def latent(data_shape):
model = Sequential()
model.add(Dense(mc._OUT_DIM, activation='relu', input_shape=(data_shape,),\
kernel_regularizer=regularizers.l2(mc._L2_REGULARIZE_RATE)))
model.add(Dense(mc._OUT_DIM, activation='relu', input_shape=(data_shape,),\
kernel_regularizer=regularizers.l2(mc._L2_REGULARIZE_RATE)))
return model
def build_policy_head(self, input_layer):
policy_head = layers.Conv2D(2, (1, 1),
data_format=self.data_format,
kernel_regularizer=regularizers.l2(
self.reg_alpha))(input_layer)
policy_head = layers.BatchNormalization(
axis=self.batch_norm_axis)(policy_head)
policy_head = layers.Activation("relu")(policy_head)
policy_head = layers.Flatten()(policy_head)
policy_head = layers.Dense(BOARD_SIZE * BOARD_SIZE + 1,
kernel_initializer='random_uniform',
bias_initializer='ones',
kernel_regularizer=regularizers.l2(
self.reg_alpha))(policy_head)
policy_head = layers.Activation("softmax",
name="output_policy")(policy_head)
return policy_head
def build_convolutional_block(self, input_layer):
conv_layer = layers.Conv2D(CONV_FILTERS,
CONV_KERNEL,
data_format=self.data_format,
padding='same',
kernel_regularizer=regularizers.l2(
self.reg_alpha))(input_layer)
conv_layer = layers.BatchNormalization(
axis=self.batch_norm_axis)(conv_layer)
conv_layer = layers.Activation("relu")(conv_layer)
return conv_layer
def build_residual_block(self, input_layer):
reslayer = layers.Conv2D(CONV_FILTERS,
CONV_KERNEL,
data_format=self.data_format,
padding='same',
kernel_regularizer=regularizers.l2(
self.reg_alpha))(input_layer)
reslayer = layers.BatchNormalization(
axis=self.batch_norm_axis)(reslayer)
reslayer = layers.Activation("relu")(reslayer)
reslayer = layers.Conv2D(CONV_FILTERS,
CONV_KERNEL,
data_format=self.data_format,
padding='same',
kernel_regularizer=regularizers.l2(
self.reg_alpha))(reslayer)
reslayer = layers.BatchNormalization(
axis=self.batch_norm_axis)(reslayer)
assert reslayer.shape.as_list() == input_layer.shape.as_list()
reslayer = layers.Add()([reslayer, input_layer])
reslayer = layers.Activation("relu")(reslayer)
return reslayer
def build_value_head(self, input_layer):
value_head = layers.Conv2D(1, (1, 1),
data_format=self.data_format,
kernel_regularizer=regularizers.l2(
self.reg_alpha))(input_layer)
value_head = layers.BatchNormalization(
axis=self.batch_norm_axis)(value_head)
value_head = layers.Activation("relu")(value_head)
value_head = layers.Flatten()(value_head)
value_head = layers.Dense(DENSE_SIZE,
"relu",
kernel_initializer='random_uniform',
bias_initializer='ones',
kernel_regularizer=regularizers.l2(
self.reg_alpha))(value_head)
value_head = layers.Dense(1,
"tanh",
name="output_value",
kernel_initializer='random_uniform',
bias_initializer='ones',
kernel_regularizer=regularizers.l2(
self.reg_alpha))(value_head)
return value_head
def __init__(self, learning_rate, layers, functions, optimizer_name,
beta=0.0, dropout=1.0):
self.n_input = layers[0]
self.n_hidden = layers[1:-1]
self.n_output = layers[-1]
self.model = Sequential()
if len(self.n_hidden) == 0:
# single layer
self.model.add(Dense(self.n_output, activation=functions[0],
kernel_regularizer=regularizers.l2(beta),
input_shape=(self.n_input,)))
elif len(self.n_hidden) == 1:
# hidden layer
self.model.add(Dense(self.n_hidden[0], activation=functions[0],
kernel_regularizer=regularizers.l2(beta),
input_shape=(self.n_input,)))
self.model.add(Dropout(dropout))
# output layer
self.model.add(Dense(self.n_output, activation=functions[1],
kernel_regularizer=regularizers.l2(beta)))
else:
# the first hidden layer
self.model.add(Dense(self.n_hidden[0], activation=functions[0],
kernel_regularizer=regularizers.l2(beta),
input_shape=(self.n_input,)))
self.model.add(Dropout(dropout))
# the second hidden layer
self.model.add(Dense(self.n_hidden[1], activation=functions[1],
kernel_regularizer=regularizers.l2(beta)))
self.model.add(Dropout(dropout))
# the output layer
self.model.add(Dense(self.n_output, activation=functions[2],
kernel_regularizer=regularizers.l2(beta)))
self.model.summary()
if optimizer_name == 'Adam': optimizer = Adam(learning_rate)
#self.model.compile(loss='mean_squared_error',
# optimizer=optimizer,
# metrics=['accuracy'])
self.model.compile(loss='categorical_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
def cross_val_model(params):
keras_model = None
# Destroys the current TF graph and creates a new one.
# Useful to avoid clutter from old models / layers.
K.clear_session()
self._set_xpu_config(num_cpus, num_gpus)
self.span = W = params.span
(X_miss, X_train), (X_miss_val, X_test) = self.train_test_split(
dataset,
train_size=train_size,
abnormal=abnormal,
)
if len(X_train) == 0:
raise errors.NoData("insufficient training data")
if len(X_test) == 0:
raise errors.NoData("insufficient validation data")
# expected input data shape: (batch_size, timesteps,)
# network parameters
input_shape = (W, )
intermediate_dim = params.intermediate_dim
latent_dim = params.latent_dim
# VAE model = encoder + decoder
# build encoder model
main_input = Input(shape=input_shape)
# bool vector to flag missing data points
aux_input = Input(shape=input_shape)
aux_output = Lambda(lambda x: x)(aux_input)
x = Dense(intermediate_dim,
kernel_regularizer=regularizers.l2(0.01),
activation='relu')(main_input)
z_mean = Dense(latent_dim, name='z_mean')(x)
z_log_var = Dense(latent_dim, name='z_log_var')(x)
# use reparameterization trick to push the sampling out as input
# note that "output_shape" isn't necessary with the TensorFlow backend
z = Lambda(sampling, output_shape=(latent_dim, ),
name='z')([z_mean, z_log_var])
# build decoder model
x = Dense(intermediate_dim,
kernel_regularizer=regularizers.l2(0.01),
activation='relu',
name='dense_1')(z)
main_output = Dense(W, activation='linear', name='dense_2')(x)
# instantiate Donut model
keras_model = _Model([main_input, aux_input],
[main_output, aux_output],
name='donut')
add_loss(keras_model, W)
optimizer_cls = None
if params.optimizer == 'adam':
optimizer_cls = tf.keras.optimizers.Adam()
keras_model.compile(optimizer=optimizer_cls, )
_stop = EarlyStopping(
monitor='val_loss',
patience=5,
verbose=_verbose,
mode='auto',
)
keras_model.fit_generator(
generator(X_train, X_miss, batch_size, keras_model),
epochs=num_epochs,
steps_per_epoch=len(X_train) / batch_size,
verbose=_verbose,
validation_data=([X_test, X_miss_val], None),
callbacks=[_stop],
workers=0, # https://github.com/keras-team/keras/issues/5511
)
# How well did it do?
score = keras_model.evaluate(
[X_test, X_miss_val],
batch_size=batch_size,
verbose=_verbose,
)
self.current_eval += 1
if progress_cb is not None:
progress_cb(self.current_eval, max_evals)
return score, keras_model
from __future__ import absolute_import, division, print_function
import numpy as np
import tensorflow.contrib.keras.api.keras.backend as K
from tensorflow.contrib.keras.api.keras.models import Model
from tensorflow.contrib.keras.api.keras.layers import Input
from tensorflow.contrib.keras.api.keras.layers import Conv2D, \
Dense, Activation, Lambda, Reshape, Permute, Add, Concatenate, \
BatchNormalization, Dropout
from tensorflow.contrib.keras.api.keras.optimizers import Adam
from tensorflow.contrib.keras.api.keras.regularizers import l2
from layers.edm import EDM
from layers.comb_matrix import CombMatrix
from utils.scoping import Scoping
CONV2D_ARGS = {'padding': 'same', 'data_format': 'channels_last', 'kernel_regularizer': l2(5e-4)}
class _DMNN(object):
def __init__(self, config):
self.name = config.model_type + '_' + config.model_version
self.data_set = config.data_set
self.batch_size = config.batch_size
self.num_actions = config.num_actions
self.seq_len = config.pick_num if config.pick_num > 0 else (
config.crop_len if config.crop_len > 0 else None)
self.njoints = config.njoints
self.body_members = config.body_members
self.dropout = config.dropout
real_seq = Input(
batch_shape=(self.batch_size, self.njoints, self.seq_len, 3),
