Add ZeroSumNormal distribution (#6121)

Also:
* Refactor get_steps to work with multivariate support shapes
* Replace get_steps by get_support_shape_1d in timeseries.py

Co-authored-by: Ricardo Vieira <[email protected]>

Author: AlexAndorra

docs/source/api/distributions/multivariate.rst

@@ -8,6 +8,7 @@ Multivariate
 
    MvNormal
    MvStudentT
+   ZeroSumNormal
    Dirichlet
    Multinomial
    DirichletMultinomial

docs/source/api/distributions/transforms.rst

@@ -33,6 +33,7 @@ Specific Transform Classes
     LogExpM1
     Ordered
     SumTo1
+    ZeroSumTransform
 
 
 Transform Composition Classes

pymc/distributions/__init__.py

@@ -99,6 +99,7 @@
     StickBreakingWeights,
     Wishart,
     WishartBartlett,
+    ZeroSumNormal,
 )
 from pymc.distributions.simulator import Simulator
 from pymc.distributions.timeseries import (
@@ -116,8 +117,8 @@
     "Uniform",
     "Flat",
     "HalfFlat",
-    "TruncatedNormal",
     "Normal",
+    "TruncatedNormal",
     "Beta",
     "Kumaraswamy",
     "Exponential",
@@ -160,6 +161,7 @@
     "Continuous",
     "Discrete",
     "MvNormal",
+    "ZeroSumNormal",
     "MatrixNormal",
     "KroneckerNormal",
     "MvStudentT",

pymc/distributions/multivariate.py

@@ -18,6 +18,7 @@
 import warnings
 
 from functools import reduce
+from typing import Optional
 
 import aesara
 import aesara.tensor as at
@@ -63,15 +64,17 @@
     _change_dist_size,
     broadcast_dist_samples_to,
     change_dist_size,
+    get_support_shape,
     rv_size_is_none,
     to_tuple,
 )
-from pymc.distributions.transforms import Interval, _default_transform
+from pymc.distributions.transforms import Interval, ZeroSumTransform, _default_transform
 from pymc.math import kron_diag, kron_dot
 from pymc.util import check_dist_not_registered
 
 __all__ = [
     "MvNormal",
+    "ZeroSumNormal",
     "MvStudentT",
     "Dirichlet",
     "Multinomial",
@@ -2380,3 +2383,205 @@ def logp(value, alpha, K):
             K > 0,
             msg="alpha > 0, K > 0",
         )
+
+
+class ZeroSumNormalRV(SymbolicRandomVariable):
+    """ZeroSumNormal random variable"""
+
+    _print_name = ("ZeroSumNormal", "\\operatorname{ZeroSumNormal}")
+    default_output = 0
+
+
+class ZeroSumNormal(Distribution):
+    r"""
+    ZeroSumNormal distribution, i.e Normal distribution where one or
+    several axes are constrained to sum to zero.
+    By default, the last axis is constrained to sum to zero.
+    See `zerosum_axes` kwarg for more details.
+
+    .. math::
+
+        \begin{align*}
+            ZSN(\sigma) = N \Big( 0, \sigma^2 (I - \tfrac{1}{n}J) \Big) \\
+            \text{where} \ ~ J_{ij} = 1 \ ~ \text{and} \\
+            n = \text{nbr of zero-sum axes}
+        \end{align*}
+
+    Parameters
+    ----------
+    sigma : tensor_like of float
+        Scale parameter (sigma > 0).
+        It's actually the standard deviation of the underlying, unconstrained Normal distribution.
+        Defaults to 1 if not specified.
+        For now, ``sigma`` has to be a scalar, to ensure the zero-sum constraint.
+    zerosum_axes: int, defaults to 1
+        Number of axes along which the zero-sum constraint is enforced, starting from the rightmost position.
+        Defaults to 1, i.e the rightmost axis.
+    dims: sequence of strings, optional
+        Dimension names of the distribution. Works the same as for other PyMC distributions.
+        Necessary if ``shape`` is not passed.
+    shape: tuple of integers, optional
+        Shape of the distribution. Works the same as for other PyMC distributions.
+        Necessary if ``dims`` or ``observed`` is not passed.
+
+    Warnings
+    --------
+    ``sigma`` has to be a scalar, to ensure the zero-sum constraint.
+    The ability to specifiy a vector of ``sigma`` may be added in future versions.
+
+    ``zerosum_axes`` has to be > 0. If you want the behavior of ``zerosum_axes = 0``,
+    just use ``pm.Normal``.
+
+    Examples
+    --------
+    Define a `ZeroSumNormal` variable, with `sigma=1` and
+    `zerosum_axes=1`  by default::
+
+        COORDS = {
+            "regions": ["a", "b", "c"],
+            "answers": ["yes", "no", "whatever", "don't understand question"],
+        }
+        with pm.Model(coords=COORDS) as m:
+            # the zero sum axis will be 'answers'
+            v = pm.ZeroSumNormal("v", dims=("regions", "answers"))
+
+        with pm.Model(coords=COORDS) as m:
+            # the zero sum axes will be 'answers' and 'regions'
+            v = pm.ZeroSumNormal("v", dims=("regions", "answers"), zerosum_axes=2)
+
+        with pm.Model(coords=COORDS) as m:
+            # the zero sum axes will be the last two
+            v = pm.ZeroSumNormal("v", shape=(3, 4, 5), zerosum_axes=2)
+    """
+    rv_type = ZeroSumNormalRV
+
+    def __new__(cls, *args, zerosum_axes=None, support_shape=None, dims=None, **kwargs):
+        if dims is not None or kwargs.get("observed") is not None:
+            zerosum_axes = cls.check_zerosum_axes(zerosum_axes)
+
+            support_shape = get_support_shape(
+                support_shape=support_shape,
+                shape=None,  # Shape will be checked in `cls.dist`
+                dims=dims,
+                observed=kwargs.get("observed", None),
+                ndim_supp=zerosum_axes,
+            )
+
+        return super().__new__(
+            cls, *args, zerosum_axes=zerosum_axes, support_shape=support_shape, dims=dims, **kwargs
+        )
+
+    @classmethod
+    def dist(cls, sigma=1, zerosum_axes=None, support_shape=None, **kwargs):
+        zerosum_axes = cls.check_zerosum_axes(zerosum_axes)
+
+        sigma = at.as_tensor_variable(floatX(sigma))
+        if sigma.ndim > 0:
+            raise ValueError("sigma has to be a scalar")
+
+        support_shape = get_support_shape(
+            support_shape=support_shape,
+            shape=kwargs.get("shape"),
+            ndim_supp=zerosum_axes,
+        )
+
+        if support_shape is None:
+            if zerosum_axes > 0:
+                raise ValueError("You must specify dims, shape or support_shape parameter")
+            # TODO: edge-case doesn't work for now, because at.stack in get_support_shape fails
+            # else:
+            #     support_shape = () # because it's just a Normal in that case
+        support_shape = at.as_tensor_variable(intX(support_shape))
+
+        assert zerosum_axes == at.get_vector_length(
+            support_shape
+        ), "support_shape has to be as long as zerosum_axes"
+
+        return super().dist(
+            [sigma], zerosum_axes=zerosum_axes, support_shape=support_shape, **kwargs
+        )
+
+    @classmethod
+    def check_zerosum_axes(cls, zerosum_axes: Optional[int]) -> int:
+        if zerosum_axes is None:
+            zerosum_axes = 1
+        if not isinstance(zerosum_axes, int):
+            raise TypeError("zerosum_axes has to be an integer")
+        if not zerosum_axes > 0:
+            raise ValueError("zerosum_axes has to be > 0")
+        return zerosum_axes
+
+    @classmethod
+    def rv_op(cls, sigma, zerosum_axes, support_shape, size=None):
+
+        shape = to_tuple(size) + tuple(support_shape)
+        normal_dist = ignore_logprob(pm.Normal.dist(sigma=sigma, shape=shape))
+
+        if zerosum_axes > normal_dist.ndim:
+            raise ValueError("Shape of distribution is too small for the number of zerosum axes")
+
+        normal_dist_, sigma_, support_shape_ = (
+            normal_dist.type(),
+            sigma.type(),
+            support_shape.type(),
+        )
+
+        # Zerosum-normaling is achieved by substracting the mean along the given zerosum_axes
+        zerosum_rv_ = normal_dist_
+        for axis in range(zerosum_axes):
+            zerosum_rv_ -= zerosum_rv_.mean(axis=-axis - 1, keepdims=True)
+
+        return ZeroSumNormalRV(
+            inputs=[normal_dist_, sigma_, support_shape_],
+            outputs=[zerosum_rv_, support_shape_],
+            ndim_supp=zerosum_axes,
+        )(normal_dist, sigma, support_shape)
+
+
+@_change_dist_size.register(ZeroSumNormalRV)
+def change_zerosum_size(op, normal_dist, new_size, expand=False):
+
+    normal_dist, sigma, support_shape = normal_dist.owner.inputs
+
+    if expand:
+        original_shape = tuple(normal_dist.shape)
+        old_size = original_shape[: len(original_shape) - op.ndim_supp]
+        new_size = tuple(new_size) + old_size
+
+    return ZeroSumNormal.rv_op(
+        sigma=sigma, zerosum_axes=op.ndim_supp, support_shape=support_shape, size=new_size
+    )
+
+
+@_moment.register(ZeroSumNormalRV)
+def zerosumnormal_moment(op, rv, *rv_inputs):
+    return at.zeros_like(rv)
+
+
+@_default_transform.register(ZeroSumNormalRV)
+def zerosum_default_transform(op, rv):
+    zerosum_axes = tuple(np.arange(-op.ndim_supp, 0))
+    return ZeroSumTransform(zerosum_axes)
+
+
+@_logprob.register(ZeroSumNormalRV)
+def zerosumnormal_logp(op, values, normal_dist, sigma, support_shape, **kwargs):
+    (value,) = values
+    shape = value.shape
+    zerosum_axes = op.ndim_supp
+
+    _deg_free_support_shape = at.inc_subtensor(shape[-zerosum_axes:], -1)
+    _full_size = at.prod(shape)
+    _degrees_of_freedom = at.prod(_deg_free_support_shape)
+
+    zerosums = [
+        at.all(at.isclose(at.mean(value, axis=-axis - 1), 0, atol=1e-9))
+        for axis in range(zerosum_axes)
+    ]
+
+    out = at.sum(
+        pm.logp(normal_dist, value) * _degrees_of_freedom / _full_size,
+        axis=tuple(np.arange(-zerosum_axes, 0)),
+    )
+
+    return check_parameters(out, *zerosums, msg="at.mean(value, axis=zerosum_axes) == 0")