def construct_CNN(target_length,numConv,kernel_num,kernel_size,dropout,maxpool = False,
maxpool_size=1 , add_dense_layer = False, dense_unit = None,normalization = False):
# create model
model=Sequential()
model.add(Dropout(0.2))
if numConv != len(kernel_size) and len(kernel_num) != len(kernel_size):
print('Incompatible number of kernel sizes with number of Conv layer!')
print('Incompatible number of filters with number of Conv layer!')
#Construct Convolutional Layers
for n in range(numConv):
model.add(Conv1D(kernel_num[n], kernel_size = kernel_size[n],padding = 'same',
input_shape = (4, target_length/(maxpool_size ** n)), activation = 'relu'))
model.add(Dropout(dropout))
if maxpool:
model.add(MaxPooling1D(pool_size = maxpool_size, padding='same'))
model.add(Dropout(dropout))
if normalization:
model.add(BatchNormalization())
# Flatten the network
model.add(Flatten())
model.add(Dropout(dropout))
#Construct Dense Layer
if add_dense_layer:
for n in range(len(dense_unit)):
model.add(Dense(dense_unit[n],activation = 'relu'))
model.add(Dropout(0.2))
model.add(Dense(1))
return model
def create_model1(optimizer,metric=['accuracy']):
##CNN 1
model_1 = Sequential()
#Convolutional Layers
model_1.add(Reshape((62, 1), input_shape=(62,)))
model_1.add(Conv1D(31, kernel_size=2, strides=2, padding="same", activation = 'relu'))
#Dense Layers
model_1.add(Flatten())
model_1.add(Dense(units=15, activation='relu'))
model_1.add(Dense(units=1, activation='sigmoid'))
model_1.compile(loss='binary_crossentropy', optimizer=optimizer, metrics=metric)
return model_1
def _init_conv(self, filters, name):
return Conv1D(
filters=filters,
kernel_size=self.kernel_size,
strides=self.strides,
padding='same',
dilation_rate=self.dilation_rate,
activation=self.activation,
use_bias=self.use_bias,
kernel_initializer=self.kernel_initializer,
bias_initializer=self.bias_initializer,
kernel_regularizer=self.kernel_regularizer,
bias_regularizer=self.bias_regularizer,
activity_regularizer=self.activity_regularizer,
kernel_constraint=self.kernel_constraint,
bias_constraint=self.bias_constraint,
name=name,
)
def create_model(use_case,
model_num,
optimizer,
dropout=False,
metric=['accuracy']):
if use_case == "V2_noClaim" or use_case == "V2_withClaim":
if model_num == 1:
model_1 = Sequential()
#Convolutional Layers
model_1.add(Reshape((62, 1), input_shape=(62, )))
model_1.add(
Conv1D(31,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
#Dense Layers
model_1.add(Flatten())
model_1.add(Dense(units=15, activation='relu'))
if dropout is not False:
model_1.add(Dropout(dropout))
model_1.add(Dense(units=1, activation='sigmoid'))
model_1.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=metric)
return model_1
elif model_num == 2:
model_2 = Sequential()
#Convolutional Layers
model_2.add(Reshape((62, 1), input_shape=(62, )))
model_2.add(
Conv1D(60,
kernel_size=3,
strides=1,
padding="same",
activation='relu'))
model_2.add(
Conv1D(30,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
#Dense Layers
model_2.add(Flatten())
model_2.add(Dense(units=15, activation='relu'))
if dropout is not False:
model_2.add(Dropout(dropout))
model_2.add(Dense(units=1, activation='sigmoid'))
model_2.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
return model_2
elif model_num == 3:
model_3 = Sequential()
#Convolutional Layers
model_3.add(Reshape((62, 1), input_shape=(62, )))
model_3.add(
Conv1D(61,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
model_3.add(
Conv1D(60,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
model_3.add(
Conv1D(30,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
#Dense Layers
model_3.add(Flatten())
model_3.add(Dense(units=15, activation='relu'))
if dropout is not False:
model_3.add(Dropout(dropout))
model_3.add(Dense(units=1, activation='sigmoid'))
model_3.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
return model_3
elif model_num == 4:
model_4 = Sequential()
#Convolutional Layers
model_4.add(Reshape((62, 1), input_shape=(62, )))
model_4.add(
Conv1D(60,
kernel_size=3,
strides=1,
padding="same",
activation='relu'))
model_4.add(
Conv1D(30,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
#Dense Layers
model_4.add(Flatten())
model_4.add(Dense(units=15, activation='relu'))
if dropout is not False:
model_4.add(Dropout(dropout))
model_4.add(Dense(units=10, activation='relu'))
if dropout is not False:
model_4.add(Dropout(dropout))
model_4.add(Dense(units=5, activation='relu'))
if dropout is not False:
model_4.add(Dropout(dropout))
model_4.add(Dense(units=1, activation='sigmoid'))
model_4.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
return model_4
elif model_num == 5:
model_5 = Sequential()
#Convolutional Layers
model_5.add(Reshape((62, 1), input_shape=(62, )))
model_5.add(
Conv1D(31,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
model_5.add(
Conv1D(30,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
#Dense Layers
model_5.add(Flatten())
model_5.add(Dense(units=15, activation='relu'))
if dropout is not False:
model_5.add(Dropout(dropout))
model_5.add(Dense(units=1, activation='sigmoid'))
model_5.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
return model_5
elif model_num == 6:
model_6 = Sequential()
#Convolutional Layers
model_6.add(Reshape((62, 1), input_shape=(62, )))
model_6.add(
Conv1D(61,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
model_6.add(
Conv1D(60,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
model_6.add(
Conv1D(30,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
model_6.add(
Conv1D(10,
kernel_size=3,
strides=3,
padding="same",
activation='relu'))
#Dense Layers
model_6.add(Flatten())
model_6.add(Dense(units=15, activation='relu'))
if dropout is not False:
model_6.add(Dropout(dropout))
model_6.add(Dense(units=1, activation='sigmoid'))
model_6.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
return model_6
elif model_num == 7:
model_7 = Sequential()
#Convolutional Layers
model_7.add(Reshape((62, 1), input_shape=(62, )))
model_7.add(
Conv1D(61,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
model_7.add(
Conv1D(60,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
model_7.add(
Conv1D(60,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
model_7.add(
Conv1D(20,
kernel_size=3,
strides=3,
padding="same",
activation='relu'))
#Dense Layers
model_7.add(Flatten())
model_7.add(Dense(units=30, activation='relu'))
if dropout is not False:
model_7.add(Dropout(dropout))
model_7.add(Dense(units=15, activation='relu'))
if dropout is not False:
model_7.add(Dropout(dropout))
model_7.add(Dense(units=1, activation='sigmoid'))
model_7.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
return model_7
elif model_num == 8:
model_8 = Sequential()
#Convolutional Layers
model_8.add(Reshape((62, 1), input_shape=(62, )))
model_8.add(
Conv1D(61,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
model_8.add(
Conv1D(60,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
model_8.add(
Conv1D(59,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
model_8.add(
Conv1D(40,
kernel_size=20,
strides=1,
padding="same",
activation='relu'))
model_8.add(
Conv1D(20,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
#Dense Layers
model_8.add(Flatten())
model_8.add(Dense(units=30, activation='relu'))
if dropout is not False:
model_8.add(Dropout(dropout))
model_8.add(Dense(units=1, activation='sigmoid'))
model_8.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
return model_8
elif model_num == 9:
model_9 = Sequential()
#Convolutional Layers
model_9.add(Reshape((62, 1), input_shape=(62, )))
model_9.add(
Conv1D(61,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
model_9.add(
Conv1D(60,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
model_9.add(
Conv1D(59,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
model_9.add(
Conv1D(40,
kernel_size=20,
strides=1,
padding="same",
activation='relu'))
model_9.add(
Conv1D(20,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
#Dense Layers
model_9.add(Flatten())
model_9.add(Dense(units=30, activation='relu'))
if dropout is not False:
model_9.add(Dropout(dropout))
model_9.add(Dense(units=1, activation='sigmoid'))
model_9.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
return model_9
elif model_num == 10:
model_10 = Sequential()
#Convolutional Layers
model_10.add(Reshape((62, 1), input_shape=(62, )))
model_10.add(
Conv1D(31,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
model_10.add(
Conv1D(20,
kernel_size=11,
strides=1,
padding="same",
activation='relu'))
#Dense Layers
model_10.add(Flatten())
model_10.add(Dense(units=30, activation='relu'))
if dropout is not False:
model_10.add(Dropout(dropout))
model_10.add(Dense(units=1, activation='sigmoid'))
model_10.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
return model_10
else:
print("Wrong model num : architecture not defined")
elif use_case == "V1_withClaim":
if use_case == "V1_withClaim":
input_dim = 97
# elif use_case == "V1_noClaim":
# input_dim = 55
else:
print("Error - Incorrect data")
if model_num == 1:
model_1 = Sequential()
#Convolutional Layers
model_1.add(Reshape((input_dim, 1), input_shape=(input_dim, )))
model_1.add(
Conv1D(20,
kernel_size=5,
strides=5,
padding="same",
activation='relu'))
#Dense Layers
model_1.add(Flatten())
model_1.add(Dense(units=30, activation='relu'))
if dropout is not False:
model_1.add(Dropout(dropout))
model_1.add(Dense(units=15, activation='relu'))
if dropout is not False:
model_1.add(Dropout(dropout))
model_1.add(Dense(units=1, activation='sigmoid'))
model_1.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
return model_1
elif model_num == 2:
model_2 = Sequential()
#Convolutional Layers
model_2.add(Reshape((input_dim, 1), input_shape=(input_dim, )))
model_2.add(
Conv1D(48,
kernel_size=3,
strides=2,
padding="same",
activation='relu'))
model_2.add(
Conv1D(20,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
#Dense Layers
model_2.add(Flatten())
model_2.add(Dense(units=30, activation='relu'))
if dropout is not False:
model_2.add(Dropout(dropout))
model_2.add(Dense(units=15, activation='relu'))
if dropout is not False:
model_2.add(Dropout(dropout))
model_2.add(Dense(units=1, activation='sigmoid'))
model_2.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
return model_2
elif model_num == 3:
model_3 = Sequential()
#Convolutional Layers
model_3.add(Reshape((input_dim, 1), input_shape=(input_dim, )))
model_3.add(
Conv1D(48,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
model_3.add(
Conv1D(40,
kernel_size=8,
strides=1,
padding="same",
activation='relu'))
model_3.add(
Conv1D(20,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
#Dense Layers
model_3.add(Flatten())
model_3.add(Dense(units=30, activation='relu'))
if dropout is not False:
model_3.add(Dropout(dropout))
model_3.add(Dense(units=15, activation='relu'))
if dropout is not False:
model_3.add(Dropout(dropout))
model_3.add(Dense(units=1, activation='sigmoid'))
model_3.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
return model_3
elif model_num == 4:
model_4 = Sequential()
#Convolutional Layers
model_4.add(Reshape((input_dim, 1), input_shape=(input_dim, )))
model_4.add(
Conv1D(25,
kernel_size=4,
strides=4,
padding="same",
activation='relu'))
#Dense Layers
model_4.add(Flatten())
model_4.add(Dense(units=30, activation='relu'))
if dropout is not False:
model_4.add(Dropout(dropout))
model_4.add(Dense(units=15, activation='relu'))
if dropout is not False:
model_4.add(Dropout(dropout))
model_4.add(Dense(units=1, activation='sigmoid'))
model_4.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
return model_4
elif model_num == 5:
model_5 = Sequential()
#Convolutional Layers
model_5.add(Reshape((input_dim, 1), input_shape=(input_dim, )))
model_5.add(
Conv1D(33,
kernel_size=3,
strides=3,
padding="same",
activation='relu'))
model_5.add(
Conv1D(20,
kernel_size=14,
strides=1,
padding="same",
activation='relu'))
#Dense Layers
model_5.add(Flatten())
model_5.add(Dense(units=30, activation='relu'))
if dropout is not False:
model_5.add(Dropout(dropout))
model_5.add(Dense(units=15, activation='relu'))
if dropout is not False:
model_5.add(Dropout(dropout))
model_5.add(Dense(units=1, activation='sigmoid'))
model_5.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
return model_5
elif model_num == 6:
model_6 = Sequential()
#Convolutional Layers
model_6.add(Reshape((input_dim, 1), input_shape=(input_dim, )))
model_6.add(
Conv1D(48,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
model_6.add(
Conv1D(40,
kernel_size=8,
strides=1,
padding="same",
activation='relu'))
model_6.add(
Conv1D(20,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
model_6.add(
Conv1D(10,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
#Dense Layers
model_6.add(Flatten())
model_6.add(Dense(units=30, activation='relu'))
if dropout is not False:
model_6.add(Dropout(dropout))
model_6.add(Dense(units=15, activation='relu'))
if dropout is not False:
model_6.add(Dropout(dropout))
model_6.add(Dense(units=1, activation='sigmoid'))
model_6.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
return model_6
else:
print("Wrong model num : architecture not defined")
elif use_case == "V1_noClaim":
if model_num == 1:
model_1 = Sequential()
#Convolutional Layers
model_1.add(Reshape((55, 1), input_shape=(55, )))
model_1.add(
Conv1D(50,
kernel_size=6,
strides=1,
padding="same",
activation='relu'))
model_1.add(
Conv1D(40,
kernel_size=11,
strides=1,
padding="same",
activation='relu'))
model_1.add(
Conv1D(20,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
#Dense Layers
model_1.add(Flatten())
model_1.add(Dense(units=30, activation='relu'))
if dropout is not False:
model_1.add(Dropout(dropout))
model_1.add(Dense(units=15, activation='relu'))
if dropout is not False:
model_1.add(Dropout(dropout))
model_1.add(Dense(units=1, activation='sigmoid'))
model_1.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
return model_1
elif model_num == 2:
model_2 = Sequential()
#Convolutional Layers
model_2.add(Reshape((55, 1), input_shape=(55, )))
model_2.add(
Conv1D(11,
kernel_size=5,
strides=5,
padding="same",
activation='relu'))
#Dense Layers
model_2.add(Flatten())
model_2.add(Dense(units=30, activation='relu'))
if dropout is not False:
model_2.add(Dropout(dropout))
model_2.add(Dense(units=15, activation='relu'))
if dropout is not False:
model_2.add(Dropout(dropout))
model_2.add(Dense(units=1, activation='sigmoid'))
model_2.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
return model_2
elif model_num == 3:
model_3 = Sequential()
#Convolutional Layers
model_3.add(Reshape((55, 1), input_shape=(55, )))
model_3.add(
Conv1D(50,
kernel_size=5,
strides=1,
padding="same",
activation='relu'))
model_3.add(
Conv1D(24,
kernel_size=4,
strides=5,
padding="same",
activation='relu'))
#Dense Layers
model_3.add(Flatten())
model_3.add(Dense(units=35, activation='relu'))
if dropout is not False:
model_3.add(Dropout(dropout))
model_3.add(Dense(units=15, activation='relu'))
if dropout is not False:
model_3.add(Dropout(dropout))
model_3.add(Dense(units=1, activation='sigmoid'))
model_3.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
return model_3
elif model_num == 4:
model_4 = Sequential()
#Convolutional Layers
model_4.add(Reshape((55, 1), input_shape=(55, )))
model_4.add(
Conv1D(50,
kernel_size=5,
strides=1,
padding="same",
activation='relu'))
model_4.add(
Conv1D(24,
kernel_size=5,
strides=2,
padding="same",
activation='relu'))
#Dense Layers
model_4.add(Flatten())
model_4.add(Dense(units=30, activation='relu'))
if dropout is not False:
model_4.add(Dropout(dropout))
model_4.add(Dense(units=15, activation='relu'))
if dropout is not False:
model_4.add(Dropout(dropout))
model_4.add(Dense(units=1, activation='sigmoid'))
model_4.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
return model_4
elif model_num == 5:
model_5 = Sequential()
#Convolutional Layers
model_5.add(Reshape((55, 1), input_shape=(55, )))
model_5.add(
Conv1D(28,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
model_5.add(
Conv1D(20,
kernel_size=9,
strides=1,
padding="same",
activation='relu'))
#Dense Layers
model_5.add(Flatten())
model_5.add(Dense(units=30, activation='relu'))
if dropout is not False:
model_5.add(Dropout(dropout))
model_5.add(Dense(units=15, activation='relu'))
if dropout is not False:
model_5.add(Dropout(dropout))
model_5.add(Dense(units=1, activation='sigmoid'))
model_5.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
return model_5
elif model_num == 6:
model_6 = Sequential()
#Convolutional Layers
model_6.add(Reshape((55, 1), input_shape=(55, )))
model_6.add(
Conv1D(48,
kernel_size=8,
strides=1,
padding="same",
activation='relu'))
model_6.add(
Conv1D(40,
kernel_size=8,
strides=1,
padding="same",
activation='relu'))
model_6.add(
Conv1D(20,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
model_6.add(
Conv1D(10,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
#Dense Layers
model_6.add(Flatten())
model_6.add(Dense(units=30, activation='relu'))
if dropout is not False:
model_6.add(Dropout(dropout))
model_6.add(Dense(units=15, activation='relu'))
if dropout is not False:
model_6.add(Dropout(dropout))
model_6.add(Dense(units=1, activation='sigmoid'))
model_6.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
return model_6
else:
print("Wrong model num : architecture not defined")
else:
print("Wrong Use case! You entered :")
print(use_case)
#0.0 1463
#1.0 486
pd.value_counts(y_test)
#0.0 794
#1.0 256
############################################################################
###Modelling
os.chdir(model_path)
os.getcwd()
##CNN 1
model = Sequential()
#Convolutional Layers
model.add(Reshape((55, 1)))
model.add(
Conv1D(50, kernel_size=5, strides=1, padding="same", activation='relu'))
#Dense Layers
model.add(Flatten())
model.add(Dense(units=60, input_dim=55, activation='relu'))
#model.add(Dropout(0.1))
model.add(Dense(units=30, activation='relu'))
#model.add(Dropout(0.1))
model.add(Dense(units=15, activation='relu'))
#model.add(Dropout(0.3))
model.add(Dense(units=1, activation='sigmoid'))
model.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
def __init__(self, fl, mode, hparams):
"""
Initialises new DNN model based on input features_dim, labels_dim, hparams
:param features_dim: Number of input feature nodes. Integer
:param labels_dim: Number of output label nodes. Integer
:param hparams: Dict containing hyperparameter information. Dict can be created using create_hparams() function.
hparams includes: hidden_layers: List containing number of nodes in each hidden layer. [10, 20] means 10 then 20 nodes.
"""
# self.features_dim = fl.features_c_dim
# self.labels_dim = fl.labels_dim # Assuming that each task has only 1 dimensional output
self.features_dim = fl.features_c_dim + 1 # 1 for the positional argument
self.labels_dim = 1
self.numel = fl.labels.shape[1] + 1
self.hparams = hparams
self.mode = mode
self.normalise_labels = fl.normalise_labels
self.labels_scaler = fl.labels_scaler
features_in = Input(shape=(self.features_dim, ),
name='main_features_c_input')
# Selection of model
if mode == 'ann':
model = ann(self.features_dim, self.labels_dim, self.hparams)
x = model(features_in)
self.model = Model(inputs=features_in, outputs=x)
elif mode == 'ann2':
model_1 = ann(self.features_dim, 50, self.hparams)
x = model_1(features_in)
model_end = ann(50, 50, self.hparams)
end = model_end(x)
end_node = Dense(units=1,
activation='linear',
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='output_layer')(end)
model_2 = ann(50, self.labels_dim - 1, self.hparams)
x = model_2(x)
self.model = Model(inputs=features_in, outputs=[end_node, x])
elif mode == 'ann3':
x = Dense(units=hparams['pre'],
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Pre_' + str(0))(features_in)
x = Dense(units=hparams['pre'],
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Pre_' + str(1))(x)
x = Dense(units=hparams['pre'],
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Pre_' + str(2))(x)
# x = BatchNormalization()(x)
x = Dense(units=1,
activation='linear',
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Pre_set_19')(x)
self.model = Model(inputs=features_in, outputs=x)
elif mode == 'conv1':
x = Dense(units=hparams['pre'],
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='shared' + str(1))(features_in)
x = Dense(units=hparams['pre'],
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Pre_' + str(1))(x)
#x = BatchNormalization()(x)
x = Dense(units=19,
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Pre_set_19')(x)
#x = BatchNormalization()(x)
x = Reshape(target_shape=(19, 1))(x)
x = Conv1D(filters=hparams['filters'],
kernel_size=3,
strides=1,
padding='same',
activation='relu')(x)
#x = BatchNormalization()(x)
x = Conv1D(filters=hparams['filters'] * 2,
kernel_size=3,
strides=1,
padding='same',
activation='relu')(x)
x = Conv1D(filters=hparams['filters'] * 4,
kernel_size=3,
strides=1,
padding='same',
activation='relu')(x)
#x = Permute((2,1))(x)
#x = GlobalAveragePooling1D()(x)
x = TimeDistributed(Dense(1, activation='linear'))(x)
x = Reshape(target_shape=(19, ))(x)
self.model = Model(inputs=features_in, outputs=x)
elif mode == 'conv2':
x = Dense(units=10,
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Shared_e_' + str(1))(features_in)
x = Dense(units=10,
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Shared_e_' + str(2))(x)
end = Dense(units=10,
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Dense_e_' + str(1))(x)
end = Dense(units=10,
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Dense_e_' + str(2))(end)
end_node = Dense(units=1,
activation='linear',
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='output_layer')(end)
x = Dense(units=80,
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Pre_' + str(1))(x)
x = Reshape(target_shape=(80, 1))(x)
x = Conv1D(filters=8,
kernel_size=3,
strides=1,
padding='same',
activation='relu')(x)
x = MaxPooling1D(pool_size=2)(x)
x = Conv1D(filters=16,
kernel_size=3,
strides=1,
padding='same',
activation='relu')(x)
x = MaxPooling1D(pool_size=2)(x)
#x = Permute((2,1))(x)
#x = GlobalAveragePooling1D()(x)
x = TimeDistributed(Dense(1, activation='linear'))(x)
x = Reshape(target_shape=(20, ))(x)
self.model = Model(inputs=features_in, outputs=[end_node, x])
elif mode == 'lstm':
x = Dense(units=20,
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Shared_e_' + str(1))(features_in)
x = Dense(units=20,
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Shared_e_' + str(2))(x)
end = Dense(units=20,
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Dense_e_' + str(1))(x)
end = Dense(units=20,
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Dense_e_' + str(2))(end)
end_node = Dense(units=1,
activation='linear',
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='output_layer')(end)
x = Dense(units=20,
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Pre_' + str(1))(x)
x = Dense(units=20,
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Pre_' + str(2))(x)
x = RepeatVector(n=20)(x)
x = LSTM(units=30, activation='relu', return_sequences=True)(x)
x = LSTM(units=30, activation='relu', return_sequences=True)(x)
x = TimeDistributed(Dense(1))(x)
x = Reshape(target_shape=(20, ))(x)
'''
x = Permute((2,1))(x)
x = GlobalAveragePooling1D()(x)
'''
self.model = Model(inputs=features_in, outputs=[end_node, x])
optimizer = Adam(clipnorm=1)
self.model.compile(optimizer=optimizer, loss='mean_squared_error')
def __init__(self, fl, mode, hparams):
"""
Initialises new DNN model based on input features_dim, labels_dim, hparams
:param features_dim: Number of input feature nodes. Integer
:param labels_dim: Number of output label nodes. Integer
:param hparams: Dict containing hyperparameter information. Dict can be created using create_hparams() function.
hparams includes: hidden_layers: List containing number of nodes in each hidden layer. [10, 20] means 10 then 20 nodes.
"""
self.features_dim = fl.features_c_dim
self.labels_dim = fl.labels_dim # Assuming that each task has only 1 dimensional output
self.hparams = hparams
self.mode = mode
self.normalise_labels = fl.normalise_labels
self.labels_scaler = fl.labels_scaler
features_in = Input(shape=(self.features_dim, ),
name='main_features_c_input')
# Selection of model
if mode == 'ann':
model = ann(self.features_dim, self.labels_dim, self.hparams)
x = model(features_in)
self.model = Model(inputs=features_in, outputs=x)
elif mode == 'ann2':
model_1 = ann(self.features_dim, 50, self.hparams)
x = model_1(features_in)
model_end = ann(50, 50, self.hparams)
end = model_end(x)
end_node = Dense(units=1,
activation='linear',
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='output_layer')(end)
model_2 = ann(50, self.labels_dim - 1, self.hparams)
x = model_2(x)
self.model = Model(inputs=features_in, outputs=[end_node, x])
elif mode == 'ann3':
x = Dense(units=hparams['pre'],
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Pre_' + str(0))(features_in)
x = Dense(units=hparams['pre'],
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Pre_' + str(1))(x)
x = Dense(units=hparams['pre'],
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Pre_' + str(2))(x)
# x = BatchNormalization()(x)
x = Dense(units=self.labels_dim,
activation='linear',
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Final')(x)
self.model = Model(inputs=features_in, outputs=x)
elif mode == 'conv1':
if fl.label_type == 'gf20':
final_dim = 20
else:
final_dim = 19
x = Dense(units=hparams['pre'],
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='shared' + str(1))(features_in)
x = Dense(units=hparams['pre'],
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Pre_' + str(1))(x)
#x = BatchNormalization()(x)
x = Dense(units=final_dim,
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Pre_set_19')(x)
#x = BatchNormalization()(x)
x = Reshape(target_shape=(final_dim, 1))(x)
x = Conv1D(filters=hparams['filters'],
kernel_size=3,
strides=1,
padding='same',
activation='relu')(x)
#x = BatchNormalization()(x)
x = Conv1D(filters=hparams['filters'] * 2,
kernel_size=3,
strides=1,
padding='same',
activation='relu')(x)
x = Conv1D(filters=hparams['filters'] * 4,
kernel_size=3,
strides=1,
padding='same',
activation='relu')(x)
#x = Permute((2,1))(x)
#x = GlobalAveragePooling1D()(x)
x = TimeDistributed(Dense(1, activation='linear'))(x)
x = Reshape(target_shape=(final_dim, ))(x)
self.model = Model(inputs=features_in, outputs=x)
elif mode == 'conv2':
x = Dense(units=10,
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Shared_e_' + str(1))(features_in)
x = Dense(units=10,
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Shared_e_' + str(2))(x)
end = Dense(units=10,
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Dense_e_' + str(1))(x)
end = Dense(units=10,
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Dense_e_' + str(2))(end)
end_node = Dense(units=1,
activation='linear',
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='output_layer')(end)
x = Dense(units=80,
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Pre_' + str(1))(x)
x = Reshape(target_shape=(80, 1))(x)
x = Conv1D(filters=8,
kernel_size=3,
strides=1,
padding='same',
activation='relu')(x)
x = MaxPooling1D(pool_size=2)(x)
x = Conv1D(filters=16,
kernel_size=3,
strides=1,
padding='same',
activation='relu')(x)
x = MaxPooling1D(pool_size=2)(x)
#x = Permute((2,1))(x)
#x = GlobalAveragePooling1D()(x)
x = TimeDistributed(Dense(1, activation='linear'))(x)
x = Reshape(target_shape=(20, ))(x)
self.model = Model(inputs=features_in, outputs=[end_node, x])
elif mode == 'lstm':
x = Dense(units=20,
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Shared_e_' + str(1))(features_in)
x = Dense(units=20,
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Shared_e_' + str(2))(x)
end = Dense(units=20,
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Dense_e_' + str(1))(x)
end = Dense(units=20,
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Dense_e_' + str(2))(end)
end_node = Dense(units=1,
activation='linear',
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='output_layer')(end)
x = Dense(units=20,
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Pre_' + str(1))(x)
x = Dense(units=20,
activation=hparams['activation'],
kernel_regularizer=regularizers.l1_l2(
l1=hparams['reg_l1'], l2=hparams['reg_l2']),
name='Pre_' + str(2))(x)
x = RepeatVector(n=20)(x)
x = LSTM(units=30, activation='relu', return_sequences=True)(x)
x = LSTM(units=30, activation='relu', return_sequences=True)(x)
x = TimeDistributed(Dense(1))(x)
x = Reshape(target_shape=(20, ))(x)
'''
x = Permute((2,1))(x)
x = GlobalAveragePooling1D()(x)
'''
self.model = Model(inputs=features_in, outputs=[end_node, x])
optimizer = Adam(learning_rate=hparams['learning_rate'], clipnorm=1)
def weighted_mse(y_true, y_pred):
loss_weights = np.sqrt(np.arange(1, 20))
#loss_weights = np.arange(1, 20)
return K.mean(K.square(y_pred - y_true) * loss_weights, axis=-1)
def haitao_error(y_true, y_pred):
diff = K.abs(
(y_true - y_pred) /
K.reshape(K.clip(K.abs(y_true[:, -1]), K.epsilon(), None),
(-1, 1)))
return 100. * K.mean(diff, axis=-1)
if hparams['loss'] == 'mape':
self.model.compile(optimizer=optimizer,
loss=MeanAbsolutePercentageError())
elif hparams['loss'] == 'haitao':
self.model.compile(optimizer=optimizer, loss=haitao_error)
elif hparams['loss'] == 'mse':
self.model.compile(optimizer=optimizer, loss='mean_squared_error')
#1.0 486
pd.value_counts(y_test)
#0.0 794
#1.0 256
############################################################################
###Modelling
os.chdir(model_path)
os.getcwd()
##CNN 1
model_1 = Sequential()
#Convolutional Layers
model_1.add(Reshape((55, 1), input_shape=(55, )))
model_1.add(
Conv1D(50, kernel_size=6, strides=1, padding="same", activation='relu'))
model_1.add(
Conv1D(40, kernel_size=11, strides=1, padding="same", activation='relu'))
model_1.add(
Conv1D(20, kernel_size=2, strides=2, padding="same", activation='relu'))
#Dense Layers
model_1.add(Flatten())
model_1.add(Dense(units=30, activation='relu'))
model_1.add(Dense(units=15, activation='relu'))
model_1.add(Dense(units=1, activation='sigmoid'))
model_1.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
#1.0 138
pd.value_counts(y_test)
#0.0 106
#1.0 72
############################################################################
###Modelling
os.chdir(model_path)
os.getcwd()
##CNN 1
model_1 = Sequential()
#Convolutional Layers
model_1.add(Reshape((97, 1), input_shape=(97, )))
model_1.add(
Conv1D(20, kernel_size=5, strides=5, padding="same", activation='relu'))
#Dense Layers
model_1.add(Flatten())
model_1.add(Dense(units=30, activation='relu'))
model_1.add(Dense(units=15, activation='relu'))
model_1.add(Dense(units=1, activation='sigmoid'))
model_1.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
# Fit the model_1
model_1.fit(X_train, y_train, epochs=20, batch_size=20)
#predictions
def model_self_learning(java_matrix, parola_segno):
import numpy as np
import pandas as pd
from csv import writer
from pandas import read_csv
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense
from tensorflow.keras.layers import Flatten
from tensorflow.keras.layers import Dropout
from tensorflow.python.keras.layers.convolutional import Conv1D
from tensorflow.python.keras.layers.convolutional import MaxPooling1D
from tensorflow.keras.models import Model
from tensorflow.python.keras.layers.advanced_activations import LeakyReLU
from tensorflow.keras.utils import to_categorical
from tensorflow.keras.models import model_from_json
from os.path import dirname, join
python_matrix = [[]]
row = 0
for x in java_matrix:
if (row == 0):
python_matrix = [x]
else:
python_matrix = python_matrix + [x]
row = row + 1
i = 0
dataset_path = join(dirname(__file__), 'Dataset_segni.csv')
with open(dataset_path, 'a+', newline='') as write_obj:
csv_writer = writer(write_obj)
for x in python_matrix:
python_matrix[i] = python_matrix[i] + [parola_segno]
csv_writer.writerow(python_matrix[i])
i = i + 1
write_obj.close()
df = pd.read_csv(dataset_path)
trad = df["Traduzione"]
trad.head()
gest = []
gest = gest + [trad[0]]
num = 0
numTrad = 0
check = False
for x in trad:
for i in gest:
if i == x:
check = True
if check == False:
if x != gest[num]:
num = num + 1
gest = gest + [trad[numTrad]]
numTrad = numTrad + 1
check = False
i = 0
for x in gest:
df.loc[df["Traduzione"] == gest[i], "label"] = i
i = i + 1
y = df.iloc[:, -1]
print(y)
x = df.iloc[:, :-2]
print(x)
y = to_categorical(y)
x = np.array(x[:])
print(x)
from sklearn.model_selection import train_test_split
x_train, x_test, y_train, y_test = train_test_split(x,
y,
test_size=0.3,
random_state=120)
x_train = x_train.reshape(x_train.shape[0], x_train.shape[1], 1)
x_test = x_test.reshape(x_test.shape[0], x_test.shape[1], 1)
batch_size = 64
num_classes = len(gest)
epochs = 10
input_shape = (x_train.shape[1], 1)
model = Sequential()
model.add(
Conv1D(filters=32,
kernel_size=3,
activation='linear',
input_shape=input_shape))
model.add(LeakyReLU(alpha=0.1))
model.add(MaxPooling1D(pool_size=(2), padding='same'))
model.add(
Conv1D(filters=64, kernel_size=3, activation='linear', padding='same'))
model.add(LeakyReLU(alpha=0.1))
model.add(MaxPooling1D(pool_size=(2), padding='same'))
model.add(
Conv1D(filters=128, kernel_size=3, activation='linear',
padding='same'))
model.add(LeakyReLU(alpha=0.1))
model.add(MaxPooling1D(pool_size=(2), padding='same'))
model.add(Flatten())
model.add(Dense(128, activation='linear'))
model.add(LeakyReLU(alpha=0.1))
model.add(Dense(num_classes, activation='softmax'))
model.summary()
model.compile(optimizer='adam',
loss='categorical_crossentropy',
metrics=['accuracy'])
# fit model
history = model.fit(x_train,
y_train,
batch_size=batch_size,
epochs=10,
verbose=1,
validation_data=(x_test, y_test))
model_path = join(dirname(__file__), 'model_cnn.h5')
model.save(model_path)
return python_matrix
def create_model(model_num, optimizer, dropout=False, metric=['accuracy']):
if model_num == 1:
model_1 = Sequential()
#Convolutional Layers
model_1.add(Reshape((62, 1), input_shape=(62, )))
model_1.add(
Conv1D(31,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
#Dense Layers
model_1.add(Flatten())
model_1.add(Dense(units=15, activation='relu'))
if dropout is not False:
model_1.add(Dropout(dropout))
model_1.add(Dense(units=1, activation='sigmoid'))
model_1.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=metric)
return model_1
elif model_num == 2:
model_2 = Sequential()
#Convolutional Layers
model_2.add(Reshape((62, 1), input_shape=(62, )))
model_2.add(
Conv1D(60,
kernel_size=3,
strides=1,
padding="same",
activation='relu'))
model_2.add(
Conv1D(30,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
#Dense Layers
model_2.add(Flatten())
model_2.add(Dense(units=15, activation='relu'))
if dropout is not False:
model_2.add(Dropout(dropout))
model_2.add(Dense(units=1, activation='sigmoid'))
model_2.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
return model_2
elif model_num == 3:
model_3 = Sequential()
#Convolutional Layers
model_3.add(Reshape((62, 1), input_shape=(62, )))
model_3.add(
Conv1D(61,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
model_3.add(
Conv1D(60,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
model_3.add(
Conv1D(30,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
#Dense Layers
model_3.add(Flatten())
model_3.add(Dense(units=15, activation='relu'))
if dropout is not False:
model_3.add(Dropout(dropout))
model_3.add(Dense(units=1, activation='sigmoid'))
model_3.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
return model_3
elif model_num == 4:
model_4 = Sequential()
#Convolutional Layers
model_4.add(Reshape((62, 1), input_shape=(62, )))
model_4.add(
Conv1D(60,
kernel_size=3,
strides=1,
padding="same",
activation='relu'))
model_4.add(
Conv1D(30,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
#Dense Layers
model_4.add(Flatten())
model_4.add(Dense(units=15, activation='relu'))
if dropout is not False:
model_4.add(Dropout(dropout))
model_4.add(Dense(units=10, activation='relu'))
if dropout is not False:
model_4.add(Dropout(dropout))
model_4.add(Dense(units=5, activation='relu'))
if dropout is not False:
model_4.add(Dropout(dropout))
model_4.add(Dense(units=1, activation='sigmoid'))
model_4.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
return model_4
elif model_num == 5:
model_5 = Sequential()
#Convolutional Layers
model_5.add(Reshape((62, 1), input_shape=(62, )))
model_5.add(
Conv1D(31,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
model_5.add(
Conv1D(30,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
#Dense Layers
model_5.add(Flatten())
model_5.add(Dense(units=15, activation='relu'))
if dropout is not False:
model_5.add(Dropout(dropout))
model_5.add(Dense(units=1, activation='sigmoid'))
model_5.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
return model_5
elif model_num == 6:
model_6 = Sequential()
#Convolutional Layers
model_6.add(Reshape((62, 1), input_shape=(62, )))
model_6.add(
Conv1D(61,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
model_6.add(
Conv1D(60,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
model_6.add(
Conv1D(30,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
model_6.add(
Conv1D(10,
kernel_size=3,
strides=3,
padding="same",
activation='relu'))
#Dense Layers
model_6.add(Flatten())
model_6.add(Dense(units=15, activation='relu'))
if dropout is not False:
model_6.add(Dropout(dropout))
model_6.add(Dense(units=1, activation='sigmoid'))
model_6.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
return model_6
elif model_num == 7:
model_7 = Sequential()
#Convolutional Layers
model_7.add(Reshape((62, 1), input_shape=(62, )))
model_7.add(
Conv1D(61,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
model_7.add(
Conv1D(60,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
model_7.add(
Conv1D(60,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
model_7.add(
Conv1D(20,
kernel_size=3,
strides=3,
padding="same",
activation='relu'))
#Dense Layers
model_7.add(Flatten())
model_7.add(Dense(units=30, activation='relu'))
if dropout is not False:
model_7.add(Dropout(dropout))
model_7.add(Dense(units=15, activation='relu'))
if dropout is not False:
model_7.add(Dropout(dropout))
model_7.add(Dense(units=1, activation='sigmoid'))
model_7.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
return model_7
elif model_num == 8:
model_8 = Sequential()
#Convolutional Layers
model_8.add(Reshape((62, 1), input_shape=(62, )))
model_8.add(
Conv1D(61,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
model_8.add(
Conv1D(60,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
model_8.add(
Conv1D(59,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
model_8.add(
Conv1D(40,
kernel_size=20,
strides=1,
padding="same",
activation='relu'))
model_8.add(
Conv1D(20,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
#Dense Layers
model_8.add(Flatten())
model_8.add(Dense(units=30, activation='relu'))
if dropout is not False:
model_8.add(Dropout(dropout))
model_8.add(Dense(units=1, activation='sigmoid'))
model_8.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
return model_8
elif model_num == 9:
model_9 = Sequential()
#Convolutional Layers
model_9.add(Reshape((62, 1), input_shape=(62, )))
model_9.add(
Conv1D(61,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
model_9.add(
Conv1D(60,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
model_9.add(
Conv1D(59,
kernel_size=2,
strides=1,
padding="same",
activation='relu'))
model_9.add(
Conv1D(40,
kernel_size=20,
strides=1,
padding="same",
activation='relu'))
model_9.add(
Conv1D(20,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
#Dense Layers
model_9.add(Flatten())
model_9.add(Dense(units=30, activation='relu'))
if dropout is not False:
model_9.add(Dropout(dropout))
model_9.add(Dense(units=1, activation='sigmoid'))
model_9.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
return model_9
elif model_num == 10:
model_10 = Sequential()
#Convolutional Layers
model_10.add(Reshape((62, 1), input_shape=(62, )))
model_10.add(
Conv1D(31,
kernel_size=2,
strides=2,
padding="same",
activation='relu'))
model_10.add(
Conv1D(20,
kernel_size=11,
strides=1,
padding="same",
activation='relu'))
#Dense Layers
model_10.add(Flatten())
model_10.add(Dense(units=30, activation='relu'))
if dropout is not False:
model_10.add(Dropout(dropout))
model_10.add(Dense(units=1, activation='sigmoid'))
model_10.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
return model_10
else:
print("Wrong model num : architecture not defined")
#0.0 627 #1.0 523 pd.value_counts(y_test) #0.0 360 #1.0 260 ############################################################################ ###Modelling os.chdir(model_path) os.getcwd() ##CNN 1 model_1 = Sequential() #Convolutional Layers model_1.add(Reshape((62, 1), input_shape=(62,))) model_1.add(Conv1D(31, kernel_size=2, strides=2, padding="same", activation = 'relu')) #Dense Layers model_1.add(Flatten()) model_1.add(Dense(units=15, activation='relu')) model_1.add(Dense(units=1, activation='sigmoid')) model_1.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy']) #Fit the model model_1.fit(X_train, y_train, epochs=20, batch_size=20) #Report Info model_name = "CNN" defined_params = "" file_name = "CNN_1"
def main():
print("Loading samples and labels")
samples, labels, _ = load_files("data")
print("Loaded {} samples".format(samples.shape[0]))
sequence_dim = 100
print("Converting to sequences of length {}".format(sequence_dim))
samples, labels = make_sequences(samples, labels, sequence_dim)
print("Number of samples from sequences: {}".format(samples.shape[0]))
lb = LabelBinarizer()
labels = lb.fit_transform(labels)
# flattened samples for Decision Tree
flatSamples = samples.reshape(samples.shape[0], -1) #tree!
(trainSamples, testSamples, trainLabels,
testLabels) = train_test_split(flatSamples,
labels,
test_size=0.25,
random_state=42)
print("=" * 20)
print("Building DecisionTree model")
model = DecisionTreeClassifier()
model.fit(trainSamples, trainLabels)
treeResults = model.predict(testSamples)
print(
confusion_matrix(testLabels.argmax(axis=1),
treeResults.argmax(axis=1)))
print(
classification_report(testLabels.argmax(axis=1),
treeResults.argmax(axis=1)))
treeAcc = accuracy_score(testLabels.argmax(axis=1),
treeResults.argmax(axis=1))
print("Accuracy Tree: {:.2f}".format(treeAcc))
print("Cohen's Kappa {:.2f}".format(
cohen_kappa_score(testLabels.argmax(axis=1),
treeResults.argmax(axis=1))))
print("=" * 20)
print("Building CNN model")
(trainSamples, testSamples, trainLabels,
testLabels) = train_test_split(samples,
labels,
test_size=0.25,
random_state=42)
inputShape = (samples.shape[1], samples.shape[2])
model = Sequential()
model.add(Conv1D(32, 10, padding="same", input_shape=inputShape))
model.add(Activation("relu"))
model.add(BatchNormalization())
model.add(Dropout(0.2))
model.add(Conv1D(64, 10, padding="same"))
model.add(Activation("relu"))
model.add(BatchNormalization())
model.add(Dropout(0.2))
model.add(Conv1D(128, 10, padding="same"))
model.add(Activation("relu"))
model.add(Dropout(0.2))
model.add(Flatten(input_shape=inputShape))
model.add(Dense(128, activation='sigmoid'))
model.add(Dense(64, activation='sigmoid'))
model.add(Dense(labels.shape[1], activation='softmax'))
model.compile(loss='categorical_crossentropy',
optimizer="adam",
metrics=['accuracy'])
EPOCHS = 10
BATCH = 128
model.fit(trainSamples,
trainLabels,
batch_size=BATCH,
epochs=EPOCHS,
validation_data=(testSamples, testLabels))
cnnResults = model.predict(testSamples)
print(
confusion_matrix(testLabels.argmax(axis=1), cnnResults.argmax(axis=1)))
print(
classification_report(testLabels.argmax(axis=1),
cnnResults.argmax(axis=1),
target_names=lb.classes_))
print("CNN Accuracy: {:.2f}".format(
accuracy_score(testLabels.argmax(axis=1), cnnResults.argmax(axis=1))))
print("Cohen's Kappa {:.2f}".format(
cohen_kappa_score(testLabels.argmax(axis=1),
cnnResults.argmax(axis=1))))
input("")
def evaluate_model(trainX, trainy, testX, testy, testy_norm):
"""
Create, fit and evaluate a model
:param trainX: (array)
:param trainy: (array)
:param testX: (array)
:param testy: (array)
:param testy_norm: (array)
:return:
accurancy (float)
loss (float)
"""
verbose, epochs, batch_size = 1, 60, 16 # 16
trainX, testX = scale_data(trainX, testX)
# trainX, testX = Magnitude(trainX,testX)
# trainX, testX = AutoCorallation(trainX, testX)
n_timesteps, n_features, n_outputs = trainX.shape[1], trainX.shape[
2], trainy.shape[1]
print(testX.shape)
print(testy.shape)
model = Sequential()
# Small structure
model.add(
Conv1D(32,
5,
activation='relu',
padding='same',
input_shape=(n_timesteps, n_features)))
model.add(MaxPooling1D(pool_size=2))
model.add(Conv1D(64, 5, activation='relu', padding='same'))
model.add(MaxPooling1D(pool_size=2))
model.add(Conv1D(128, 5, activation='relu', padding='same'))
model.add(SpatialDropout1D(0.5))
model.add(MaxPooling1D(pool_size=2))
model.add(Flatten())
model.add(Dense(128, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(64, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(2, activation='relu'))
model.add(Dense(n_outputs, activation='softmax'))
model.summary()
plot_model(model, 'model_info.png', show_shapes=True)
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
# fit network
tensorboard = TensorBoard(log_dir="logs_3xconv/{}".format(time()),
histogram_freq=1,
write_images=True)
history = model.fit(trainX,
trainy,
epochs=epochs,
batch_size=batch_size,
verbose=verbose,
validation_split=0.15,
shuffle=True,
callbacks=[tensorboard])
# evaluate model
loss, accuracy = model.evaluate(testX,
testy,
batch_size=batch_size,
verbose=0)
export_model(model)
predictions = model.predict_classes(testX)
print(metrics.classification_report(testy_norm, predictions))
confusion_matrix = metrics.confusion_matrix(y_true=testy_norm,
y_pred=predictions)
print(confusion_matrix)
normalised_confusion_matrix = np.array(
confusion_matrix, dtype=np.float32) / np.sum(confusion_matrix) * 100
print("")
print("Confusion matrix (normalised to % of total test data):")
print(normalised_confusion_matrix)
width = 12
height = 12
# fig, ax = plt.subplots()
plt.figure(figsize=(width, height))
plt.imshow(normalised_confusion_matrix,
interpolation='nearest',
cmap=plt.cm.rainbow)
plt.title("Confusion matrix \n(normalized to the entire test set [%])")
plt.colorbar()
tick_marks = np.arange(2)
LABELS = ["Dynamic", "Static"]
plt.xticks(tick_marks, LABELS, rotation=90)
plt.yticks(tick_marks, LABELS)
plt.tight_layout()
plt.ylabel('Real value')
plt.xlabel('Prediction value')
plt.figure()
plt.plot(history.history['loss'])
plt.plot(history.history['val_loss'])
plt.title('Model loss')
plt.ylabel('Loss')
plt.xlabel('Epoch')
plt.legend(['Training', 'Validation'], loc='upper left')
plt.figure()
plt.plot(history.history['accuracy'])
plt.plot(history.history['val_accuracy'])
plt.title('Model accurancy')
plt.ylabel('Accurancy')
plt.xlabel('Epoch')
plt.legend(['Training', 'Validation'], loc='upper left')
plt.show()
return accuracy, loss
#0.0 171
pd.value_counts(y_test)
#1.0 152
#0.0 88
############################################################################
###Modelling
os.chdir(model_path)
os.getcwd()
##CNN 1
model_1 = Sequential()
#Convolutional Layers
model_1.add(Reshape((62, 1), input_shape=(62, )))
model_1.add(
Conv1D(31, kernel_size=2, strides=2, padding="same", activation='relu'))
#Dense Layers
model_1.add(Flatten())
model_1.add(Dense(units=15, activation='relu'))
model_1.add(Dense(units=1, activation='sigmoid'))
model_1.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
#Fit the model
model_1.fit(X_train, y_train, epochs=20, batch_size=20)
#Report Info
model_name = "CNN"
from tensorflow.python.keras.layers import Flatten
from tensorflow.python.keras.layers.embeddings import Embedding
from tensorflow.python.keras.preprocessing import sequence
from tensorflow.python.keras.layers.convolutional import Conv1D
from tensorflow.python.keras.layers.convolutional import MaxPooling1D
top_words = 3000
(X_train, y_train), (X_test, y_test) = imdb.load_data(num_words=top_words)
max_words = 300
X_train = sequence.pad_sequences(X_train, maxlen=max_words)
X_test = sequence.pad_sequences(X_test, maxlen=max_words)
model = Sequential()
model.add(Embedding(top_words, 32, input_length=max_words))
model.add(Conv1D(filters=64, kernel_size=3, padding='same', activation='relu'))
model.add(Flatten())
model.add(Dense(250, activation='relu'))
model.add(Dense(250, activation='relu'))
model.add(Dropout(0.001, seed=0))
model.add(Dense(250, activation='relu'))
model.add(Dense(1, activation='sigmoid'))
model.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
model.fit(X_train,
y_train,
validation_data=(X_test, y_test),
epochs=5,
batch_size=128,