def create_model(self):
# create a SET-MLP model for CIFAR10 with 3 hidden layers
self.model = Sequential()
self.model.add(Flatten(input_shape=(32, 32, 3)))
self.model.add(
Dense(4000,
name="sparse_1",
kernel_constraint=MaskWeights(self.wm1),
weights=self.w1))
self.model.add(SReLU(name="srelu1", weights=self.wSRelu1))
self.model.add(Dropout(0.3))
self.model.add(
Dense(1000,
name="sparse_2",
kernel_constraint=MaskWeights(self.wm2),
weights=self.w2))
self.model.add(SReLU(name="srelu2", weights=self.wSRelu2))
self.model.add(Dropout(0.3))
self.model.add(
Dense(4000,
name="sparse_3",
kernel_constraint=MaskWeights(self.wm3),
weights=self.w3))
self.model.add(SReLU(name="srelu3", weights=self.wSRelu3))
self.model.add(Dropout(0.3))
self.model.add(
Dense(self.num_classes, name="dense_4", weights=self.w4)
) #please note that there is no need for a sparse output layer as the number of classes is much smaller than the number of input hidden neurons
self.model.add(Activation('softmax'))
def create_model(self):
# create a dense MLP model for CIFAR10 with 3 hidden layers
self.model = Sequential()
self.model.add(Flatten(input_shape=(32, 32, 3)))
self.model.add(Dense(4000, name="dense_1", weights=self.w1))
self.model.add(SReLU(name="srelu1", weights=self.wSRelu1))
self.model.add(Dropout(0.3))
self.model.add(Dense(1000, name="dense_2", weights=self.w2))
self.model.add(SReLU(name="srelu2", weights=self.wSRelu2))
self.model.add(Dropout(0.3))
self.model.add(Dense(4000, name="dense_3", weights=self.w3))
self.model.add(SReLU(name="srelu3", weights=self.wSRelu3))
self.model.add(Dropout(0.3))
self.model.add(Dense(self.num_classes, name="dense_4", weights=self.w4))
self.model.add(Activation('softmax'))
def create_model(self):
K.set_learning_phase(1)
self.model = Sequential()
if len(self.input_shape) > 2:
self.model.add(Flatten(input_shape=self.input_shape))
if (self.typemlp == "dense"):
if len(self.input_shape) <= 2:
self.model.add(
Dense(self.hidden_layer_sizes[0],
name="dense_1",
weights=self.w1,
input_shape=self.input_shape))
else:
self.model.add(
Dense(self.hidden_layer_sizes[0],
name="dense_1",
weights=self.w1))
self.model.add(SReLU(name="srelu1", weights=self.wSRelu1))
self.model.add(Dropout(0.3))
self.model.add(
Dense(self.hidden_layer_sizes[1],
name="dense_2",
weights=self.w2))
self.model.add(SReLU(name="srelu2", weights=self.wSRelu2))
self.model.add(Dropout(0.3))
self.model.add(
Dense(self.hidden_layer_sizes[2],
name="dense_3",
weights=self.w3))
self.model.add(SReLU(name="srelu3", weights=self.wSRelu3))
self.model.add(Dropout(0.3))
self.model.add(
Dense(self.num_classes, name="dense_4", weights=self.w4))
self.model.add(Activation('softmax'))
else:
if len(self.input_shape) <= 2:
self.model.add(
Dense(self.hidden_layer_sizes[0],
name="sparse_1",
kernel_constraint=MaskWeights(self.wm1),
weights=self.w1,
input_shape=self.input_shape))
else:
self.model.add(
Dense(self.hidden_layer_sizes[0],
name="sparse_1",
kernel_constraint=MaskWeights(self.wm1),
weights=self.w1))
self.model.add(SReLU(name="srelu1", weights=self.wSRelu1))
self.model.add(Dropout(0.3))
self.model.add(
Dense(self.hidden_layer_sizes[1],
name="sparse_2",
kernel_constraint=MaskWeights(self.wm2),
weights=self.w2))
self.model.add(SReLU(name="srelu2", weights=self.wSRelu2))
self.model.add(Dropout(0.3))
self.model.add(
Dense(self.hidden_layer_sizes[2],
name="sparse_3",
kernel_constraint=MaskWeights(self.wm3),
weights=self.w3))
self.model.add(SReLU(name="srelu3", weights=self.wSRelu3))
self.model.add(Dropout(0.3))
self.model.add(
Dense(self.num_classes, name="dense_4", weights=self.w4))
self.model.add(Activation('softmax'))