def __init__(self,
do_convolution,
dim_latent,
num_examples,
dim_hidden,
dropout_rate=0.1,
beta=1.0,
negative_sampling=True,
*args,
**kwargs):
super(Encoder, self).__init__(*args, **kwargs)
self.conv = (convolution.ConvHead(
base_filters=32,
num_examples=sum(num_examples),
dropout_rate=dropout_rate,
) if do_convolution else dense.identity())
self.hidden_1 = dense.Dense(
units=dim_hidden,
num_examples=sum(num_examples),
dropout_rate=dropout_rate,
activation="elu",
name="encoder_hidden_1",
)
self.hidden_2 = dense.Dense(
units=dim_hidden,
num_examples=sum(num_examples),
dropout_rate=dropout_rate,
activation="elu",
name="encoder_hidden_2",
)
self.hidden_3 = dense.Dense(
units=dim_hidden,
num_examples=num_examples,
dropout_rate=dropout_rate,
num_branches=2,
activation="elu",
name="encoder_hidden_3",
)
self.hidden_4 = dense.Dense(
units=dim_hidden,
num_examples=num_examples,
dropout_rate=dropout_rate,
num_branches=2,
activation="elu",
name="encoder_hidden_4",
)
self.sampler = samplers.NormalSampler(
dim_output=dim_latent,
num_branches=2,
num_examples=num_examples,
beta=beta / 2 if negative_sampling else beta,
)
self.negative_sampling = negative_sampling
def __init__(self, dim_output, num_branches, num_examples, beta, *args,
**kwargs):
super(NormalSampler, self).__init__(*args, **kwargs)
self.mu = dense.Dense(units=dim_output,
num_examples=num_examples,
dropout_rate=0.0,
num_branches=num_branches,
activation='linear',
name='normal_mu')
self.sigma = dense.Dense(
units=dim_output,
num_examples=num_examples,
dropout_rate=0.0,
num_branches=num_branches,
activation='softplus',
kernel_initializer=tf.keras.initializers.RandomNormal(stddev=0.02),
name='normal_sigma')
self.beta = beta
def __init__(self, dim_output, num_branches, num_examples, beta, *args,
**kwargs):
super(CategoricalSampler, self).__init__(*args, **kwargs)
self.logits = dense.Dense(units=dim_output,
num_examples=num_examples,
dropout_rate=0.0,
num_branches=num_branches,
activation='linear',
name='categorical_logits')
self.beta = beta
def __init__(self, dim_output, num_branches, num_examples, beta, *args,
**kwargs):
super(BernoulliSampler, self).__init__(*args, **kwargs)
self.logits = dense.Dense(units=dim_output,
num_examples=num_examples,
dropout_rate=0.0,
num_branches=num_branches,
activation='linear',
name='encoder_mu')
self.beta = beta
def __init__(self,
do_convolution,
num_examples,
dim_hidden,
regression,
dropout_rate=0.1,
beta=1.0,
mode="tarnet",
*args,
**kwargs):
super(TARNet, self).__init__(*args, **kwargs)
self.conv = (convolution.ConvHead(
base_filters=32,
num_examples=sum(num_examples),
dropout_rate=0.1,
) if do_convolution else dense.identity())
self.hidden_1 = dense.Dense(
units=dim_hidden,
num_examples=sum(num_examples),
dropout_rate=dropout_rate,
activation="elu",
name="tarnet_hidden_1",
)
self.hidden_2 = dense.Dense(
units=dim_hidden,
num_examples=sum(num_examples),
dropout_rate=dropout_rate,
activation="elu",
name="tarnet_hidden_2",
)
self.hidden_3 = dense.Dense(
units=dim_hidden,
num_examples=sum(num_examples),
dropout_rate=dropout_rate,
activation="elu",
name="tarnet_hidden_3",
)
self.hidden_4 = dense.Dense(
units=dim_hidden,
num_examples=num_examples,
dropout_rate=dropout_rate,
num_branches=2,
activation="elu",
name="encoder_hidden_4",
)
self.hidden_5 = dense.Dense(
units=dim_hidden,
num_examples=num_examples,
dropout_rate=0.5,
num_branches=2,
activation="elu",
name="encoder_hidden_5",
)
y_sampler = samplers.NormalSampler if regression else samplers.BernoulliSampler
self.y_sampler = y_sampler(dim_output=1,
num_branches=2,
num_examples=num_examples,
beta=0.0)
self.regression = regression
self.beta = beta
if mode == "dragon":
self.t_sampler = samplers.CategoricalSampler(
2, num_branches=1, num_examples=sum(num_examples), beta=0.0)
self.mode = mode
self.y_loss = (tf.keras.losses.MeanSquaredError()
if regression else tf.keras.losses.BinaryCrossentropy())
def __init__(self,
dim_x,
dim_t,
dim_y,
regression,
num_examples,
dim_hidden,
dropout_rate=0.1,
*args,
**kwargs):
super(Decoder, self).__init__(*args, **kwargs)
do_convolution = isinstance(dim_x, (tuple, list))
self.x_hidden_1 = dense.Dense(
units=dim_hidden,
num_examples=sum(num_examples),
dropout_rate=dropout_rate,
num_branches=1,
activation="elu",
name="decoder_x_hidden_1",
)
self.x_hidden_2 = dense.Dense(
units=dim_hidden,
num_examples=sum(num_examples),
dropout_rate=dropout_rate if do_convolution else 0.1,
num_branches=1,
activation="elu",
name="decoder_x_hidden_2",
)
self.x_conv = (convolution.ConvTail(
base_filters=32,
num_examples=sum(num_examples),
dropout_rate=dropout_rate,
) if do_convolution else dense.identity())
self.t_hidden_1 = dense.Dense(
units=dim_hidden,
num_examples=sum(num_examples),
dropout_rate=dropout_rate,
num_branches=1,
activation="elu",
name="decoder_t_hidden_1",
)
self.t_hidden_2 = dense.Dense(
units=dim_hidden,
num_examples=sum(num_examples),
dropout_rate=0.5,
num_branches=1,
activation="elu",
name="decoder_t_hidden_2",
)
self.t_sampler = samplers.CategoricalSampler(
dim_output=dim_t,
num_branches=1,
num_examples=sum(num_examples),
beta=0.0)
self.y_hidden_1 = dense.Dense(
units=dim_hidden,
num_examples=sum(num_examples),
dropout_rate=dropout_rate,
activation="elu",
name="decoder_y_hidden_1",
)
self.y_hidden_2 = dense.Dense(
units=dim_hidden,
num_examples=sum(num_examples),
dropout_rate=dropout_rate,
activation="elu",
name="decoder_y_hidden_2",
)
self.y_hidden_3 = dense.Dense(
units=dim_hidden,
num_examples=num_examples,
dropout_rate=dropout_rate,
num_branches=2,
activation="elu",
name="decoder_y_hidden_3",
)
self.y_hidden_4 = dense.Dense(
units=dim_hidden,
num_examples=num_examples,
dropout_rate=0.5,
num_branches=2,
activation="elu",
name="decoder_y_hidden_4",
)
y_sampler = samplers.NormalSampler if regression else samplers.BernoulliSampler
self.y_sampler = y_sampler(dim_output=dim_y,
num_branches=2,
num_examples=num_examples,
beta=0.0)