def model_no_masking(discrete_time, init_alpha, max_beta):
model = Sequential()
model.add(TimeDistributed(Dense(2), input_shape=(n_timesteps, n_features)))
model.add(Lambda(wtte.output_lambda, arguments={"init_alpha": init_alpha,
"max_beta_value": max_beta}))
if discrete_time:
loss = wtte.loss(kind='discrete').loss_function
else:
loss = wtte.loss(kind='continuous').loss_function
model.compile(loss=loss, optimizer=RMSprop(lr=lr))
return model
def model_masking(discrete_time,init_alpha,max_beta):
model = Sequential()
model.add(Masking(mask_value=mask_value,input_shape=(n_timesteps, n_features)))
model.add(TimeDistributed(Dense(2)))
model.add(Lambda(wtte.output_lambda, arguments={"init_alpha":init_alpha,
"max_beta_value":max_beta}))
if discrete_time:
loss = wtte.loss(kind='discrete',reduce_loss=False).loss_function
else:
loss = wtte.loss(kind='continuous',reduce_loss=False).loss_function
model.compile(loss=loss, optimizer=RMSprop(lr=lr),sample_weight_mode='temporal')
return model
def model_no_masking(discrete_time,init_alpha,max_beta):
model = Sequential()
model.add(TimeDistributed(Dense(2), input_shape=(n_timesteps, n_features)))
model.add(Lambda(wtte.output_lambda, arguments={"init_alpha":init_alpha,
"max_beta_value":max_beta}))
if discrete_time:
loss = wtte.loss(kind='discrete').loss_function
else:
loss = wtte.loss(kind='continuous').loss_function
model.compile(loss=loss, optimizer=RMSprop(lr=lr))
return model
def model_masking(discrete_time, init_alpha, max_beta):
model = Sequential()
model.add(Masking(mask_value=mask_value,
input_shape=(n_timesteps, n_features)))
model.add(TimeDistributed(Dense(2)))
model.add(Lambda(wtte.output_lambda, arguments={"init_alpha": init_alpha,
"max_beta_value": max_beta}))
if discrete_time:
loss = wtte.loss(kind='discrete', reduce_loss=False).loss_function
else:
loss = wtte.loss(kind='continuous', reduce_loss=False).loss_function
model.compile(loss=loss, optimizer=RMSprop(
lr=lr), sample_weight_mode='temporal')
return model
model = Sequential()
model.add(
GRU(1,
input_shape=(xg_train.shape[1:]),
activation='tanh',
return_sequences=True))
model.add(Dense(2))
model.add(
Lambda(wtte.output_lambda,
arguments={
"init_alpha": init_alpha,
"max_beta_value": 4.0
}))
loss = wtte.loss(kind='discrete').loss_function
#model.load_weights('load_weight.hdf5')
model.compile(loss=loss, optimizer=adam(lr=.01))
# checkpoints
filepath = 'gen_cp/{epoch:02d}-{val_loss:.2f}.hdf5'
checkpoint = ModelCheckpoint(filepath,
monitor='val_loss',
verbose=1,
save_best_only=True,
save_weights_only=False,
mode='auto',
period=1)
## Base model
model = Sequential()
model.add(Masking(mask_value=mask_value,input_shape=(None, n_features)))
model.add(GRU(10,activation='tanh',return_sequences=True,recurrent_dropout=0.1,unroll=False))
model.add(BatchNormalization(axis=-1, momentum=0.9, epsilon=0.01))
model.add(TimeDistributed(Dense(10,activation='tanh')))
## Wtte-RNN part
model.add(TimeDistributed(Dense(2)))
model.add(Lambda(wtte.output_lambda, arguments={"init_alpha":init_alpha,
"max_beta_value":2.0,
"alpha_kernel_scalefactor":0.5
}))
loss = wtte.loss(kind='discrete',reduce_loss=False).loss_function
model.compile(loss=loss, optimizer=adam(lr=.01,clipvalue=0.5),sample_weight_mode='temporal')
model.summary()
K.set_value(model.optimizer.lr, 0.01)
model.fit(x_train, y_train,
epochs=300,
batch_size=300,
verbose=1,
validation_data=(x_valid, y_valid,sample_weights_valid),
sample_weight = sample_weights_train,
callbacks=[nanterminator,history,weightwatcher,reduce_lr])
## model 학습이 강제 종료됨 : loss nan
"""
Train on 924 samples, validate on 105 samples
def build_model(self, X_train, y_train, size_dyn, size_sta):
"""Build time to event model using a GRU network.
Args:
X_train (object): training set input features of shape
[n_examples, prefix, n_features]
y_train (object): training set labels of shape
[n_examples, n_features]
size_dyn (int): GRU units size.
size_sta (int): Static branch hidden layer size (optional)
Returns:
initialize self.model
"""
logger.info('Initializing time to event model ...')
# check if we have static features
static_flag = False
if X_train[1].shape[2] != 0:
static_flag = True
X_train_static = X_train[1]
n_features_static = X_train_static.shape[-1]
# Static model
static_input = Input(shape=(n_features_static),
name='static_input')
dense_static1 = Dense(size_sta,
name='hidden_static1')(static_input)
bs1 = BatchNormalization()(dense_static1)
# dense_static2 = Dense(size_sta//2, name='hidden_static2')(bs1)
# bs2 = BatchNormalization()(dense_static2)
static_output = Dense(1,
name='static_output',
activation='sigmoid')(bs1)
X_train = X_train[0]
tte_mean_train = np.nanmean(y_train[:, 0].astype('float'))
mean_u = np.nanmean(y_train[:, 1].astype('float'))
init_alpha = -1.0 / np.log(1.0 - 1.0 / (tte_mean_train + 1.0))
init_alpha = init_alpha / mean_u
n_features = X_train.shape[-1]
# Fixing seeds
np.random.seed(sd)
tf.random.set_seed(sd)
# Main model
main_input = Input(shape=(None, n_features), name='main_input')
l1 = GRU(size_dyn,
activation='tanh',
recurrent_dropout=0.25,
return_sequences=True)(main_input)
b1 = BatchNormalization()(l1)
l2 = GRU(size_dyn // 2,
activation='tanh',
recurrent_dropout=0.25,
return_sequences=False)(b1)
b2 = BatchNormalization()(l2)
if static_flag:
dynamic_output = Dense(2, name='Dense_main')(b2)
merged = Concatenate()([dynamic_output, static_output])
l4 = Dense(2, name='output')(merged)
else:
l4 = Dense(2, name='output')(b2)
output = Lambda(wtte.output_lambda,
name="lambda_layer",
arguments={
"init_alpha": init_alpha,
"max_beta_value": 100,
"scalefactor": 0.5
})(l4)
loss = wtte.loss(kind='continuous', reduce_loss=False).loss_function
np.random.seed(sd)
tf.random.set_seed(sd)
if static_flag:
self.model = Model(inputs=[main_input, static_input],
outputs=[output])
else:
self.model = Model(inputs=[main_input], outputs=[output])
self.model.compile(loss=loss,
optimizer=Nadam(lr=0.001,
beta_1=0.9,
beta_2=0.999,
epsilon=1e-08,
schedule_decay=0.004,
clipvalue=3))
# self.model.compile(loss=loss, optimizer=RMSprop(
# lr=0.001, clipvalue=3))
return
