def form_discriminator():
# ,recurrent_regularizer=l2(0.01)
model = Sequential()
model.add(
LSTM(input_shape=(seq_length, 1),
units=ncell,
unit_forget_bias=True,
return_sequences=True,
kernel_regularizer=l2(0.01),
recurrent_regularizer=l2(0.01),
bias_regularizer=l2(0.00)))
for i in range(nlayer - 1):
model.add(
LSTM(units=ncell,
unit_forget_bias=True,
return_sequences=True,
kernel_regularizer=l2(0.00),
recurrent_regularizer=l2(0.00),
bias_regularizer=l2(0.00)))
model.add(Dense(units=1))
model.add(Activation('sigmoid'))
model.add(pooling.AveragePooling1D(pool_size=seq_length, strides=None))
model.summary()
model_json = model.to_json()
with open('{0}/model_dis.json'.format(filepath), 'w') as f:
f = json.dump(model_json, f)
return model
def build_network(self):
model = Sequential()
model.add(Bidirectional(LSTM(units=cell_num,unit_forget_bias=True,return_sequences=True),input_shape=(signal_len,1)))
for i in range(lstm_num-1):
model.add(Bidirectional(LSTM(units=cell_num,unit_forget_bias=True,return_sequences=True)))
model.add(Dense(units=1,activation='sigmoid'))
model.add(pooling.AveragePooling1D(pool_size=signal_len,strides=None))
return model
def build_time_discriminator(self):
# with tf.device('/cpu:0'):
input = Input(shape = (seq_length, feature_count + class_num))
model =LSTM(units = ncell, use_bias=True, unit_forget_bias = True, return_sequences = True, recurrent_regularizer = l2(0.01))(input)
for i in range(nlayer - 1):
model = LSTM(units = ncell, use_bias=True, unit_forget_bias = True, return_sequences = True, recurrent_regularizer = l2(0.01))(model)
# model = LSTM(units = ncell, use_bias=True, unit_forget_bias = True, return_sequences = True, recurrent_regularizer = l2(0.01))(model)
# model = LSTM(units = 1, use_bias=True, activation='sigmoid', unit_forget_bias = True, return_sequences = True, recurrent_regularizer = l2(0.01))(model)
model = Dense(units=1, activation='sigmoid')(model)
model = pooling.AveragePooling1D(pool_size = seq_length, strides = None)(model)
return Model(inputs=input, outputs=model)
def create_emb_layer(self):
iw = Input(shape=(self.max_ngram_num, ),
dtype='int32',
name="inputword")
emb_in = embeddings.Embedding(output_dim=self.vector_size,
input_dim=self.ngram_size,
init="uniform",
name="input_layer")
vv_iw = emb_in(iw)
conv_l = convolutional.Convolution1D(self.vector_size,
30,
border_mode='same')
conv = conv_l(vv_iw)
pool = pooling.AveragePooling1D(self.max_ngram_num, border_mode="same")
pool_res = pool(conv)
return ([iw], emb_in, pool_res)
def build_discriminator(self):
input = Input(shape=(seq_length, output_count))
model = LSTM(units=ncell,
use_bias=True,
unit_forget_bias=False,
return_sequences=True,
recurrent_regularizer=l2(0.01))(input)
for i in range(nlayer - 1):
model = LSTM(units=ncell,
use_bias=True,
unit_forget_bias=False,
return_sequences=True,
recurrent_regularizer=l2(0.01))(model)
model = Dense(units=1, activation='sigmoid')(model)
model = pooling.AveragePooling1D(pool_size=seq_length,
strides=None)(model)
return Model(input, model)
def create_emb_layer(self):
iw = Input(shape=(self.max_ngram_num, ),
dtype='int32',
name="inputword")
emb_in = embeddings.Embedding(output_dim=self.vector_size,
input_dim=self.ngram_size,
init="uniform",
mask_zero=True,
name="input_layer")
vv_iw = emb_in(iw)
zm = ZeroMaskedEntries()
zm.build((None, self.max_ngram_num, self.vector_size))
zero_masked_emd = zm(vv_iw)
conv_l = convolutional.Convolution1D(self.vector_size,
30,
border_mode='same')
conv = conv_l(zero_masked_emd)
pool = pooling.AveragePooling1D(self.max_ngram_num, border_mode="same")
pool_res = pool(conv)
return ([iw], emb_in, pool_res)
def build_freq_discriminator(self):
len_src = self.y.shape[1]
t = np.argmin(0, 1, 1/len_src)
omega = 2*np.pi*np.arange(0, len_src)
t = np.ones((t.shape[0], omega.shape[0]))*t
cos_mat = np.cos(omega*t.T)
sin_mat = np.sin(omega*t.T)
# with tf.device('/cpu:0'):
input = Input(shape = (None, feature_count))
# class_label = Lambda(lambda x: x[:, :5], output_shape=(5,))(model_input)
# fft_layer = x1 = Lambda(lambda x: x[:, :5], output_shape=(5,))(input)
model =LSTM(units = ncell, use_bias=True, unit_forget_bias = True, return_sequences = True, recurrent_regularizer = l2(0.01))(input)
for i in range(nlayer - 1):
model = LSTM(units = ncell, use_bias=True, unit_forget_bias = True, return_sequences = True, recurrent_regularizer = l2(0.01))(model)
# model = LSTM(units = ncell, use_bias=True, unit_forget_bias = True, return_sequences = True, recurrent_regularizer = l2(0.01))(model)
# model = LSTM(units = 1, use_bias=True, activation='sigmoid', unit_forget_bias = True, return_sequences = True, recurrent_regularizer = l2(0.01))(model)
model = Dense(units=1, activation='sigmoid')(model)
model = pooling.AveragePooling1D(pool_size = seq_length, strides = None)(model)
return Model(inputs=input, outputs=model)
def create_emb_layer(self):
iw3 = Input(shape=(self.max_ngram_one_class, ),
dtype='int32',
name="inputword3")
iw4 = Input(shape=(self.max_ngram_one_class, ),
dtype='int32',
name="inputword4")
iw5 = Input(shape=(self.max_ngram_one_class, ),
dtype='int32',
name="inputword5")
iw6 = Input(shape=(self.max_ngram_one_class, ),
dtype='int32',
name="inputword6")
emb_in = embeddings.Embedding(output_dim=self.vector_size,
input_dim=self.ngram_size,
init="uniform",
mask_zero=True,
name="input_layer")
vv_iw3 = emb_in(iw3)
vv_iw4 = emb_in(iw4)
vv_iw5 = emb_in(iw5)
vv_iw6 = emb_in(iw6)
zm = ZeroMaskedEntries()
zm.build((None, self.max_ngram_num, self.vector_size))
zero_masked_emd3 = zm(vv_iw3)
zero_masked_emd4 = zm(vv_iw4)
zero_masked_emd5 = zm(vv_iw5)
zero_masked_emd6 = zm(vv_iw6)
conv_l3 = convolutional.Convolution1D(self.vector_size,
10,
border_mode='same')
conv3 = conv_l3(zero_masked_emd3)
conv_l4 = convolutional.Convolution1D(self.vector_size,
10,
border_mode='same')
conv4 = conv_l4(zero_masked_emd4)
conv_l5 = convolutional.Convolution1D(self.vector_size,
10,
border_mode='same')
conv5 = conv_l5(zero_masked_emd5)
conv_l6 = convolutional.Convolution1D(self.vector_size,
10,
border_mode='same')
conv6 = conv_l6(zero_masked_emd6)
pool3 = pooling.AveragePooling1D(self.max_ngram_one_class,
border_mode="same")
pool_res3 = pool3(conv3)
pool4 = pooling.AveragePooling1D(self.max_ngram_one_class,
border_mode="same")
pool_res4 = pool4(conv4)
pool5 = pooling.AveragePooling1D(self.max_ngram_one_class,
border_mode="same")
pool_res5 = pool5(conv5)
pool6 = pooling.AveragePooling1D(self.max_ngram_one_class,
border_mode="same")
pool_res6 = pool6(conv6)
merge_conv = Merge(mode='ave', concat_axis=1)
merged = merge_conv([pool_res3, pool_res4, pool_res5, pool_res6])
return ([iw3, iw4, iw5, iw6], emb_in, merged)