利用MultiDigitNumberRecognition.java来验证新版本

20191129

beta3运行自带的dl4j-example/org.deeplearning4j.example/catcharecognition/MultiDigitNumberRecognition.java是正常的，但是beta5运行同样的文件就有问题。

 

20191122

上述文件在beta6被放到了dl4j-example/org.deeplearning4j.examples/convolution/catcharecognition/中。

 

20200210

beta6运行上述文件，没有配置虚拟机参数，4个epoch计算完以后，除了报以下错误以外，其余运行正常。程序停止前的混淆矩阵分布不记得了，没空重新测试。

Exception in thread "main" org.nd4j.linalg.exception.ND4JIllegalStateException: Axis array [1] contains values above array rank (rank=1)    at org.nd4j.linalg.api.shape.Shape.normalizeAxis(Shape.java:3394)    at org.nd4j.linalg.api.ops.BaseOp.defineDimensions(BaseOp.java:344)    at org.nd4j.linalg.api.ops.BaseIndexAccumulation.<init>(BaseIndexAccumulation.java:95)    at org.nd4j.linalg.api.ops.impl.indexaccum.IMax.<init>(IMax.java:47)    at org.nd4j.linalg.factory.Nd4j.argMax(Nd4j.java:578)    at org.deeplearning4j.examples.convolution.captcharecognition.MultiDigitNumberRecognition.modelPredict(MultiDigitNumberRecognition.java:177)    at org.deeplearning4j.examples.convolution.captcharecognition.MultiDigitNumberRecognition.main(MultiDigitNumberRecognition.java:103)

 

20200527

beta7运行MultiDigitNumberRecognition.java后和20200210beta6的情况一样。

程序停止前的混淆矩阵似乎有问题。

  0  1  2  3  4  5  6  7  8  9
-------------------------------
  0  0  0  0  0  0  0  0  0 22 | 0 = 0
  0  0  0  0  0  0  0  0  0  0 | 1 = 1
  0  0  0  0  0  0  0  0  0 13 | 2 = 2
  0  0  0  0  0  0  0  0  0  7 | 3 = 3
  0  0  0  0  0  0  0  0  0 16 | 4 = 4
  0  0  0  0  0  0  0  0  0  6 | 5 = 5
  0  0  0  0  0  0  0  0  0 14 | 6 = 6
  0  0  0  0  0  0  0  0  0  0 | 7 = 7
  0  0  0  0  0  0  0  0  0 15 | 8 = 8
  0  0  0  0  0  0  0  0  0 15 | 9 = 9

暂时就讨论到这里。