Add compile_forward_sampling_function

Author: lucianopaz

RELEASE-NOTES.md

@@ -139,6 +139,7 @@ This includes API changes we did not warn about since at least `3.11.0` (2021-01
 - `pymc.sampling_jax` samplers support `log_likelihood`, `observed_data`, and `sample_stats` in returned InferenceData object (see [#5189](https://github.com/pymc-devs/pymc/pull/5189))
 - Adding support for `pm.Deterministic` in `pymc.sampling_jax` (see [#5182](https://github.com/pymc-devs/pymc/pull/5182))
 - Added an alternative parametrization, `logit_p` to `pm.Binomial` and `pm.Categorical` distributions (see [5637](https://github.com/pymc-devs/pymc/pull/5637)).
+- Added the low level `compile_forward_sampling_function` method to compile the aesara function responsible for generating forward samples (see [#5759](https://github.com/pymc-devs/pymc/pull/5759)).
 - ...
 
 

pymc/sampling.py

@@ -43,16 +43,22 @@
 import numpy as np
 import xarray
 
-from aesara.graph.basic import Constant, Variable
-from aesara.tensor import TensorVariable
+from aesara import tensor as at
+from aesara.graph.basic import Apply, Constant, Variable, general_toposort, walk
+from aesara.graph.fg import FunctionGraph
+from aesara.tensor.random.op import RandomVariable
+from aesara.tensor.random.var import (
+    RandomGeneratorSharedVariable,
+    RandomStateSharedVariable,
+)
 from aesara.tensor.sharedvar import SharedVariable
 from arviz import InferenceData
 from fastprogress.fastprogress import progress_bar
 from typing_extensions import TypeAlias
 
 import pymc as pm
 
-from pymc.aesaraf import change_rv_size, compile_pymc, inputvars, walk_model
+from pymc.aesaraf import change_rv_size, compile_pymc
 from pymc.backends.arviz import _DefaultTrace
 from pymc.backends.base import BaseTrace, MultiTrace
 from pymc.backends.ndarray import NDArray
@@ -75,6 +81,7 @@
     get_default_varnames,
     get_untransformed_name,
     is_transformed_name,
+    point_wrapper,
 )
 from pymc.vartypes import discrete_types
 
@@ -83,6 +90,7 @@
 __all__ = [
     "sample",
     "iter_sample",
+    "compile_forward_sampling_function",
     "sample_posterior_predictive",
     "sample_posterior_predictive_w",
     "init_nuts",
@@ -1534,6 +1542,147 @@ def stop_tuning(step):
     return step
 
 
+def get_vars_in_point_list(trace, model):
+    """Get the list of Variable instances in the model that have values stored in the trace."""
+    if not isinstance(trace, MultiTrace):
+        names_in_trace = list(trace[0])
+    else:
+        names_in_trace = trace.varnames
+    vars_in_trace = [model[v] for v in names_in_trace]
+    return vars_in_trace
+
+
+def compile_forward_sampling_function(
+    outputs: List[Variable],
+    vars_in_trace: List[Variable],
+    basic_rvs: Optional[List[Variable]] = None,
+    givens_dict: Optional[Dict[Variable, Any]] = None,
+    **kwargs,
+) -> Callable[..., Union[np.ndarray, List[np.ndarray]]]:
+    """Compile a function to draw samples, conditioned on the values of some variables.
+
+    The goal of this function is to walk the aesara computational graph from the list
+    of output nodes down to the root nodes, and then compile a function that will produce
+    values for these output nodes. The compiled function will take as inputs the subset of
+    variables in the ``vars_in_trace`` that are deemed to not be **volatile**.
+
+    Volatile variables are variables whose values could change between runs of the
+    compiled function or after inference has been run. These variables are:
+
+    - Variables in the outputs list
+    - ``SharedVariable`` instances that are not ``RandomStateSharedVariable`` or ``RandomGeneratorSharedVariable``
+    - Basic RVs that are not in the ``vars_in_trace`` list
+    - Variables that are keys in the ``givens_dict``
+    - Variables that have volatile inputs
+
+    Where by basic RVs we mean ``Variable`` instances produced by a ``RandomVariable`` ``Op``
+    that are in the ``basic_rvs`` list.
+
+    Concretely, this function can be used to compile a function to sample from the
+    posterior predictive distribution of a model that has variables that are conditioned
+    on ``MutableData`` instances. The variables that depend on the mutable data will be
+    considered volatile, and as such, they wont be included as inputs into the compiled function.
+    This means that if they have values stored in the posterior, these values will be ignored
+    and new values will be computed (in the case of deterministics and potentials) or sampled
+    (in the case of random variables).
+
+    This function also enables a way to impute values for any variable in the computational
+    graph that produces the desired outputs: the ``givens_dict``. This dictionary can be used
+    to set the ``givens`` argument of the aesara function compilation. This will essentially
+    replace a node in the computational graph with any other expression that has the same
+    type as the desired node. Passing variables in the givens_dict is considered an intervention
+    that might lead to different variable values from those that could have been seen during
+    inference, as such, **any variable that is passed in the ``givens_dict`` will be considered
+    volatile**.
+
+    Parameters
+    ----------
+    outputs : List[aesara.graph.basic.Variable]
+        The list of variables that will be returned by the compiled function
+    vars_in_trace : List[aesara.graph.basic.Variable]
+        The list of variables that are assumed to have values stored in the trace
+    basic_rvs : Optional[List[aesara.graph.basic.Variable]]
+        A list of random variables that are defined in the model. This list (which could be the
+        output of ``model.basic_RVs``) should have a reference to the variables that should
+        be considered as random variable instances. This includes variables that have
+        a ``RandomVariable`` owner op, but also unpure random variables like Mixtures, or
+        Censored distributions. If ``None``, only pure random variables will be considered
+        as potential random variables.
+    givens_dict : Optional[Dict[aesara.graph.basic.Variable, Any]]
+        A dictionary that maps tensor variables to the values that should be used to replace them
+        in the compiled function. The types of the key and value should match or an error will be
+        raised during compilation.
+    """
+    if givens_dict is None:
+        givens_dict = {}
+
+    if basic_rvs is None:
+        basic_rvs = []
+
+    # We need a function graph to walk the clients and propagate the volatile property
+    fg = FunctionGraph(outputs=outputs, clone=False)
+
+    # Walk the graph from inputs to outputs and tag the volatile variables
+    nodes: List[Variable] = general_toposort(
+        fg.outputs, deps=lambda x: x.owner.inputs if x.owner else []
+    )
+    volatile_nodes: Set[Any] = set()
+    for node in nodes:
+        if (
+            node in fg.outputs
+            or node in givens_dict
+            or (  # SharedVariables, except RandomState/Generators
+                isinstance(node, SharedVariable)
+                and not isinstance(node, (RandomStateSharedVariable, RandomGeneratorSharedVariable))
+            )
+            or (  # Basic RVs that are not in the trace
+                node.owner
+                and isinstance(node.owner.op, RandomVariable)
+                and node in basic_rvs
+                and node not in vars_in_trace
+            )
+            or (  # Variables that have any volatile input
+                node.owner and any(inp in volatile_nodes for inp in node.owner.inputs)
+            )
+        ):
+            volatile_nodes.add(node)
+
+    # Collect the function inputs by walking the graph from the outputs. Inputs will be:
+    # 1. Random variables that are not volatile
+    # 2. Variables that have no owner and are not constant or shared
+    inputs = []
+
+    def expand(node):
+        if (
+            (
+                node.owner is None and not isinstance(node, (Constant, SharedVariable))
+            )  # Variables without owners that are not constant or shared
+            or node in vars_in_trace  # Variables in the trace
+        ) and node not in volatile_nodes:
+            # This test will include variables without owners, and that are not constant
+            # or shared, because these nodes will never be considered volatile
+            inputs.append(node)
+        if node.owner:
+            return node.owner.inputs
+
+    # walk produces a generator, so we have to actually exhaust the generator in a list to walk
+    # the entire graph
+    list(walk(fg.outputs, expand))
+
+    # Populate the givens list
+    givens = [
+        (
+            node,
+            value
+            if isinstance(value, (Variable, Apply))
+            else at.constant(value, dtype=getattr(node, "dtype", None), name=node.name),
+        )
+        for node, value in givens_dict.items()
+    ]
+
+    return compile_pymc(inputs, fg.outputs, givens=givens, on_unused_input="ignore", **kwargs)
+
+
 def sample_posterior_predictive(
     trace,
     samples: Optional[int] = None,
@@ -1718,38 +1867,23 @@ def sample_posterior_predictive(
             return trace
         return {}
 
-    inputs: Sequence[TensorVariable]
-    input_names: Sequence[str]
-    if not isinstance(_trace, MultiTrace):
-        names_in_trace = list(_trace[0])
-    else:
-        names_in_trace = _trace.varnames
-    inputs_and_names = [
-        (rv, rv.name)
-        for rv in walk_model(vars_to_sample, walk_past_rvs=True)
-        if rv not in vars_to_sample
-        and rv in model.named_vars.values()
-        and not isinstance(rv, (Constant, SharedVariable))
-        and rv.name in names_in_trace
-    ]
-    if inputs_and_names:
-        inputs, input_names = zip(*inputs_and_names)
-    else:
-        inputs, input_names = [], []
-
     if size is not None:
         vars_to_sample = [change_rv_size(v, size, expand=True) for v in vars_to_sample]
+    vars_in_trace = get_vars_in_point_list(_trace, model)
 
     if compile_kwargs is None:
         compile_kwargs = {}
-
-    sampler_fn = compile_pymc(
-        inputs,
-        vars_to_sample,
-        allow_input_downcast=True,
-        accept_inplace=True,
-        on_unused_input="ignore",
-        **compile_kwargs,
+    compile_kwargs.setdefault("allow_input_downcast", True)
+    compile_kwargs.setdefault("accept_inplace", True)
+
+    sampler_fn = point_wrapper(
+        compile_forward_sampling_function(
+            outputs=vars_to_sample,
+            vars_in_trace=vars_in_trace,
+            basic_rvs=model.basic_RVs,
+            givens_dict=None,
+            **compile_kwargs,
+        )
     )
 
     ppc_trace_t = _DefaultTrace(samples)
@@ -1775,7 +1909,7 @@ def sample_posterior_predictive(
             else:
                 param = _trace[idx % len_trace]
 
-            values = sampler_fn(*(param[n] for n in input_names))
+            values = sampler_fn(**param)
 
             for k, v in zip(vars_, values):
                 ppc_trace_t.insert(k.name, v, idx)
@@ -2063,16 +2197,16 @@ def sample_prior_predictive(
             names.append(rv_var.name)
             vars_to_sample.append(rv_var)
 
-    inputs = [i for i in inputvars(vars_to_sample) if not isinstance(i, (Constant, SharedVariable))]
-
     if compile_kwargs is None:
         compile_kwargs = {}
+    compile_kwargs.setdefault("allow_input_downcast", True)
+    compile_kwargs.setdefault("accept_inplace", True)
 
-    sampler_fn = compile_pymc(
-        inputs,
+    sampler_fn = compile_forward_sampling_function(
         vars_to_sample,
-        allow_input_downcast=True,
-        accept_inplace=True,
+        vars_in_trace=[],
+        basic_rvs=model.basic_RVs,
+        givens_dict=None,
         **compile_kwargs,
     )