--network_type=cnn # Images are 128x128x3 --input_shape 128 128 3 # Convolutional modules are defined by: kernel size, padding type, # number of filters, activation function, and pooling size # Kernel sizes at each of four layers. Since the input is an image, # then the true kernel sizes are 3x3, 3x3, 5x5, and 5x5, respectively --conv_kernel_size 3 3 5 5 --conv_padding=valid # Number of filters at each layer --conv_number_filters 8 16 16 32 --conv_activation=elu # Max pooling at each layer (0 = none; 2 = 2x2 pooling) --conv_pool_size 0 2 0 0 # Following global max pooling, the input is encoded # as a vector. This vector is further processed using # a sequence of fully-connected layers --number_hidden_units 20 10 --hidden_activation=elu # Two classes --output_shape 2 # The classes are exclusive, requiring softmax --output_activation=softmax