Merge pull request #184 from danielward27/beta_and_sigmoid

Add Beta distribution and Sigmoid transform

flowjax/bijections/__init__.py

@@ -12,6 +12,7 @@
 from .planar import Planar
 from .power import Power
 from .rational_quadratic_spline import RationalQuadraticSpline
+from .sigmoid import Sigmoid
 from .softplus import SoftPlus
 from .tanh import LeakyTanh, Tanh
 from .utils import EmbedCondition, Flip, Identity, Invert, Partial, Permute, Reshape
@@ -40,6 +41,7 @@
     "Reshape",
     "Scale",
     "Scan",
+    "Sigmoid",
     "SoftPlus",
     "Stack",
     "Tanh",

flowjax/bijections/sigmoid.py

@@ -0,0 +1,36 @@
+"""Sigmoid bijection."""
+
+from typing import ClassVar
+
+import jax.numpy as jnp
+from jax import nn
+from jax.scipy.special import logit
+
+from flowjax.bijections.bijection import AbstractBijection
+
+
+class Sigmoid(AbstractBijection):
+    r"""Sigmoid bijection :math:`y = \sigma(x) = \frac{1}{1 + \exp(-x)}`.
+
+    Args:
+        shape: The shape of the transform.
+    """
+
+    shape: tuple[int, ...] = ()
+    cond_shape: ClassVar[None] = None
+
+    def transform(self, x, condition=None):
+        return nn.sigmoid(x)
+
+    def transform_and_log_det(self, x, condition=None):
+        y = nn.sigmoid(x)
+        log_det = jnp.sum(nn.log_sigmoid(x) + nn.log_sigmoid(-x))
+        return y, log_det
+
+    def inverse(self, y, condition=None):
+        return logit(y)
+
+    def inverse_and_log_det(self, y, condition=None):
+        x = logit(y)
+        log_det = -jnp.sum(nn.log_sigmoid(x) + nn.log_sigmoid(-x))
+        return x, log_det

flowjax/distributions.py

@@ -780,3 +780,29 @@ def __init__(self, concentration: ArrayLike, scale: ArrayLike):
         concentration, scale = jnp.broadcast_arrays(concentration, scale)
         self.base_dist = _StandardGamma(concentration)
         self.bijection = Scale(scale)
+
+
+class Beta(AbstractDistribution):
+    """Beta distribution.
+
+    Args:
+        alpha: The alpha shape parameter.
+        beta: The beta shape parameter.
+    """
+
+    alpha: Array | AbstractUnwrappable[Array]
+    beta: Array | AbstractUnwrappable[Array]
+    shape: tuple[int, ...]
+    cond_shape: ClassVar[None] = None
+
+    def __init__(self, alpha: ArrayLike, beta: ArrayLike):
+        alpha, beta = jnp.broadcast_arrays(alpha, beta)
+        self.alpha = Parameterize(softplus, inv_softplus(alpha))
+        self.beta = Parameterize(softplus, inv_softplus(beta))
+        self.shape = alpha.shape
+
+    def _sample(self, key, condition=None):
+        return jr.beta(key, self.alpha, self.beta)
+
+    def _log_prob(self, x, condition=None):
+        return jstats.beta.logpdf(x, self.alpha, self.beta).sum()

tests/test_bijections/test_bijections.py

@@ -30,6 +30,7 @@
     Reshape,
     Scale,
     Scan,
+    Sigmoid,
     SoftPlus,
     Stack,
     Tanh,
@@ -67,6 +68,7 @@
     "LeakyTanh (broadcast max_val)": lambda: LeakyTanh(1, (2, 3)),
     "Loc": lambda: Loc(jnp.arange(DIM)),
     "Exp": lambda: Exp((DIM,)),
+    "Sigmoid": lambda: Sigmoid((DIM,)),
     "SoftPlus": lambda: SoftPlus((DIM,)),
     "TriangularAffine (lower)": lambda: TriangularAffine(
         jnp.arange(DIM),

tests/test_distributions.py

@@ -13,6 +13,7 @@
 from flowjax.distributions import (
     AbstractDistribution,
     AbstractTransformed,
+    Beta,
     Cauchy,
     Exponential,
     Gamma,
@@ -62,6 +63,7 @@
         eqx.filter_vmap(Normal)(jnp.arange(3 * prod(shape)).reshape(3, *shape)),
         weights=jnp.arange(3) + 1,
     ),
+    "Beta": lambda shape: Beta(jnp.ones(shape), jnp.ones(shape)),
 }