def Conv_Dense_0001_a(DataCenter):
tfInitial.initialize_placeholders(DataCenter)
scope = DataCenter.new_model_scope
last_layer = DataCenter.train_output_data.shape[1]
with tf.variable_scope(scope):
x = DataCenter.x_placeholder
nn = tf.layers.conv1d(x, filters=64, kernel_size=5, name='conv1_1')
nn = tf.layers.conv1d(nn, filters=64, kernel_size=5, name='conv1_2')
nn = tf.layers.conv1d(nn, filters=64, kernel_size=5, name='conv1_3')
nn = tf.layers.max_pooling1d(nn,
pool_size=4,
strides=2,
name='max_pool1')
nn = tf.layers.batch_normalization(nn, name='batchnorm_1')
nn = tf.layers.conv1d(nn,
filters=128,
kernel_size=5,
activation=tf.nn.relu,
name='conv2_1')
nn = tf.layers.conv1d(nn,
filters=128,
kernel_size=5,
activation=tf.nn.relu,
name='conv2_2')
nn = tf.layers.conv1d(nn,
filters=128,
kernel_size=5,
activation=tf.nn.relu,
name='conv2_3')
nn = tf.layers.max_pooling1d(nn,
pool_size=4,
strides=2,
name='max_pool2')
nn = tf.layers.batch_normalization(nn, name='batchnorm_2')
nn = tf.layers.conv1d(nn,
filters=256,
kernel_size=5,
activation=tf.nn.relu,
name='conv3_1')
nn = tf.layers.max_pooling1d(nn,
pool_size=4,
strides=2,
name='max_pool3')
nn = tf.layers.batch_normalization(nn, name='batchnorm_3')
nn = tf.layers.flatten(nn, name='flatten_1')
nn = tf.layers.dense(nn, 512, activation=tf.nn.relu, name='dense_1')
nn = tf.layers.dense(nn,
last_layer,
activation=tf.nn.softmax,
name='dense_3')
return nn
def Conv_Dense_0001_b(DataCenter):
tfInitial.initialize_placeholders(DataCenter)
scope = DataCenter.new_model_scope
last_layer = DataCenter.train_output_data.shape[1]
activ = tf.nn.relu
with tf.variable_scope(scope):
x = DataCenter.x_placeholder
conv1_7_7 = tf.layers.conv1d(x,
64,
7,
strides=2,
padding='same',
activation=activ,
name='conv1_7_7_s2')
pool1_3_3 = tf.layers.max_pooling1d(conv1_7_7, 3, strides=2)
pool1_3_3 = tf.layers.batch_normalization(pool1_3_3,
name='batchnorm_1')
conv2_3_3_reduce = tf.layers.conv1d(pool1_3_3,
64,
1,
padding='same',
activation=activ,
name='conv2_3_3_reduce')
conv2_3_3 = tf.layers.conv1d(conv2_3_3_reduce,
192,
3,
padding='same',
activation=activ,
name='conv2_3_3')
conv2_3_3 = tf.layers.batch_normalization(conv2_3_3)
pool2_3_3 = tf.layers.max_pooling1d(conv2_3_3,
pool_size=3,
strides=2,
name='pool2_3_3_s2')
inception_3a_1_1 = tf.layers.conv1d(pool2_3_3,
64,
1,
padding='same',
activation=activ,
name='inception_3a_1_1')
inception_3a_3_3_reduce = tf.layers.conv1d(
pool2_3_3,
96,
1,
padding='same',
activation=activ,
name='inception_3a_3_3_reduce')
inception_3a_3_3 = tf.layers.conv1d(inception_3a_3_3_reduce,
128,
padding='same',
kernel_size=3,
activation=activ,
name='inception_3a_3_3')
inception_3a_5_5_reduce = tf.layers.conv1d(
pool2_3_3,
16,
padding='same',
kernel_size=1,
activation=activ,
name='inception_3a_5_5_reduce')
inception_3a_5_5 = tf.layers.conv1d(inception_3a_5_5_reduce,
32,
padding='same',
kernel_size=5,
activation=activ,
name='inception_3a_5_5')
inception_3a_pool = tf.layers.max_pooling1d(pool2_3_3,
pool_size=3,
strides=1,
padding='same')
inception_3a_pool_1_1 = tf.layers.conv1d(inception_3a_pool,
32,
kernel_size=1,
padding='same',
activation=activ,
name='inception_3a_pool_1_1')
# merge the inception_3a__
inception_3a_output = tf.concat([
inception_3a_1_1, inception_3a_3_3, inception_3a_5_5,
inception_3a_pool_1_1
],
axis=2)
inception_3b_1_1 = tf.layers.conv1d(inception_3a_output,
128,
kernel_size=1,
padding='same',
activation=activ,
name='inception_3b_1_1')
inception_3b_3_3_reduce = tf.layers.conv1d(
inception_3a_output,
128,
kernel_size=1,
padding='same',
activation=activ,
name='inception_3b_3_3_reduce')
inception_3b_3_3 = tf.layers.conv1d(inception_3b_3_3_reduce,
192,
kernel_size=3,
padding='same',
activation=activ,
name='inception_3b_3_3')
inception_3b_5_5_reduce = tf.layers.conv1d(
inception_3a_output,
32,
kernel_size=1,
padding='same',
activation=activ,
name='inception_3b_5_5_reduce')
inception_3b_5_5 = tf.layers.conv1d(inception_3b_5_5_reduce,
96,
kernel_size=5,
padding='same',
activation=activ,
name='inception_3b_5_5')
inception_3b_pool = tf.layers.max_pooling1d(inception_3a_output,
pool_size=3,
strides=1,
padding='same',
name='inception_3b_pool')
inception_3b_pool_1_1 = tf.layers.conv1d(inception_3b_pool,
64,
kernel_size=1,
padding='same',
activation=activ,
name='inception_3b_pool_1_1')
# merge the inception_3b_*
inception_3b_output = tf.concat([
inception_3b_1_1, inception_3b_3_3, inception_3b_5_5,
inception_3b_pool_1_1
],
axis=2,
name='inception_3b_output')
pool3_3_3 = tf.layers.max_pooling1d(inception_3b_output,
pool_size=3,
strides=2,
name='pool3_3_3')
inception_4a_1_1 = tf.layers.conv1d(pool3_3_3,
192,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4a_1_1')
inception_4a_3_3_reduce = tf.layers.conv1d(
pool3_3_3,
96,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4a_3_3_reduce')
inception_4a_3_3 = tf.layers.conv1d(inception_4a_3_3_reduce,
208,
kernel_size=3,
padding='same',
activation=activ,
name='inception_4a_3_3')
inception_4a_5_5_reduce = tf.layers.conv1d(
pool3_3_3,
16,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4a_5_5_reduce')
inception_4a_5_5 = tf.layers.conv1d(inception_4a_5_5_reduce,
48,
kernel_size=5,
padding='same',
activation=activ,
name='inception_4a_5_5')
inception_4a_pool = tf.layers.max_pooling1d(pool3_3_3,
pool_size=3,
strides=1,
padding='same',
name='inception_4a_pool')
inception_4a_pool_1_1 = tf.layers.conv1d(inception_4a_pool,
64,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4a_pool_1_1')
inception_4a_output = tf.concat([
inception_4a_1_1, inception_4a_3_3, inception_4a_5_5,
inception_4a_pool_1_1
],
axis=2,
name='inception_4a_output')
inception_4b_1_1 = tf.layers.conv1d(inception_4a_output,
160,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4b_1_1')
inception_4b_3_3_reduce = tf.layers.conv1d(
inception_4a_output,
112,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4b_3_3_reduce')
inception_4b_3_3 = tf.layers.conv1d(inception_4b_3_3_reduce,
224,
kernel_size=3,
padding='same',
activation=activ,
name='inception_4b_3_3')
inception_4b_5_5_reduce = tf.layers.conv1d(
inception_4a_output,
24,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4b_5_5_reduce')
inception_4b_5_5 = tf.layers.conv1d(inception_4b_5_5_reduce,
64,
kernel_size=5,
padding='same',
activation=activ,
name='inception_4b_5_5')
inception_4b_pool = tf.layers.max_pooling1d(inception_4a_output,
pool_size=3,
strides=1,
padding='same',
name='inception_4b_pool')
inception_4b_pool_1_1 = tf.layers.conv1d(inception_4b_pool,
64,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4b_pool_1_1')
inception_4b_output = tf.concat([
inception_4b_1_1, inception_4b_3_3, inception_4b_5_5,
inception_4b_pool_1_1
],
axis=2,
name='inception_4b_output')
inception_4c_1_1 = tf.layers.conv1d(inception_4b_output,
128,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4c_1_1')
inception_4c_3_3_reduce = tf.layers.conv1d(
inception_4b_output,
128,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4c_3_3_reduce')
inception_4c_3_3 = tf.layers.conv1d(inception_4c_3_3_reduce,
256,
kernel_size=3,
padding='same',
activation=activ,
name='inception_4c_3_3')
inception_4c_5_5_reduce = tf.layers.conv1d(
inception_4b_output,
24,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4c_5_5_reduce')
inception_4c_5_5 = tf.layers.conv1d(inception_4c_5_5_reduce,
64,
kernel_size=5,
padding='same',
activation=activ,
name='inception_4c_5_5')
inception_4c_pool = tf.layers.max_pooling1d(inception_4b_output,
pool_size=3,
strides=1,
padding='same')
inception_4c_pool_1_1 = tf.layers.conv1d(inception_4c_pool,
64,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4c_pool_1_1')
inception_4c_output = tf.concat([
inception_4c_1_1, inception_4c_3_3, inception_4c_5_5,
inception_4c_pool_1_1
],
axis=2,
name='inception_4c_output')
inception_4d_1_1 = tf.layers.conv1d(inception_4c_output,
112,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4d_1_1')
inception_4d_3_3_reduce = tf.layers.conv1d(
inception_4c_output,
144,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4d_3_3_reduce')
inception_4d_3_3 = tf.layers.conv1d(inception_4d_3_3_reduce,
288,
kernel_size=3,
padding='same',
activation=activ,
name='inception_4d_3_3')
inception_4d_5_5_reduce = tf.layers.conv1d(
inception_4c_output,
32,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4d_5_5_reduce')
inception_4d_5_5 = tf.layers.conv1d(inception_4d_5_5_reduce,
64,
kernel_size=5,
padding='same',
activation=activ,
name='inception_4d_5_5')
inception_4d_pool = tf.layers.max_pooling1d(inception_4c_output,
pool_size=3,
strides=1,
padding='same',
name='inception_4d_pool')
inception_4d_pool_1_1 = tf.layers.conv1d(inception_4d_pool,
64,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4d_pool_1_1')
inception_4d_output = tf.concat([
inception_4d_1_1, inception_4d_3_3, inception_4d_5_5,
inception_4d_pool_1_1
],
axis=2,
name='inception_4d_output')
inception_4e_1_1 = tf.layers.conv1d(inception_4d_output,
256,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4e_1_1')
inception_4e_3_3_reduce = tf.layers.conv1d(
inception_4d_output,
160,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4e_3_3_reduce')
inception_4e_3_3 = tf.layers.conv1d(inception_4e_3_3_reduce,
320,
kernel_size=3,
padding='same',
activation=activ,
name='inception_4e_3_3')
inception_4e_5_5_reduce = tf.layers.conv1d(
inception_4d_output,
32,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4e_5_5_reduce')
inception_4e_5_5 = tf.layers.conv1d(inception_4e_5_5_reduce,
128,
kernel_size=5,
padding='same',
activation=activ,
name='inception_4e_5_5')
inception_4e_pool = tf.layers.max_pooling1d(inception_4d_output,
pool_size=3,
strides=1,
padding='same',
name='inception_4e_pool')
inception_4e_pool_1_1 = tf.layers.conv1d(inception_4e_pool,
128,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4e_pool_1_1')
inception_4e_output = tf.concat([
inception_4e_1_1, inception_4e_3_3, inception_4e_5_5,
inception_4e_pool_1_1
],
axis=2)
pool4_3_3 = tf.layers.max_pooling1d(inception_4e_output,
pool_size=3,
strides=2,
name='pool_3_3')
inception_5a_1_1 = tf.layers.conv1d(pool4_3_3,
256,
kernel_size=1,
padding='same',
activation=activ,
name='inception_5a_1_1')
inception_5a_3_3_reduce = tf.layers.conv1d(
pool4_3_3,
160,
kernel_size=1,
padding='same',
activation=activ,
name='inception_5a_3_3_reduce')
inception_5a_3_3 = tf.layers.conv1d(inception_5a_3_3_reduce,
320,
kernel_size=3,
padding='same',
activation=activ,
name='inception_5a_3_3')
inception_5a_5_5_reduce = tf.layers.conv1d(
pool4_3_3,
32,
kernel_size=1,
padding='same',
activation=activ,
name='inception_5a_5_5_reduce')
inception_5a_5_5 = tf.layers.conv1d(inception_5a_5_5_reduce,
128,
kernel_size=5,
padding='same',
activation=activ,
name='inception_5a_5_5')
inception_5a_pool = tf.layers.max_pooling1d(pool4_3_3,
pool_size=3,
strides=1,
padding='same',
name='inception_5a_pool')
inception_5a_pool_1_1 = tf.layers.conv1d(inception_5a_pool,
128,
kernel_size=1,
padding='same',
activation=activ,
name='inception_5a_pool_1_1')
inception_5a_output = tf.concat([
inception_5a_1_1, inception_5a_3_3, inception_5a_5_5,
inception_5a_pool_1_1
],
axis=2)
inception_5b_1_1 = tf.layers.conv1d(inception_5a_output,
384,
kernel_size=1,
padding='same',
activation=activ,
name='inception_5b_1_1')
inception_5b_3_3_reduce = tf.layers.conv1d(
inception_5a_output,
192,
kernel_size=1,
padding='same',
activation=activ,
name='inception_5b_3_3_reduce')
inception_5b_3_3 = tf.layers.conv1d(inception_5b_3_3_reduce,
384,
kernel_size=3,
padding='same',
activation=activ,
name='inception_5b_3_3')
inception_5b_5_5_reduce = tf.layers.conv1d(
inception_5a_output,
48,
kernel_size=1,
padding='same',
activation=activ,
name='inception_5b_5_5_reduce')
inception_5b_5_5 = tf.layers.conv1d(inception_5b_5_5_reduce,
128,
kernel_size=5,
padding='same',
activation=activ,
name='inception_5b_5_5')
inception_5b_pool = tf.layers.max_pooling1d(inception_5a_output,
pool_size=3,
strides=1,
padding='same',
name='inception_5b_pool')
inception_5b_pool_1_1 = tf.layers.conv1d(inception_5b_pool,
128,
kernel_size=1,
padding='same',
activation=activ,
name='inception_5b_pool_1_1')
inception_5b_output = tf.concat([
inception_5b_1_1, inception_5b_3_3, inception_5b_5_5,
inception_5b_pool_1_1
],
axis=2)
pool5_7_7 = tf.layers.max_pooling1d(inception_5b_output,
pool_size=7,
strides=1)
nn = tf.layers.flatten(pool5_7_7)
nn = tf.layers.dense(nn, 1024)
nn = tf.layers.dense(nn, 1024)
nn = tf.layers.dense(nn,
last_layer,
activation=tf.nn.softmax,
name='output')
return nn
def Conv_Dense_0001_c(DataCenter, trainable=False):
tfInitial.initialize_placeholders(DataCenter)
if trainable == False:
print('Trainable is set to False for Transfer Layers')
scope = DataCenter.transfer_model_scope
last_layer = DataCenter.train_output_data.shape[1]
with tf.variable_scope(scope):
x = DataCenter.x_placeholder
nn = tf.layers.conv1d(x,
filters=64,
kernel_size=9,
name='conv1_1',
trainable=trainable)
nn = tf.layers.conv1d(nn,
filters=64,
kernel_size=9,
name='conv1_2',
trainable=trainable)
nn = tf.layers.conv1d(nn,
filters=64,
kernel_size=9,
name='conv1_3',
trainable=trainable)
nn = tf.layers.max_pooling1d(nn,
pool_size=2,
strides=2,
name='max_pool1')
nn = tf.layers.dropout(nn, 0.5)
nn = tf.layers.batch_normalization(nn,
name='batchnorm_1',
trainable=trainable)
nn = tf.layers.conv1d(nn,
filters=128,
kernel_size=7,
activation=tf.nn.relu,
name='conv2_1',
trainable=trainable)
nn = tf.layers.conv1d(nn,
filters=128,
kernel_size=7,
activation=tf.nn.relu,
name='conv2_2',
trainable=trainable)
nn = tf.layers.conv1d(nn,
filters=128,
kernel_size=7,
activation=tf.nn.relu,
name='conv2_3',
trainable=trainable)
nn = tf.layers.max_pooling1d(nn,
pool_size=2,
strides=2,
name='max_pool2')
nn = tf.layers.dropout(nn, 0.5)
nn = tf.layers.batch_normalization(nn,
name='batchnorm_2',
trainable=trainable)
nn = tf.layers.conv1d(nn,
filters=256,
kernel_size=5,
activation=tf.nn.relu,
name='conv3_1',
trainable=trainable)
nn = tf.layers.conv1d(nn,
filters=256,
kernel_size=5,
activation=tf.nn.relu,
name='conv3_2',
trainable=trainable)
nn = tf.layers.conv1d(nn,
filters=256,
kernel_size=5,
activation=tf.nn.relu,
name='conv3_3',
trainable=trainable)
nn = tf.layers.max_pooling1d(nn,
pool_size=2,
strides=2,
name='max_pool3')
nn = tf.layers.dropout(nn, 0.5)
nn = tf.layers.batch_normalization(nn,
name='batchnorm_3',
trainable=trainable)
nn = tf.layers.flatten(nn, name='flatten_1')
nn = tf.layers.dropout(nn, rate=0.5)
nn = tf.layers.batch_normalization(nn, name='batchnorm_6')
nn = tf.layers.dense(nn, 512, activation=tf.nn.relu, name='dense_1')
nn = tf.layers.dropout(nn, 0.3)
nn = tf.layers.dense(nn, 512, activation=tf.nn.relu, name='dense_2')
nn = tf.layers.dropout(nn, 0.3)
nn = tf.layers.dense(nn,
last_layer,
activation=tf.nn.tanh,
name='output')
return nn
def Conv_Dense_0001_d(DataCenter, trainable=False):
tfInitial.initialize_placeholders(DataCenter)
if trainable == False:
print('Trainable is set to False for Transfer Layers')
scope = DataCenter.transfer_model_scope
last_layer = DataCenter.train_output_data.shape[1]
kern_init = tf.initializers.truncated_normal(stddev=0.05)
with tf.variable_scope(scope):
x = DataCenter.x_placeholder
nn = tf.layers.conv1d(x,
filters=4,
kernel_size=15,
name='conv1_1',
padding='same',
trainable=trainable,
kernel_initializer=kern_init)
nn = tf.layers.conv1d(nn,
filters=4,
kernel_size=15,
name='conv1_2',
padding='same',
trainable=trainable,
kernel_initializer=kern_init)
nn = tf.layers.conv1d(nn,
filters=4,
kernel_size=15,
name='conv1_3',
padding='same',
trainable=trainable,
kernel_initializer=kern_init)
nn = tf.layers.max_pooling1d(nn,
pool_size=2,
strides=2,
name='max_pool1')
nn = tf.layers.dropout(nn, 0.5)
nn = tf.layers.batch_normalization(nn,
name='batchnorm_1',
trainable=trainable)
nn = tf.layers.conv1d(nn,
filters=8,
kernel_size=15,
activation=tf.nn.relu,
padding='same',
name='conv2_1',
trainable=trainable,
kernel_initializer=kern_init)
nn = tf.layers.conv1d(nn,
filters=8,
kernel_size=15,
activation=tf.nn.relu,
padding='same',
name='conv2_2',
trainable=trainable,
kernel_initializer=kern_init)
nn = tf.layers.conv1d(nn,
filters=8,
kernel_size=15,
activation=tf.nn.relu,
padding='same',
name='conv2_3',
trainable=trainable,
kernel_initializer=kern_init)
nn = tf.layers.max_pooling1d(nn,
pool_size=2,
strides=2,
name='max_pool2')
nn = tf.layers.dropout(nn, 0.5)
nn = tf.layers.batch_normalization(nn,
name='batchnorm_2',
trainable=trainable)
nn = tf.layers.conv1d(nn,
filters=16,
kernel_size=15,
activation=tf.nn.relu,
padding='same',
name='conv3_1',
trainable=trainable,
kernel_initializer=kern_init)
nn = tf.layers.conv1d(nn,
filters=16,
kernel_size=15,
activation=tf.nn.relu,
padding='same',
name='conv3_2',
trainable=trainable,
kernel_initializer=kern_init)
nn = tf.layers.conv1d(nn,
filters=16,
kernel_size=15,
activation=tf.nn.relu,
padding='same',
name='conv3_3',
trainable=trainable,
kernel_initializer=kern_init)
nn = tf.layers.dropout(nn, 0.5)
nn = tf.layers.batch_normalization(nn,
name='batchnorm_3',
trainable=trainable)
nn = tf.layers.flatten(nn, name='flatten_1')
nn = tf.layers.dense(nn,
64,
activation=tf.nn.relu,
name='dense_1',
kernel_initializer=kern_init)
nn = tf.layers.dropout(nn, 0.3)
nn = tf.layers.dense(nn,
32,
activation=tf.nn.relu,
name='dense_2',
kernel_initializer=kern_init)
nn = tf.layers.dropout(nn, 0.3)
nn = tf.layers.dense(nn,
last_layer,
activation=tf.nn.tanh,
name='output',
kernel_initializer=kern_init)
return nn
def Conv_Dense_0001_b(DataCenter, trainable=False):
tfInitial.initialize_placeholders(DataCenter)
if trainable == False:
print('Trainable is set to False for Transfer Layers')
activ = tf.nn.relu
scope = DataCenter.transfer_model_scope
last_layer = DataCenter.train_output_data.shape[1]
with tf.variable_scope(scope):
x = DataCenter.x_placeholder
conv1_7_7 = tf.layers.conv1d(x,
64,
7,
strides=2,
padding='same',
activation=activ,
name='conv1_7_7_s2',
trainable=trainable)
pool1_3_3 = tf.layers.max_pooling1d(conv1_7_7, 3, strides=2)
pool1_3_3 = tf.layers.batch_normalization(pool1_3_3,
name='batchnorm_1',
trainable=trainable)
conv2_3_3_reduce = tf.layers.conv1d(pool1_3_3,
64,
1,
padding='same',
activation=activ,
name='conv2_3_3_reduce',
trainable=trainable)
conv2_3_3 = tf.layers.conv1d(conv2_3_3_reduce,
192,
3,
padding='same',
activation=activ,
name='conv2_3_3',
trainable=trainable)
conv2_3_3 = tf.layers.batch_normalization(conv2_3_3)
pool2_3_3 = tf.layers.max_pooling1d(conv2_3_3,
pool_size=3,
strides=2,
name='pool2_3_3_s2')
inception_3a_1_1 = tf.layers.conv1d(pool2_3_3,
64,
1,
padding='same',
activation=activ,
name='inception_3a_1_1',
trainable=trainable)
inception_3a_3_3_reduce = tf.layers.conv1d(
pool2_3_3,
96,
1,
padding='same',
activation=activ,
name='inception_3a_3_3_reduce',
trainable=trainable)
inception_3a_3_3 = tf.layers.conv1d(inception_3a_3_3_reduce,
128,
padding='same',
kernel_size=3,
activation=activ,
name='inception_3a_3_3',
trainable=trainable)
inception_3a_5_5_reduce = tf.layers.conv1d(
pool2_3_3,
16,
padding='same',
kernel_size=1,
activation=activ,
name='inception_3a_5_5_reduce',
trainable=trainable)
inception_3a_5_5 = tf.layers.conv1d(inception_3a_5_5_reduce,
32,
padding='same',
kernel_size=5,
activation=activ,
name='inception_3a_5_5',
trainable=trainable)
inception_3a_pool = tf.layers.max_pooling1d(pool2_3_3,
pool_size=3,
strides=1,
padding='same')
inception_3a_pool_1_1 = tf.layers.conv1d(inception_3a_pool,
32,
kernel_size=1,
padding='same',
activation=activ,
name='inception_3a_pool_1_1',
trainable=trainable)
# merge the inception_3a__
inception_3a_output = tf.concat([
inception_3a_1_1, inception_3a_3_3, inception_3a_5_5,
inception_3a_pool_1_1
],
axis=2)
inception_3b_1_1 = tf.layers.conv1d(inception_3a_output,
128,
kernel_size=1,
padding='same',
activation=activ,
name='inception_3b_1_1',
trainable=trainable)
inception_3b_3_3_reduce = tf.layers.conv1d(
inception_3a_output,
128,
kernel_size=1,
padding='same',
activation=activ,
name='inception_3b_3_3_reduce',
trainable=trainable)
inception_3b_3_3 = tf.layers.conv1d(inception_3b_3_3_reduce,
192,
kernel_size=3,
padding='same',
activation=activ,
name='inception_3b_3_3',
trainable=trainable)
inception_3b_5_5_reduce = tf.layers.conv1d(
inception_3a_output,
32,
kernel_size=1,
padding='same',
activation=activ,
name='inception_3b_5_5_reduce',
trainable=trainable)
inception_3b_5_5 = tf.layers.conv1d(inception_3b_5_5_reduce,
96,
kernel_size=5,
padding='same',
activation=activ,
name='inception_3b_5_5',
trainable=trainable)
inception_3b_pool = tf.layers.max_pooling1d(inception_3a_output,
pool_size=3,
strides=1,
padding='same',
name='inception_3b_pool')
inception_3b_pool_1_1 = tf.layers.conv1d(inception_3b_pool,
64,
kernel_size=1,
padding='same',
activation=activ,
name='inception_3b_pool_1_1',
trainable=trainable)
# merge the inception_3b_*
inception_3b_output = tf.concat([
inception_3b_1_1, inception_3b_3_3, inception_3b_5_5,
inception_3b_pool_1_1
],
axis=2,
name='inception_3b_output')
pool3_3_3 = tf.layers.max_pooling1d(inception_3b_output,
pool_size=3,
strides=2,
name='pool3_3_3')
inception_4a_1_1 = tf.layers.conv1d(pool3_3_3,
192,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4a_1_1',
trainable=trainable)
inception_4a_3_3_reduce = tf.layers.conv1d(
pool3_3_3,
96,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4a_3_3_reduce',
trainable=trainable)
inception_4a_3_3 = tf.layers.conv1d(inception_4a_3_3_reduce,
208,
kernel_size=3,
padding='same',
activation=activ,
name='inception_4a_3_3',
trainable=trainable)
inception_4a_5_5_reduce = tf.layers.conv1d(
pool3_3_3,
16,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4a_5_5_reduce',
trainable=trainable)
inception_4a_5_5 = tf.layers.conv1d(inception_4a_5_5_reduce,
48,
kernel_size=5,
padding='same',
activation=activ,
name='inception_4a_5_5',
trainable=trainable)
inception_4a_pool = tf.layers.max_pooling1d(pool3_3_3,
pool_size=3,
strides=1,
padding='same',
name='inception_4a_pool')
inception_4a_pool_1_1 = tf.layers.conv1d(inception_4a_pool,
64,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4a_pool_1_1',
trainable=trainable)
inception_4a_output = tf.concat([
inception_4a_1_1, inception_4a_3_3, inception_4a_5_5,
inception_4a_pool_1_1
],
axis=2,
name='inception_4a_output')
inception_4b_1_1 = tf.layers.conv1d(inception_4a_output,
160,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4b_1_1',
trainable=trainable)
inception_4b_3_3_reduce = tf.layers.conv1d(
inception_4a_output,
112,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4b_3_3_reduce',
trainable=trainable)
inception_4b_3_3 = tf.layers.conv1d(inception_4b_3_3_reduce,
224,
kernel_size=3,
padding='same',
activation=activ,
name='inception_4b_3_3',
trainable=trainable)
inception_4b_5_5_reduce = tf.layers.conv1d(
inception_4a_output,
24,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4b_5_5_reduce',
trainable=trainable)
inception_4b_5_5 = tf.layers.conv1d(inception_4b_5_5_reduce,
64,
kernel_size=5,
padding='same',
activation=activ,
name='inception_4b_5_5',
trainable=trainable)
inception_4b_pool = tf.layers.max_pooling1d(inception_4a_output,
pool_size=3,
strides=1,
padding='same',
name='inception_4b_pool')
inception_4b_pool_1_1 = tf.layers.conv1d(inception_4b_pool,
64,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4b_pool_1_1',
trainable=trainable)
inception_4b_output = tf.concat([
inception_4b_1_1, inception_4b_3_3, inception_4b_5_5,
inception_4b_pool_1_1
],
axis=2,
name='inception_4b_output')
inception_4c_1_1 = tf.layers.conv1d(inception_4b_output,
128,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4c_1_1',
trainable=trainable)
inception_4c_3_3_reduce = tf.layers.conv1d(
inception_4b_output,
128,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4c_3_3_reduce',
trainable=trainable)
inception_4c_3_3 = tf.layers.conv1d(inception_4c_3_3_reduce,
256,
kernel_size=3,
padding='same',
activation=activ,
name='inception_4c_3_3',
trainable=trainable)
inception_4c_5_5_reduce = tf.layers.conv1d(
inception_4b_output,
24,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4c_5_5_reduce',
trainable=trainable)
inception_4c_5_5 = tf.layers.conv1d(inception_4c_5_5_reduce,
64,
kernel_size=5,
padding='same',
activation=activ,
name='inception_4c_5_5',
trainable=trainable)
inception_4c_pool = tf.layers.max_pooling1d(inception_4b_output,
pool_size=3,
strides=1,
padding='same')
inception_4c_pool_1_1 = tf.layers.conv1d(inception_4c_pool,
64,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4c_pool_1_1',
trainable=trainable)
inception_4c_output = tf.concat([
inception_4c_1_1, inception_4c_3_3, inception_4c_5_5,
inception_4c_pool_1_1
],
axis=2,
name='inception_4c_output')
inception_4d_1_1 = tf.layers.conv1d(inception_4c_output,
112,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4d_1_1',
trainable=trainable)
inception_4d_3_3_reduce = tf.layers.conv1d(
inception_4c_output,
144,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4d_3_3_reduce',
trainable=trainable)
inception_4d_3_3 = tf.layers.conv1d(inception_4d_3_3_reduce,
288,
kernel_size=3,
padding='same',
activation=activ,
name='inception_4d_3_3',
trainable=trainable)
inception_4d_5_5_reduce = tf.layers.conv1d(
inception_4c_output,
32,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4d_5_5_reduce',
trainable=trainable)
inception_4d_5_5 = tf.layers.conv1d(inception_4d_5_5_reduce,
64,
kernel_size=5,
padding='same',
activation=activ,
name='inception_4d_5_5',
trainable=trainable)
inception_4d_pool = tf.layers.max_pooling1d(inception_4c_output,
pool_size=3,
strides=1,
padding='same',
name='inception_4d_pool')
inception_4d_pool_1_1 = tf.layers.conv1d(inception_4d_pool,
64,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4d_pool_1_1',
trainable=trainable)
inception_4d_output = tf.concat([
inception_4d_1_1, inception_4d_3_3, inception_4d_5_5,
inception_4d_pool_1_1
],
axis=2,
name='inception_4d_output')
inception_4e_1_1 = tf.layers.conv1d(inception_4d_output,
256,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4e_1_1',
trainable=trainable)
inception_4e_3_3_reduce = tf.layers.conv1d(
inception_4d_output,
160,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4e_3_3_reduce',
trainable=trainable)
inception_4e_3_3 = tf.layers.conv1d(inception_4e_3_3_reduce,
320,
kernel_size=3,
padding='same',
activation=activ,
name='inception_4e_3_3',
trainable=trainable)
inception_4e_5_5_reduce = tf.layers.conv1d(
inception_4d_output,
32,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4e_5_5_reduce',
trainable=trainable)
inception_4e_5_5 = tf.layers.conv1d(inception_4e_5_5_reduce,
128,
kernel_size=5,
padding='same',
activation=activ,
name='inception_4e_5_5',
trainable=trainable)
inception_4e_pool = tf.layers.max_pooling1d(inception_4d_output,
pool_size=3,
strides=1,
padding='same',
name='inception_4e_pool')
inception_4e_pool_1_1 = tf.layers.conv1d(inception_4e_pool,
128,
kernel_size=1,
padding='same',
activation=activ,
name='inception_4e_pool_1_1',
trainable=trainable)
inception_4e_output = tf.concat([
inception_4e_1_1, inception_4e_3_3, inception_4e_5_5,
inception_4e_pool_1_1
],
axis=2)
pool4_3_3 = tf.layers.max_pooling1d(inception_4e_output,
pool_size=3,
strides=2,
name='pool_3_3')
inception_5a_1_1 = tf.layers.conv1d(pool4_3_3,
256,
kernel_size=1,
padding='same',
activation=activ,
name='inception_5a_1_1',
trainable=trainable)
inception_5a_3_3_reduce = tf.layers.conv1d(
pool4_3_3,
160,
kernel_size=1,
padding='same',
activation=activ,
name='inception_5a_3_3_reduce',
trainable=trainable)
inception_5a_3_3 = tf.layers.conv1d(inception_5a_3_3_reduce,
320,
kernel_size=3,
padding='same',
activation=activ,
name='inception_5a_3_3',
trainable=trainable)
inception_5a_5_5_reduce = tf.layers.conv1d(
pool4_3_3,
32,
kernel_size=1,
padding='same',
activation=activ,
name='inception_5a_5_5_reduce',
trainable=trainable)
inception_5a_5_5 = tf.layers.conv1d(inception_5a_5_5_reduce,
128,
kernel_size=5,
padding='same',
activation=activ,
name='inception_5a_5_5',
trainable=trainable)
inception_5a_pool = tf.layers.max_pooling1d(pool4_3_3,
pool_size=3,
strides=1,
padding='same',
name='inception_5a_pool')
inception_5a_pool_1_1 = tf.layers.conv1d(inception_5a_pool,
128,
kernel_size=1,
padding='same',
activation=activ,
name='inception_5a_pool_1_1',
trainable=trainable)
inception_5a_output = tf.concat([
inception_5a_1_1, inception_5a_3_3, inception_5a_5_5,
inception_5a_pool_1_1
],
axis=2)
inception_5b_1_1 = tf.layers.conv1d(inception_5a_output,
384,
kernel_size=1,
padding='same',
activation=activ,
name='inception_5b_1_1',
trainable=trainable)
inception_5b_3_3_reduce = tf.layers.conv1d(
inception_5a_output,
192,
kernel_size=1,
padding='same',
activation=activ,
name='inception_5b_3_3_reduce',
trainable=trainable)
inception_5b_3_3 = tf.layers.conv1d(inception_5b_3_3_reduce,
384,
kernel_size=3,
padding='same',
activation=activ,
name='inception_5b_3_3',
trainable=trainable)
inception_5b_5_5_reduce = tf.layers.conv1d(
inception_5a_output,
48,
kernel_size=1,
padding='same',
activation=activ,
name='inception_5b_5_5_reduce',
trainable=trainable)
inception_5b_5_5 = tf.layers.conv1d(inception_5b_5_5_reduce,
128,
kernel_size=5,
padding='same',
activation=activ,
name='inception_5b_5_5',
trainable=trainable)
inception_5b_pool = tf.layers.max_pooling1d(inception_5a_output,
pool_size=3,
strides=1,
padding='same',
name='inception_5b_pool')
inception_5b_pool_1_1 = tf.layers.conv1d(inception_5b_pool,
128,
kernel_size=1,
padding='same',
activation=activ,
name='inception_5b_pool_1_1',
trainable=trainable)
inception_5b_output = tf.concat([
inception_5b_1_1, inception_5b_3_3, inception_5b_5_5,
inception_5b_pool_1_1
],
axis=2)
pool5_7_7 = tf.layers.max_pooling1d(inception_5b_output,
pool_size=7,
strides=1)
nn = tf.layers.flatten(pool5_7_7)
nn = tf.layers.dropout(nn, rate=0.5)
nn = tf.layers.batch_normalization(nn,
name='batchnorm_5',
trainable=True)
nn = tf.layers.dense(nn,
4096,
activation=tf.nn.relu,
name='dense_1',
trainable=True)
nn = tf.layers.dropout(nn, rate=0.5)
nn = tf.layers.batch_normalization(nn,
name='batchnorm_6',
trainable=True)
nn = tf.layers.dense(nn,
4096,
activation=tf.nn.relu,
name='dense_2',
trainable=True)
nn = tf.layers.dropout(nn, rate=0.5)
nn = tf.layers.batch_normalization(nn,
name='batchnorm_7',
trainable=True)
nn = tf.layers.dense(nn,
4096,
activation=tf.nn.relu,
name='dense_3',
trainable=True)
nn = tf.layers.dropout(nn, rate=0.5)
nn = tf.layers.batch_normalization(nn,
name='batchnorm_8',
trainable=True)
nn = tf.layers.dense(nn, last_layer, name='output')
return nn