def encode(self, inputs, training=None, mask=None):
X, y, mask = prepare_ssl_inputs(inputs, mask=mask, n_unsupervised_inputs=1)
X = X[0] # only accept single inputs now
## encode normally
h_x = self.encoder(X, training=training, mask=mask)
h_x = bk.flatten(h_x, n_outdim=2)
## prepare the auxiliary
qa_x = self.qa_dist(self.encoder_a(X, training=training),
training=training,
mask=mask)
## prepare the label embedding
qy_ax = self.classify(X, training=training, qa_x=qa_x)
## combine into q(z|axy)
h_x = self.x_to_qz(h_x, training=training)
h_a = self.a_to_qz(qa_x, training=training)
h_y = self.y_to_qz(qy_ax, training=training)
h_axy = h_x + h_y + h_a
if self.batchnorm:
h_axy = self.qz_xy_norm(h_axy, training=training)
if 0.0 < self.dropout < 1.0:
h_axy = self.qz_xy_drop(h_axy, training=training)
# conditional embedding y
h_axy = self.xy_to_qz_net(h_axy, training=training, mask=mask)
qz_axy = self.latents(h_axy, training=training, mask=mask)
return (qz_axy, qa_x, qy_ax)
def encode(self,
inputs: Union[TensorTypes, List[TensorTypes]],
training: Optional[bool] = None,
mask: Optional[TensorTypes] = None,
**kwargs) -> JointDistributionSequential:
X, y, mask = prepare_ssl_inputs(
inputs, mask=mask, n_unsupervised_inputs=self.n_observation)
if len(y) == 0:
py = self.classify(X, training=training)
else: # only support single labels model
py = coercible_tensor(VectorDeterministic(loc=y[0]))
# encode normally
h_e = [
bk.flatten(fe(X[0], training=training, mask=mask), n_outdim=2)
for fe in self.encoder
]
if len(h_e) > 1:
h_e = tf.concat(h_e, axis=-1)
else:
h_e = h_e[0]
# conditional embedding y
y_embedded = self.labels_embedder[0](py)
h_e = tf.concat([h_e, y_embedded], axis=-1)
qz_x = [fz(h_e, training=training) for fz in self.latents]
qz_x.append(py)
return qz_x
def _apply(self, X):
ndims = X.get_shape().ndims
if ndims is not None:
if ndims == self.outdim:
return X
elif ndims < self.outdim:
raise RuntimeError("Input shape: %s, cannot be flatten to %d-D"
% (str(X.get_shape()), self.outdim))
return K.flatten(X, outdim=self.outdim)
def test_flatten(self):
x = K.placeholder(shape=(None, 8, 12, 25, 18))
for i in range(1, 5):
y = K.flatten(x, outdim=i)
f = K.function(x, y)
shape1 = K.get_shape(y)
shape2 = f(np.random.rand(16, 8, 12, 25, 18)).shape
self.assertEqual(len(shape1), len(shape2))
self.assertTrue(
all(i == j for i, j in zip(shape1, shape2) if i is not None))
def _apply(self, X, h0=None, c0=None, mask=None):
batch_size = K.get_shape(X, native=True)[0]
is_bidirectional = self.direction_mode == 'bidirectional'
input_mode = ('skip' if self.input_mode == 'skip'
or self.input_mode == 'norm' else 'linear')
# ====== precompute input ====== #
# linear or norm input mode
if self.input_mode == 'norm':
X = K.dot(X, self.W_in)
# normalize all axes except the time dimension
bn = BatchNorm(axes=(0, 1),
activation=K.linear,
gamma_init=self.gamma,
beta_init=self.beta,
mean_init=self.mean,
inv_std_init=self.inv_std)
X = bn(X)
# cudnnRNN doesnt' support multiple inputs
shapeX = K.get_shape(X, native=True)
ndims = K.ndim(X)
if 'rnn' in self.rnn_mode: N = 1
elif self.rnn_mode == 'gru': N = 3
else: N = 4
newshape = [shapeX[i]
for i in range(ndims - 1)] + [self.num_units, N]
X = K.mean(K.reshape(X, newshape), axis=-1)
# ====== hidden state ====== #
num_layers = self.num_layers * 2 if is_bidirectional else self.num_layers
require_shape = (num_layers, batch_size, self.num_units)
h0 = _check_cudnn_hidden_init(h0, require_shape, self, 'h0')
c0 = _check_cudnn_hidden_init(c0, require_shape, self, 'c0')
# ====== parameters ====== #
if self.params_split:
parameters = K.concatenate([
K.flatten(i, outdim=1) for i in self.parameters
if not has_roles(i, INITIAL_STATE)
])
else:
parameters = self.params
# ====== return CuDNN RNN ====== #
results = K.rnn_dnn(X,
hidden_size=self.num_units,
rnn_mode=self.rnn_mode,
num_layers=self.num_layers,
parameters=parameters,
h0=h0,
c0=c0,
input_mode=input_mode,
direction_mode=self.direction_mode,
dropout=self.dropout,
name=self.name)
if not self.return_states:
results = results[0] # only get the output
return results
def test_norm(self):
for p in [1, 2, 'fro', np.inf]:
for axis in [None, 0, 1, (0, 1)]:
a = bk.norm(bk.flatten(x, 2), p=p, axis=axis, keepdims=True)
b = bk.norm(bk.flatten(y, 2), p=p, axis=axis, keepdims=True)
c = bk.norm(bk.flatten(z, 2), p=p, axis=axis, keepdims=True)
assert_equal(self, (p, axis), a, b, c)
a = bk.norm(bk.flatten(x, 2), p=p, axis=axis, keepdims=False)
b = bk.norm(bk.flatten(y, 2), p=p, axis=axis, keepdims=False)
c = bk.norm(bk.flatten(z, 2), p=p, axis=axis, keepdims=False)
assert_equal(self, (p, axis), a, b, c)
def classify(self,
inputs,
training=False,
qa_x: Optional[Distribution] = None) -> Distribution:
"""Return the prediction of labels"""
# prepare x
if isinstance(inputs, (tuple, list)):
inputs = inputs[0] # only support a single inputs Tensor
h_x = self.x_to_qy(bk.flatten(inputs, n_outdim=2), training=training)
# prepare a
if qa_x is None:
qa_x = self.qa_dist(self.encoder_a(inputs, training=training),
training=training)
h_a = self.a_to_qy(qa_x, training=training)
# final combination
h_ax = h_a + h_x
if self.batchnorm:
h_ax = self.qy_ax_norm(h_ax, training=training)
if 0.0 < self.dropout < 1.0:
h_ax = self.qy_ax_drop(h_ax, training=training)
h = self.classifier(h_ax, training=training)
return self.labels(h, training=training)
name='Decoder')
# ===========================================================================
# Create statistical model
# ===========================================================================
# ====== encoder ====== #
E = f_encoder(X)
# ====== latent ====== #
q_Z_given_X = B.parse_distribution(args.zdist, E, int(args.zdim), name='Z')
# [n_sample, n_batch, zdim]
q_Z_given_X_samples = q_Z_given_X.sample(nsample)
Z = [
q_Z_given_X.mean(),
tf.concat([q_Z_given_X.mean(),
tf.sqrt(q_Z_given_X.variance())], axis=-1),
K.flatten(tf.transpose(q_Z_given_X_samples, perm=(1, 0, 2)), outdim=2)
]
Z_names = ["posterior mean", "statistic pooling", "all samples flatten"]
# ====== Z prior ====== #
p_Z = B.parse_distribution(dist_name=args.zprior)
# ====== decoder ====== #
D = f_decoder(q_Z_given_X_samples)
# ====== reconstruction ====== #
p_X_given_Z = B.parse_distribution(args.xdist,
D,
int(np.prod(input_shape[1:])),
n_eventdim=1,
name='W')
# [n_sample, n_batch, feat_dim]
p_X_given_Z_mean = p_X_given_Z.mean()
# [n_batch, feat_dim]
def flatten_and_test(n): a = bk.flatten(x, n) b = bk.flatten(y, n) c = bk.flatten(z, n) assert_equal(self, n, a, b, c)
def _apply(self, x):
input_shape = K.get_shape(x)
_validate_input_shape(input_shape)
return K.flatten(x, outdim=self.outdim)
def _apply(self, X):
return K.flatten(X, outdim=self.outdim)