Several correlation coefficients

Results Graph

Monte Carlo run

  • Iterations: 30
  • epochs: 300
  • batch_size: 100
  • Optimizer: SGD. Learning Rate: 0.1
  • Data is not shuffled at each iteration

Opened data located in data/TypeA

Correlation coefficient was changed from 0.1 to 0.9 in order to create the graph.

  • Num classes: 2
  • Total Samples: 20000
  • Vector size: 128
  • Train percentage: 80%

Models:

Complex Network

Dense layer

  • input size = 128(<class ‘numpy.complex64’>) -> output size = 64;
  • act_fun = cart_relu;
  • weight init = Glorot Uniform; bias init = Zeros
  • Dropout: 0.5

Dense layer

  • input size = 64(complex64) -> output size = 2;
  • act_fun = softmax_real;
  • weight init = Glorot Uniform; bias init = Zeros
  • Dropout: None

Real Network

Dense layer

  • input size = 256(<class ‘numpy.float32’>) -> output size = 128;
  • act_fun = cart_relu;
  • weight init = Glorot Uniform; bias init = Zeros
  • Dropout: 0.5

Dense layer

  • input size = 128(<class ‘numpy.float32’>) -> output size = 2;
  • act_fun = softmax_real;
  • weight init = Glorot Uniform; bias init = Zeros
  • Dropout: None