Java 为什么DeepLearning4J CNN在INDARY输出中返回的不是概率而是0和1
我正在使用DL4J版本1.0.0-beta3，并尝试创建一个卷积神经网络，用于识别32x32个棋子图像。 以下是我用来创建和训练网络的代码：Java 为什么DeepLearning4J CNN在INDARY输出中返回的不是概率而是0和1,java,conv-neural-network,deeplearning4j,dl4j,Java,Conv Neural Network,Deeplearning4j,Dl4j,我正在使用DL4J版本1.0.0-beta3，并尝试创建一个卷积神经网络，用于识别32x32个棋子图像。 以下是我用来创建和训练网络的代码： public class BuildNetwork1 { public static void main(String[] args) throws Exception { File rootDir = new File("./CNNinput/chesscom1");
public class BuildNetwork1 {
public static void main(String[] args) throws Exception {
File rootDir = new File("./CNNinput/chesscom1");
File locationToSave = new File(rootDir, "trained.chesscom1.bin");
int height = 32;
int width = 32;
int channels = 1;
int rngseed = 777;
int numEpochs = 100;
File trainData = new File(rootDir, "training");
MultiLayerConfiguration conf = new NeuralNetConfiguration.Builder()
.seed(rngseed)
.updater(new Adam.Builder().learningRate(0.01).build())
.activation(Activation.IDENTITY)
.weightInit(WeightInit.XAVIER)
.list()
//.layer(new ConvolutionLayer.Builder(new int[] {5, 5}, new int[] {1, 1}, new int[]{0, 0}).name("cnn1").nIn(1).nOut(64).biasInit(0).build())
//.layer(new SubsamplingLayer.Builder(new int[] {2, 2}, new int[] {2, 2}).name("maxpool1").build())
//.layer(new ConvolutionLayer.Builder(new int[] {5, 5}, new int[] {1, 1}, new int[]{0, 0}).name("cnn2").nIn(64).nOut(16).biasInit(0).build())
.layer(new OutputLayer.Builder(LossFunctions.LossFunction.NEGATIVELOGLIKELIHOOD)
.nOut(13)
.activation(Activation.SOFTMAX)
.build())
.setInputType(InputType.convolutional(32, 32, 1))
.build();
MultiLayerNetwork network = new MultiLayerNetwork(conf);
network.init();
network.setListeners(new ScoreIterationListener(10));
ImageLoader loader = new ImageLoader(height, width, channels);
DataNormalization scaler = new ImagePreProcessingScaler(0, 1);
for (int e = 0; e < numEpochs; e++) {
File[] labels = trainData.listFiles();
for (int i = 0; i < labels.length; i++) {
File label = labels[i];
File[] images = label.listFiles();
for (int j = 0; j < images.length; j++) {
File imageFile = images[j];
BufferedImage image = ImageIO.read(imageFile);
INDArray input = loader.asMatrix(image).reshape(1, channels, height, width);
scaler.fit(new DataSet(input, null));
scaler.transform(input);
double[][] outputArray = new double[1][13];
outputArray[0][Integer.parseInt(label.getName())] = 1d;
INDArray output = Nd4j.create(outputArray);
network.fit(input, output);
}
}
}
boolean saveUpdater = true;
ModelSerializer.writeModel(network, locationToSave, saveUpdater);
}
}
提前多谢 您的模型对其输出非常有信心。当您向it部门展示以前可能见过的数据时，以及当您已将模型训练为非常适合该数据（通常称为过度拟合）时，可能会发生这种情况。感谢您的及时回答！当我测试棋子的任意图像时，同样的自信输出（0和1）也会出现，但当然输出是错误的。最后，无论我尝试什么，我都无法接收概率。因为您使用的是身份激活，我猜softmax会获得非常高的输入数字，这些数字会变为1/0或非常接近1或0，实际上就是这样。嗨，保罗，再次感谢您的回答！我在[0，1]中使用了scaler作为输入，我假设它可以处理这个问题，不是吗？这会将您的输入置于0..1范围内，但与权重相乘后，它可能位于任何范围内。嗨，Paul，非常感谢！我成功地产生了概率！使用Activation.TANH并取消对代码中隐藏层的注释。我还评论了updater，因为它可以很好地处理默认值。
public class CalcNetworkAll {
public static void main(String[] args) throws Exception {
int height = 32;
int width = 32;
int channels = 1;
File rootDir = new File("./CNNinput/chesscom1");
File locationToLoad = new File("./CNNinput/chesscom1/trained.chesscom1.bin");
File testData = new File(rootDir, "testing");
MultiLayerNetwork network = ModelSerializer.restoreMultiLayerNetwork(locationToLoad);
ImageLoader loader = new ImageLoader(height, width, channels);
DataNormalization scaler = new ImagePreProcessingScaler(0, 1);
File[] labels = testData.listFiles();
for (int i = 0; i < labels.length; i++) {
File label = labels[i];
File[] images = label.listFiles();
for (int j = 0; j < images.length; j++) {
File imageFile = images[j];
BufferedImage image = ImageIO.read(imageFile);
INDArray input = loader.asMatrix(image).reshape(1, channels, height, width);
scaler.fit(new DataSet(input, null));
scaler.transform(input);
INDArray output = network.output(input, false);
System.out.println(label.getName() + " => " + output);
}
}
}
}
0 => [[ 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
0 => [[ 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
0 => [[ 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
0 => [[ 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
0 => [[ 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
0 => [[ 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
0 => [[ 1.0000,8.1707e-37, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
0 => [[ 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
0 => [[ 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
0 => [[ 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
0 => [[ 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
0 => [[ 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
0 => [[ 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
0 => [[ 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
1 => [[ 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
1 => [[ 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
1 => [[ 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
1 => [[ 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
1 => [[ 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
1 => [[ 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
1 => [[ 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
1 => [[ 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
1 => [[ 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
1 => [[ 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
1 => [[ 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
1 => [[ 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
1 => [[ 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
1 => [[ 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
10 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0]]
10 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0]]
10 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0]]
10 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0]]
10 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0]]
10 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0]]
10 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0]]
10 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0]]
10 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0]]
10 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0]]
10 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0]]
10 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0]]
10 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0]]
10 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0]]
11 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0]]
11 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0]]
11 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0]]
11 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0]]
11 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0]]
11 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0]]
11 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0]]
12 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000]]
12 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000]]
12 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000]]
12 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000]]
12 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000]]
12 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000]]
12 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000]]
2 => [[ 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
2 => [[ 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
2 => [[ 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
2 => [[ 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
2 => [[ 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
2 => [[ 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
2 => [[ 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
2 => [[ 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
2 => [[ 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
2 => [[ 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
2 => [[ 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
2 => [[ 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
2 => [[ 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
2 => [[ 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
3 => [[ 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
3 => [[ 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
3 => [[ 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
3 => [[ 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
3 => [[ 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
3 => [[ 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
3 => [[ 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
3 => [[ 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
3 => [[ 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
3 => [[ 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
3 => [[ 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
3 => [[ 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
3 => [[ 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
3 => [[ 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
4 => [[ 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0]]
4 => [[ 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0]]
4 => [[ 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0]]
4 => [[ 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0]]
4 => [[ 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0]]
4 => [[ 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0]]
4 => [[ 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0]]
4 => [[ 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0]]
4 => [[ 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0]]
4 => [[ 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0]]
4 => [[ 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0]]
4 => [[ 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0]]
4 => [[ 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0]]
4 => [[ 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0, 0]]
5 => [[ 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0]]
5 => [[ 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0]]
5 => [[ 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0]]
5 => [[ 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0]]
5 => [[ 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0]]
5 => [[ 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0]]
5 => [[ 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0, 0]]
6 => [[ 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0]]
6 => [[ 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0]]
6 => [[ 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0]]
6 => [[ 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0]]
6 => [[ 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0]]
6 => [[ 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0]]
6 => [[ 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0, 0]]
7 => [[ 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0]]
7 => [[ 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0]]
7 => [[ 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0]]
7 => [[ 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0]]
7 => [[ 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0]]
7 => [[ 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0]]
7 => [[ 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0]]
7 => [[ 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0]]
7 => [[ 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0]]
7 => [[ 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0]]
7 => [[ 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0]]
7 => [[ 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0]]
7 => [[ 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0]]
7 => [[ 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0, 0]]
8 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0]]
8 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0]]
8 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0]]
8 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0]]
8 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0]]
8 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0]]
8 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0]]
8 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0]]
8 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0]]
8 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0]]
8 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0]]
8 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0]]
8 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0]]
8 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0, 0]]
9 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0]]
9 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0]]
9 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0]]
9 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0]]
9 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0]]
9 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0]]
9 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0]]
9 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0]]
9 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0]]
9 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0]]
9 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0]]
9 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0]]
9 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0]]
9 => [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0000, 0, 0, 0]]