def brainnetCNN_model_2(input_shape, n_filters, use_bias, n_outputs):
act = 'relu'
# define network architecture
# 2 E2E
# followed by 1 E2N
# followed by 1 N2G
# followed by dense
# then sigmoid for classification
# sprinkle with batch normalization and dropout
X_input = Input(shape=input_shape, name='input_layer')
E2E_1 = EdgeToEdge(n_filters[0], use_bias[0])(X_input)
E2E_1 = BatchNormalization()(E2E_1)
E2E_1 = Activation(act)(E2E_1)
E2E_2 = EdgeToEdge(n_filters[1], use_bias[1])(E2E_1)
E2E_2 = BatchNormalization()(E2E_2)
E2E_2 = Activation(act)(E2E_2)
E2N = EdgeToNode(n_filters[2], use_bias[2])(E2E_2)
E2N = Activation(act)(E2N)
N2G = NodeToGraph(n_filters[3], use_bias[3])(E2N)
N2G = Activation(act)(N2G)
N2G = Flatten()(N2G)
N2G = Dropout(0.2)(N2G)
dense_1 = Dense(n_filters[4], activation=act, use_bias=use_bias[4])(N2G)
dense_1 = Dropout(0.2)(dense_1)
dense_final = Dense(n_outputs, activation='linear',
use_bias=use_bias[5])(dense_1)
model = Model(inputs=X_input, outputs=dense_final)
return model
def custom_layer_model_stacked_B(input_shape, n_filters, use_bias):
# test my Keras layer
X_input = Input(shape=input_shape, name='input_layer')
E2E1 = EdgeToEdge(n_filters[0], use_bias[0])(X_input)
E2E2 = relu(E2E1)
E2E2 = EdgeToEdge(n_filters[1], use_bias[1])(E2E1)
E2N = EdgeToNode(n_filters[2], use_bias[2])(E2E2)
N2G = NodeToGraph(n_filters[3], use_bias[3])(E2N)
model = Model(inputs = X_input, outputs= N2G)
#model = Model(inputs = X_input, outputs= E2E2)
return model
def brainnetCNN_model(input_shape, n_filters, use_bias):
# define network architecture
# 2 E2E
# followed by 1 E2N
# followed by 1 N2G
# followed by dense
# then sigmoid for classification
X_input = Input(shape=input_shape, name='input_layer')
E2E_1 = EdgeToEdge(n_filters[0], use_bias[0])(X_input)
#E2E1 = relu(E2E1)
E2E_1 = Activation('relu')(E2E_1)
E2E2 = EdgeToEdge(n_filters[1], use_bias[1])(E2E_1)
#E2E2 = relu(E2E2)
#E2N = EdgeToNode(n_filters[2], use_bias[2])(E2E2)
# E2N = relu(E2N)
# N2G = NodeToGraph(n_filters[3], use_bias[3])(E2N)
# N2G = relu(N2G)
# dense_1 = Dense(n_filters[4], activation = 'relu', use_bias = use_bias[4])(N2G)
# dense_final = Dense(1, activation='sigmoid', use_bias = use_bias[5])(dense_1)
# model = Model(inputs = X_input, outputs= dense_final)
model = Model(inputs = X_input, outputs= E2E2)
return model
def brainnetCNN_model_3(input_shape, n_filters, use_bias):
# define network architecture
# 2 E2E
# followed by 1 E2N
# followed by 1 N2G
# followed by dense
# then sigmoid for classification
# sprinkle with batch normalization and dropout
X_input = Input(shape=input_shape, name='input_layer')
E2E_1 = EdgeToEdge(n_filters[0], use_bias[0])(X_input)
E2E_1 = BatchNormalization()(E2E_1)
E2E_1 = Activation('relu')(E2E_1)
dense_final = Dense(1, activation='linear', use_bias = use_bias[5])(E2E_1)
model = Model(inputs = X_input, outputs= dense_final)
return model
