship","truck"};
	    	
			String[] train_data_filenames = new String[] {
					"/dataset/cifar-10/data_batch_1.bin",
					"/dataset/cifar-10/data_batch_2.bin",
					"/dataset/cifar-10/data_batch_3.bin",
					"/dataset/cifar-10/data_batch_4.bin",
					"/dataset/cifar-10/data_batch_5.bin"
			};
			
			String test_data_filename = "/dataset/cifar-10/test_batch.bin";
			
			DataSet trainData = DataLoader.getImagesToDataSetByBin(train_data_filenames, 10000, 3, 32, 32, 10, true, labelSet);
	    	
			DataSet testData = DataLoader.getImagesToDataSetByBin(test_data_filename, 10000, 3, 32, 32, 10, true, labelSet);
			
			System.out.println("data is ready.");
			
			int channel = 3;
			
			int height = 32;
			
			int width = 32;
			
			CNN netWork = new CNN(new SoftmaxWithCrossEntropyLoss(), UpdaterType.adam);
			
			netWork.learnRate = 0.001d;
			
			InputLayer inputLayer = new InputLayer(channel, height, width);
			netWork.addLayer(inputLayer);
			
			ConvolutionLayer conv1 = new ConvolutionLayer(channel, 16, width, height, 3, 3, 1, 1,false);
			netWork.addLayer(conv1);
			
			BNLayer bn1 = new BNLayer();
			netWork.addLayer(bn1);
			
			ReluLayer active1 = new ReluLayer();
			netWork.addLayer(active1);

			PoolingLayer pool1 = new PoolingLayer(conv1.oChannel, conv1.oWidth, conv1.oHeight, 2, 2, 2, PoolingType.MAX_POOLING);
			netWork.addLayer(pool1);
			
			
			ConvolutionLayer conv3 = new ConvolutionLayer(pool1.oChannel, 32, pool1.oWidth, pool1.oHeight, 3, 3, 1, 1,false);
			netWork.addLayer(conv3);
			
			BNLayer bn3 = new BNLayer();
			netWork.addLayer(bn3);
			
			ReluLayer active3 = new ReluLayer();
			netWork.addLayer(active3);
			
			PoolingLayer pool2 = new PoolingLayer(conv3.oChannel, conv3.oWidth, conv3.oHeight, 2, 2, 2, PoolingType.MAX_POOLING);
			netWork.addLayer(pool2);

			
			ConvolutionLayer conv4 = new ConvolutionLayer(pool2.oChannel, 64, pool2.oWidth, pool2.oHeight, 3, 3, 1, 1,false);
			netWork.addLayer(conv4);
			
			BNLayer bn4 = new BNLayer();
			netWork.addLayer(bn4);
			
			ReluLayer active4 = new ReluLayer();
			netWork.addLayer(active4);
			
			PoolingLayer pool3 = new PoolingLayer(conv4.oChannel, conv4.oWidth, conv4.oHeight, 2, 2, 2, PoolingType.MAX_POOLING);
			netWork.addLayer(pool3);


			int fInputCount = pool3.oChannel * pool3.oWidth * pool3.oHeight;
			
			FullyLayer full1 = new FullyLayer(fInputCount, 256, true);
			netWork.addLayer(full1);
			
			ReluLayer active9 = new ReluLayer();
			netWork.addLayer(active9);

			DropoutLayer drop1 = new DropoutLayer(0.5d);
			netWork.addLayer(drop1);

			FullyLayer full2 = new FullyLayer(full1.oWidth, 10, true);
			netWork.addLayer(full2);
			
			SoftmaxWithCrossEntropyLayer softmax = new SoftmaxWithCrossEntropyLayer(10);
			netWork.addLayer(softmax);
			
			MBSGDOptimizer optimizer = new MBSGDOptimizer(netWork, 30, 0.001d, 64, LearnRateUpdate.NONE);

			long start = System.currentTimeMillis();
			
			optimizer.train(trainData);
			
			optimizer.test(testData);
			
			System.out.println(((System.currentTimeMillis() - start) / 1000) + "s.");
			
		} catch (Exception e) {
			// TODO: handle exception
			e.printStackTrace();
		}
		
	}

未来可期

实现rcnn、rnn、resnet、yolo等算法

训练情况可视化

支持动态调参，可视化训练在这里插入图片描述

彩蛋

基于神经网络+遗传算法实现AI赛车游戏

http://119.3.123.193:8011/AICar

版本更新

omega-engine-v3

2022-06-20

1.添加gup支持，使用jcuda调用cuda的cublasSgemm矩阵乘法，参考了caffe的卷积操作已将卷积操作优化成im2col+gemm实现，计算效率得到大大提高

2.添加vgg16 demo，该模型在cifar10数据集上表现为测试数据集准确率86.45%

3.利用jdk ForkJoin框架实现任务拆分，充分利用cpu多线程，提高对数组操作与计算速度

4.参考darknet对学习率更新机制进行升级，目前已支持RANDOM、POLY、STEP、EXP、SIG等多种学习率更新方法，并且实现学习率warmup功能

5.添加basicblock模块，新增resnet模型支持，目前该模型在cifar10数据集上的表现，训练次数10，测试数据集准确率为93.22%

omega-engine-v3-gpu

2022-07-02

1.开启omega-engine-v3-gpu版本开发，该版本将实现对omega-engine的gpu全面支持.

2.全面优化卷积层计算，包括前向传播与反向传播.

2022-08-17

1.初步完成卷积层的gpu改造，使得卷积神经网络计算速度整体提升，增加im2col与col2im两个经典的核函数（Im2colKernel.cu，Col2imKernel.cu）.

2.添加cuda内存管理器，用于管理整体显存的生命周期，减少频繁申请显存的操作，减少主机与显卡之间的数据传输.

2022-09-02

1.修改bn层计算dmean公式,减少计算量

2.更换数据存储方式，以便使用gpu计算，减少4维数组与1维数组之间的转换，获得成倍的计算效率提升

3.全面优化gpu计算，更新cuda核函数实现，使得训练与预测计算效获得大大提升

4.后续版本将进一步优化gpu版本，预计将整个计算过程搬迁入gpu计算，从而减少主机与设备(显卡)之间传输，希望进一步获得更快的计算速度

javy / Omega-AI .gitee-modal { width: 500px !important; }