def complex_model_DFT(shape=(500,2),kernel_size=5,use_bias=False):
input_layer = Input(shape=shape)
scalar_layer = Complex_deconv(use_bias=use_bias,real_kernel_initializer='ones')(input_layer)
conv1 = ComplexConv1D(filters=1,kernel_size=kernel_size,padding='same',use_bias=use_bias)(scalar_layer)
fft1 = fft_layer1D()(conv1)
deconv1 = Complex_deconv(use_bias=use_bias,real_kernel_initializer='ones')(fft1)
deconv2 = ComplexConv1D(filters=1,kernel_size=kernel_size,padding='same',use_bias=use_bias)(deconv1)
model = Model(input_layer,deconv2)
return model
def conv_real_model(shape=(500, 2), use_bias=False):
'''
This model is similar to the conv_model, but the kernel itself is real number
'''
input_layer = Input(shape=shape)
conv1 = Complex_deconv(use_bias=use_bias,
real_kernel_initializer='ones')(input_layer)
#conv1 = Complex_real_Conv1D(filters=1,kernel_size=3,padding='same',activation='linear', kernel_initializer=TruncatedNormal(mean=0.33),use_bias=False)(input_layer)
#conv2 = Complex_real_Conv1D(filters=50,kernel_size=3,padding='same',activation='linear', kernel_initializer=TruncatedNormal(mean=0.33),use_bias=False)(conv1)
#conv3 = Complex_real_Conv1D(filters=1,kernel_size=1,padding='same',activation='linear',use_bias=False)(conv2)
fft1 = fft_layer1D()(conv1)
model = Model(input_layer, fft1)
return model
def much_complex_model(shape=(500, 2), use_bias=False):
input_layer = Input(shape=shape)
conv1 = ComplexConv1D(filters=64,
kernel_size=3,
padding='same',
use_bias=use_bias)(input_layer)
conv2 = ComplexConv1D(filters=128, kernel_size=3, padding='same')(conv1)
drop1 = Dropout(0.3)(conv2)
conv3 = ComplexConv1D(filters=256, kernel_size=3, padding='same')(drop1)
drop2 = Dropout(0.4)(conv3)
conv4 = ComplexConv1D(filters=128, kernel_size=3, padding='same')(drop2)
merge1 = concatenate([conv2, conv4], axis=2)
conv5 = ComplexConv1D(filters=64, kernel_size=3, padding='same')(merge1)
merge2 = concatenate([conv1, conv5])
fft1 = fft_layer1D()(merge2)
output = ComplexConv1D(1, 1, padding='same')(fft1)
model = Model(input_layer, output)
return model
def conv_model(shape=(500, 2), use_bias=True):
'''
to make sure that the input here is
'''
input_layer = Input(shape=shape)
conv1 = ComplexConv1D(
filters=100,
kernel_size=5,
padding='same',
activation='linear',
kernel_initializer=TruncatedNormal(mean=0.33))(input_layer)
conv2 = ComplexConv1D(filters=50,
kernel_size=3,
padding='same',
activation='linear',
kernel_initializer=TruncatedNormal(mean=0.33))(conv1)
conv3 = ComplexConv1D(filters=1,
kernel_size=1,
padding='same',
activation='linear',
use_bias=False)(conv2)
fft1 = fft_layer1D()(conv3)
model = Model(input_layer, fft1)
return model