def trainable_gamma(shape, min_concentration=1e-3, min_scale=1e-5, name=None):
"""Learnable Gamma via concentration and scale parameterization."""
with tf.variable_scope(None, default_name="trainable_gamma"):
unconstrained_concentration = tf.get_variable(
"unconstrained_concentration", shape,
initializer=tf.random_normal_initializer(mean=0.5, stddev=0.1))
unconstrained_scale = tf.get_variable(
"unconstrained_scale", shape,
initializer=tf.random_normal_initializer(stddev=0.1))
concentration = tf.maximum(tf.nn.softplus(unconstrained_concentration),
min_concentration)
rate = tf.maximum(1. / tf.nn.softplus(unconstrained_scale), 1. / min_scale)
rv = ed.Gamma(concentration=concentration, rate=rate, name=name)
return rv
def deep_exponential_family(data_size, feature_size, units, shape):
"""A multi-layered topic model over a documents-by-terms matrix."""
w2 = ed.Gamma(0.1, 0.3, sample_shape=[units[2], units[1]], name="w2")
w1 = ed.Gamma(0.1, 0.3, sample_shape=[units[1], units[0]], name="w1")
w0 = ed.Gamma(0.1, 0.3, sample_shape=[units[0], feature_size], name="w0")
z2 = ed.Gamma(0.1, 0.1, sample_shape=[data_size, units[2]], name="z2")
z1 = ed.Gamma(shape, shape / tf.matmul(z2, w2), name="z1")
z0 = ed.Gamma(shape, shape / tf.matmul(z1, w1), name="z0")
x = ed.Poisson(tf.matmul(z0, w0), name="x")
return x
def deep_exponential_family(data_size, feature_size, units, shape):
# units表示每层维数的大小，从unit[2]开始到unit[0]，最后输出feature_size
# data_size为batch_size的意思
w2 = ed.Gamma(0.1, 0.3, sample_shape=[units[2], units[1]],
name="w2") # 前两个位置为参数
w1 = ed.Gamma(0.1, 0.3, sample_shape=[units[1], units[0]], name="w1")
w0 = ed.Gamma(0.1, 0.3, sample_shape=[units[0], feature_size], name="w0")
# z2相当于是原始分布，用z2结合参数传到生成x
z2 = ed.Gamma(0.1, 0.1, sample_shape=[data_size, units[2]], name="z2")
z1 = ed.Gamma(shape, shape / tf.matmul(z2, w2),
name="z1") # z1的形状跟着两个参数的形状跑，此处相当于对rate建模，concentration不建模
z0 = ed.Gamma(shape, shape / tf.matmul(z1, w1), name="z0")
x = ed.Poisson(tf.matmul(z0, w0), name="x")
return x, w2, w1, w0, z2, z1, z0
def trainable_gamma(shape, name=None):
with tf.variable_scope(None, default_name="trainable_gamma"):
return ed.Gamma(tf.nn.softplus(tf.get_variable("shape", shape)),
1.0 /
tf.nn.softplus(tf.get_variable("scale", shape)),
name=name)
def latent_gamma(shape, concentration, rate):
W = ed.Gamma(concentration=concentration, rate=rate, sample_shape=shape)
return W
