def build(self, hp):
maxnorm = hp.Choice('maxnorm', values=self.hyperparam['maxnorm'])
resnet_model = Resnet(input_shape=(self.seqlen, self.channels), norm_max=maxnorm)
if self.pretrained_weights is not None:
resnet_model.set_weights(self.pretrained_weights)
inp = Input(shape=(self.seqlen, self.channels))
enc_inp = resnet_model(inp)
dense_units = hp.Int('preclassification', min_value = self.hyperparam['dense_units']['min'],\
max_value = self.hyperparam['dense_units']['max'], step = self.hyperparam['dense_units']['step'])
dense_out = Dense(units = dense_units, activation='relu',
kernel_constraint=MaxNorm(maxnorm,axis=[0,1]),
bias_constraint=MaxNorm(maxnorm,axis=0),
kernel_initializer=glorot_uniform(seed=0))(enc_inp)
dense_out = Dropout(rate=hp.Choice('dropout', values = self.hyperparam['dropout']))(dense_out)
output = Dense(self.num_classes, activation='softmax',
kernel_constraint=MaxNorm(maxnorm,axis=[0,1]),
bias_constraint=MaxNorm(maxnorm,axis=0),
kernel_initializer=glorot_uniform(seed=0))(dense_out)
model = Model(inputs=inp, outputs=output)
model.compile(optimizer=Adam(lr=hp.Choice('lr', values = self.hyperparam['lr'])),
loss=focal_loss(), metrics=['accuracy', macro_f1])
return model
def main(argv):
infile = argv[0]
outfile = argv[1]
seqlen = 1500
channels = 6
maxnorm = 20.0
# Create model
resnet_model = Resnet(input_shape=(seqlen, channels), norm_max=maxnorm)
samp1 = Input(shape=(seqlen, channels))
enc_samp1 = resnet_model(samp1)
samp2 = Input(shape=(seqlen, channels))
enc_samp2 = resnet_model(samp2)
diff_layer = Lambda(lambda tensors: K.abs(tensors[0] - tensors[1]))
diff_enc = diff_layer([enc_samp1, enc_samp2])
dense_out = Dense(50,
activation='relu',
kernel_constraint=MaxNorm(maxnorm, axis=[0, 1]),
bias_constraint=MaxNorm(maxnorm, axis=0),
kernel_initializer=glorot_uniform(seed=0))(diff_enc)
dense_out = Dropout(rate=0.2)(dense_out)
output = Dense(1,
activation='sigmoid',
kernel_constraint=MaxNorm(maxnorm, axis=[0, 1]),
bias_constraint=MaxNorm(maxnorm, axis=0),
kernel_initializer=glorot_uniform(seed=0))(dense_out)
model = Model(inputs=[samp1, samp2], outputs=output)
model.load_weights(infile)
for layer in model.layers:
if layer.name == "model":
resnet_model.set_weights(layer.get_weights())
resnet_model.save_weights(outfile)
