def logpmf(self, x, n, p):
x = tf.cast(tf.squeeze(x), dtype=tf.float32)
n = tf.cast(tf.squeeze(n), dtype=tf.float32)
p = tf.cast(tf.squeeze(p), dtype=tf.float32)
return log_gamma(n + 1.0) - log_gamma(x + 1.0) - \
log_gamma(n - x + 1.0) + \
tf.mul(x, tf.log(p)) + tf.mul(n - x, tf.log(1.0-p))
def logpdf(self, x, df, loc=0, scale=1):
x = tf.cast(tf.squeeze(x), dtype=tf.float32)
df = tf.cast(tf.squeeze(df), dtype=tf.float32)
loc = tf.cast(tf.squeeze(loc), dtype=tf.float32)
scale = tf.cast(tf.squeeze(scale), dtype=tf.float32)
z = (x - loc) / scale
return log_gamma(0.5 * (df + 1.0)) - log_gamma(0.5 * df) - \
0.5 * (tf.log(np.pi) + tf.log(df)) - tf.log(scale) - \
0.5 * (df + 1.0) * tf.log(1.0 + (1.0/df) * tf.square(z))
def logpdf(self, x, df, loc=0, scale=1):
x = tf.cast(tf.squeeze(x), dtype=tf.float32)
df = tf.squeeze(df)
loc = tf.cast(tf.squeeze(loc), dtype=tf.float32)
scale = tf.cast(tf.squeeze(scale), dtype=tf.float32)
return 0.5 * log_gamma(df + 1.0) - \
log_gamma(0.5 * df) - \
0.5 * (np.log(np.pi) + tf.log(df)) + tf.log(scale) - \
0.5 * (df + 1.0) * \
tf.log(1.0 + (1.0/df) * tf.square((x-loc)/scale))
def logpdf(self, x, df, loc=0, scale=1):
x = tf.cast(tf.squeeze(x), dtype=tf.float32)
df = tf.squeeze(df)
loc = tf.cast(tf.squeeze(loc), dtype=tf.float32)
scale = tf.cast(tf.squeeze(scale), dtype=tf.float32)
return 0.5 * log_gamma(df + 1.0) - \
log_gamma(0.5 * df) - \
0.5 * (np.log(np.pi) + tf.log(df)) + tf.log(scale) - \
0.5 * (df + 1.0) * \
tf.log(1.0 + (1.0/df) * tf.square((x-loc)/scale))
def logpmf(self, x, n, p):
"""
Arguments
----------
x: np.array or tf.Tensor
vector of length K, where x[i] is the number of outcomes
in the ith bucket
n: int or tf.Tensor
number of outcomes equal to sum x[i]
p: np.array or tf.Tensor
vector of probabilities summing to 1
"""
x = tf.cast(tf.squeeze(x), dtype=tf.float32)
n = tf.cast(tf.squeeze(n), dtype=tf.float32)
p = tf.cast(tf.squeeze(p), dtype=tf.float32)
one = tf.constant(1.0, dtype=tf.float32)
return log_gamma(n + one) - tf.reduce_sum(log_gamma(x + one)) + tf.reduce_sum(tf.mul(x, tf.log(p)))
def logpmf(self, x, n, p):
"""
Arguments
----------
x: np.array or tf.Tensor
vector of length K, where x[i] is the number of outcomes
in the ith bucket
n: int or tf.Tensor
number of outcomes equal to sum x[i]
p: np.array or tf.Tensor
vector of probabilities summing to 1
"""
x = tf.cast(tf.squeeze(x), dtype=tf.float32)
n = tf.cast(tf.squeeze(n), dtype=tf.float32)
p = tf.cast(tf.squeeze(p), dtype=tf.float32)
one = tf.constant(1.0, dtype=tf.float32)
return log_gamma(n + one) - \
tf.reduce_sum(log_gamma(x + one)) + \
tf.reduce_sum(tf.mul(x, tf.log(p)))
def logpmf(self, x, n, p):
"""
Parameters
----------
x : np.array or tf.Tensor
vector of length K, where x[i] is the number of outcomes
in the ith bucket, or matrix with column length K
n : int or tf.Tensor
number of outcomes equal to sum x[i]
p : np.array or tf.Tensor
vector of probabilities summing to 1
"""
x = tf.cast(tf.squeeze(x), dtype=tf.float32)
n = tf.cast(tf.squeeze(n), dtype=tf.float32)
p = tf.cast(tf.squeeze(p), dtype=tf.float32)
if len(get_dims(x)) == 1:
return log_gamma(n + 1.0) - \
tf.reduce_sum(log_gamma(x + 1.0)) + \
tf.reduce_sum(tf.mul(x, tf.log(p)))
else:
return log_gamma(n + 1.0) - \
tf.reduce_sum(log_gamma(x + 1.0), 1) + \
tf.reduce_sum(tf.mul(x, tf.log(p)), 1)
def logpdf(self, x, df):
x = tf.cast(tf.squeeze(x), dtype=tf.float32)
df = tf.cast(tf.squeeze(df), dtype=tf.float32)
return tf.mul(0.5*df - 1, tf.log(x)) - 0.5*x - \
tf.mul(0.5*df, tf.log(2.0)) - log_gamma(0.5*df)
def logpmf(self, x, mu):
x = tf.cast(tf.squeeze(x), dtype=tf.float32)
mu = tf.cast(tf.squeeze(mu), dtype=tf.float32)
return x * tf.log(mu) - mu - log_gamma(x + 1.0)
def logpdf(self, x, alpha, scale=1):
x = tf.cast(tf.squeeze(x), dtype=tf.float32)
alpha = tf.cast(tf.squeeze(alpha), dtype=tf.float32)
scale = tf.cast(tf.squeeze(scale), dtype=tf.float32)
return tf.mul(alpha, tf.log(scale)) - log_gamma(alpha) + \
tf.mul(-alpha-1, tf.log(x)) - tf.truediv(scale, x)
def logpdf(self, x, a, scale=1):
x = tf.cast(tf.squeeze(x), dtype=tf.float32)
a = tf.cast(tf.squeeze(a), dtype=tf.float32)
scale = tf.cast(tf.squeeze(scale), dtype=tf.float32)
return (a - 1.0) * tf.log(x) - x/scale - a * tf.log(scale) - log_gamma(a)
def _test(x):
xtf = tf.constant(x)
val_true = special.gammaln(x)
_assert_eq(log_gamma(xtf), val_true)
def logpdf(self, x, a, scale=1):
x = tf.cast(tf.squeeze(x), dtype=tf.float32)
a = tf.cast(tf.squeeze(a), dtype=tf.float32)
scale = tf.cast(tf.squeeze(scale), dtype=tf.float32)
return (a -
1.0) * tf.log(x) - x / scale - a * tf.log(scale) - log_gamma(a)
def logpmf(self, x, mu):
x = tf.squeeze(x)
mu = tf.cast(tf.squeeze(mu), dtype=tf.float32)
return x * tf.log(mu) - mu - log_gamma(x + 1.0)
def logpdf(self, x, alpha, scale=1):
x = tf.cast(tf.squeeze(x), dtype=tf.float32)
alpha = tf.cast(tf.squeeze(alpha), dtype=tf.float32)
scale = tf.cast(tf.squeeze(scale), dtype=tf.float32)
return tf.mul(alpha, tf.log(scale)) - log_gamma(alpha) + \
tf.mul(-alpha-1, tf.log(x)) - tf.truediv(scale, x)
