Test values are gone

Author: ricardoV94

docs/source/contributing/developer_guide.md

@@ -387,10 +387,10 @@ with pm.Model() as m:
 print(m.basic_RVs)    # ==> [z, x, y]
 print(m.free_RVs)     # ==> [z, x]
 
-type(m.logpt)
+type(m.logp)
 # pytensor.tensor.var.TensorVariable
 
-m.logpt.eval({x: np.random.randn(*x.tag.test_value.shape) for x in m.free_RVs})
+m.logpt.eval(m.initial_point())
 # array(-51.25369126)
 ```
 
@@ -419,11 +419,9 @@ On a high level, it allows us to build conditional logp function and its gradien
 Here is a taste of how it works in action:
 
 ```python
-inputlist = [np.random.randn(*x.tag.test_value.shape) for x in m.free_RVs]
-
 func = m.logp_dlogp_function()
 func.set_extra_values({})
-input_dict = {x.name: y for x, y in zip(m.free_RVs, inputlist)}
+input_dict = m.initial_point()
 print(input_dict)
 input_array = func.dict_to_array(input_dict)
 print(input_array)

pymc/distributions/shape_utils.py

@@ -23,10 +23,9 @@
 
 import numpy as np
 
-from pytensor import config
 from pytensor import tensor as pt
 from pytensor.graph.basic import Constant, Variable
-from pytensor.graph.op import Op, compute_test_value
+from pytensor.graph.op import Op
 from pytensor.raise_op import Assert
 from pytensor.tensor.random.op import RandomVariable
 from pytensor.tensor.shape import SpecifyShape
@@ -288,9 +287,6 @@ def change_dist_size(
     for k, v in dist.tag.__dict__.items():
         new_dist.tag.__dict__.setdefault(k, v)
 
-    if config.compute_test_value != "off":
-        compute_test_value(new_dist)
-
     return new_dist
 
 

pymc/model/core.py

@@ -1429,17 +1429,11 @@ def create_value_var(
                 # Create value variable with the same type as the RV
                 value_var = rv_var.type()
                 value_var.name = rv_var.name
-                if pytensor.config.compute_test_value != "off":
-                    value_var.tag.test_value = rv_var.tag.test_value
             else:
                 # Create value variable with the same type as the transformed RV
                 value_var = transform.forward(rv_var, *rv_var.owner.inputs).type()
                 value_var.name = f"{rv_var.name}_{transform.name}__"
                 value_var.tag.transform = transform
-                if pytensor.config.compute_test_value != "off":
-                    value_var.tag.test_value = transform.forward(
-                        rv_var, *rv_var.owner.inputs
-                    ).tag.test_value
 
         self.rvs_to_transforms[rv_var] = transform
         self.rvs_to_values[rv_var] = value_var

pymc/ode/ode.py

@@ -20,10 +20,10 @@
 import scipy
 
 from pytensor.graph.basic import Apply
-from pytensor.graph.op import Op, get_test_value
+from pytensor.graph.op import Op
 from pytensor.tensor.type import TensorType
 
-from pymc.exceptions import DtypeError, ShapeError
+from pymc.exceptions import ShapeError
 from pymc.ode import utils
 
 _log = logging.getLogger(__name__)
@@ -172,46 +172,6 @@ def __call__(self, y0, theta, return_sens=False, **kwargs):
         # use default implementation to prepare symbolic outputs (via make_node)
         states, sens = super().__call__(y0, theta, **kwargs)
 
-        if pytensor.config.compute_test_value != "off":
-            # compute test values from input test values
-            test_states, test_sens = self._simulate(
-                y0=get_test_value(y0), theta=get_test_value(theta)
-            )
-
-            # check types of simulation result
-            if not test_states.dtype == self._otypes[0].dtype:
-                raise DtypeError(
-                    "Simulated states have the wrong type.",
-                    actual=test_states.dtype,
-                    expected=self._otypes[0].dtype,
-                )
-            if not test_sens.dtype == self._otypes[1].dtype:
-                raise DtypeError(
-                    "Simulated sensitivities have the wrong type.",
-                    actual=test_sens.dtype,
-                    expected=self._otypes[1].dtype,
-                )
-
-            # check shapes of simulation result
-            expected_states_shape = (self.n_times, self.n_states)
-            expected_sens_shape = (self.n_times, self.n_states, self.n_p)
-            if not test_states.shape == expected_states_shape:
-                raise ShapeError(
-                    "Simulated states have the wrong shape.",
-                    test_states.shape,
-                    expected_states_shape,
-                )
-            if not test_sens.shape == expected_sens_shape:
-                raise ShapeError(
-                    "Simulated sensitivities have the wrong shape.",
-                    test_sens.shape,
-                    expected_sens_shape,
-                )
-
-            # attach results as test values to the outputs
-            states.tag.test_value = test_states
-            sens.tag.test_value = test_sens
-
         if return_sens:
             return states, sens
         return states

pymc/ode/utils.py

@@ -84,20 +84,16 @@ def augment_system(ode_func, n_states, n_theta):
     """
     # Present state of the system
     t_y = pt.vector("y", dtype="float64", shape=(n_states,))
-    t_y.tag.test_value = np.ones((n_states,), dtype="float64")
     # Parameter(s).  Should be vector to allow for generaliztion to multiparameter
     # systems of ODEs.  Is m dimensional because it includes all initial conditions as well as ode parameters
     t_p = pt.vector("p", dtype="float64")
-    t_p.tag.test_value = np.ones((n_states + n_theta,), dtype="float64")
     # Time.  Allow for non-autonomous systems of ODEs to be analyzed
     t_t = pt.scalar("t", dtype="float64")
-    t_t.tag.test_value = 2.459
 
     # Present state of the gradients:
     # Will always be 0 unless the parameter is the initial condition
     # Entry i,j is partial of y[i] wrt to p[j]
     dydp_vec = pt.vector("dydp", dtype="float64")
-    dydp_vec.tag.test_value = make_sens_ic(n_states, n_theta, "float64")
 
     dydp = dydp_vec.reshape((n_states, n_states + n_theta))
 

pymc/pytensorf.py

@@ -364,7 +364,6 @@ def grad_ii(i, f, x):
     )
 
 
-@pytensor.config.change_flags(compute_test_value="ignore")
 def hessian(f, vars=None, negate_output=True):
     res = jacobian(gradient(f, vars), vars)
     if negate_output:
@@ -378,7 +377,6 @@ def hessian(f, vars=None, negate_output=True):
     return res
 
 
-@pytensor.config.change_flags(compute_test_value="ignore")
 def hessian_diag1(f, v):
     g = gradient1(f, v)
     idx = pt.arange(g.shape[0], dtype="int32")
@@ -389,7 +387,6 @@ def hess_ii(i):
     return pytensor.map(hess_ii, idx, return_updates=False)
 
 
-@pytensor.config.change_flags(compute_test_value="ignore")
 def hessian_diag(f, vars=None, negate_output=True):
     if vars is None:
         vars = cont_inputs(f)
@@ -580,9 +577,6 @@ def join_nonshared_inputs(
         joined_values = np.concatenate([point[var.name].ravel() for var in inputs])
         joined_inputs = pytensor.shared(joined_values, "joined_inputs")
 
-    if pytensor.config.compute_test_value != "off":
-        joined_inputs.tag.test_value = raveled_inputs.tag.test_value
-
     replace: dict[Variable, Variable] = {}
     last_idx = 0
     for var in inputs:

pymc/variational/approximations.py

@@ -69,7 +69,6 @@ def cov(self):
     def std(self):
         return rho2sigma(self.rho)
 
-    @pytensor.config.change_flags(compute_test_value="off")
     def __init_group__(self, group):
         super().__init_group__(group)
         if not self._check_user_params():
@@ -130,7 +129,6 @@ class FullRankGroup(Group):
     short_name = "full_rank"
     alias_names = frozenset(["fr"])
 
-    @pytensor.config.change_flags(compute_test_value="off")
     def __init_group__(self, group):
         super().__init_group__(group)
         if not self._check_user_params():
@@ -201,7 +199,6 @@ class EmpiricalGroup(Group):
     __param_spec__ = {"histogram": ("s", "d")}
     short_name = "empirical"
 
-    @pytensor.config.change_flags(compute_test_value="off")
     def __init_group__(self, group):
         super().__init_group__(group)
         self._check_trace()

pymc/variational/operators.py

@@ -84,7 +84,6 @@ def __init__(self, op: KSD, tf: opvi.TestFunction):
             raise opvi.ParametrizationError("Op should be KSD")
         super().__init__(op, tf)
 
-    @pytensor.config.change_flags(compute_test_value="off")
     def __call__(self, nmc, **kwargs) -> list[Variable]:
         op: KSD = self.op
         grad = op.apply(self.tf)

pymc/variational/opvi.py

@@ -149,35 +149,11 @@ def node_property(f):
 
         def wrapper(fn):
             ff = append_name(f)(fn)
-            f_ = pytensor.config.change_flags(compute_test_value="off")(ff)
-            return property(locally_cachedmethod(f_))
+            return property(locally_cachedmethod(ff))
 
         return wrapper
     else:
-        f_ = pytensor.config.change_flags(compute_test_value="off")(f)
-        return property(locally_cachedmethod(f_))
-
-
-@pytensor.config.change_flags(compute_test_value="ignore")
-def try_to_set_test_value(node_in, node_out, s):
-    _s = s
-    if s is None:
-        s = 1
-    s = pytensor.compile.view_op(pt.as_tensor(s))
-    if not isinstance(node_in, list | tuple):
-        node_in = [node_in]
-    if not isinstance(node_out, list | tuple):
-        node_out = [node_out]
-    for i, o in zip(node_in, node_out):
-        if hasattr(i.tag, "test_value"):
-            if not hasattr(s.tag, "test_value"):
-                continue
-            else:
-                tv = i.tag.test_value[None, ...]
-                tv = np.repeat(tv, s.tag.test_value, 0)
-                if _s is None:
-                    tv = tv[0]
-                o.tag.test_value = tv
+        return property(locally_cachedmethod(f))
 
 
 class ObjectiveUpdates(pytensor.OrderedUpdates):
@@ -320,7 +296,6 @@ def add_obj_updates(
         if self.op.returns_loss:
             updates.loss = obj_target
 
-    @pytensor.config.change_flags(compute_test_value="off")
     def step_function(
         self,
         obj_n_mc=None,
@@ -408,7 +383,6 @@ def step_function(
             step_fn = compile([], [], updates=updates, random_seed=seed, **compile_kwargs)
         return step_fn
 
-    @pytensor.config.change_flags(compute_test_value="off")
     def score_function(
         self, sc_n_mc=None, more_replacements=None, compile_kwargs=None, fn_kwargs=None
     ):  # pragma: no cover
@@ -449,7 +423,6 @@ def score_function(
         seed = self.approx.rng.randint(2**30, dtype=np.int64)
         return compile([], loss, random_seed=seed, **compile_kwargs)
 
-    @pytensor.config.change_flags(compute_test_value="off")
     def __call__(self, nmc, **kwargs):
         if "more_tf_params" in kwargs:
             m = -1.0
@@ -862,7 +835,6 @@ def _input_type(self, name):
         """
         return pt.vector(name)
 
-    @pytensor.config.change_flags(compute_test_value="off")
     def __init_group__(self, group):
         """Initialize the group."""
         if not group:
@@ -1016,7 +988,6 @@ def set_size_and_deterministic(
         self, node: list[Variable], s, d: bool, more_replacements: dict | None = None
     ) -> list[Variable]: ...
 
-    @pytensor.config.change_flags(compute_test_value="off")
     def set_size_and_deterministic(
         self, node: Variable | list[Variable], s, d: bool, more_replacements: dict | None = None
     ) -> Variable | list[Variable]:
@@ -1042,7 +1013,6 @@ def set_size_and_deterministic(
         assert not (
             set(makeiter(self.input)) & set(pytensor.graph.graph_inputs(makeiter(node_out)))
         )
-        try_to_set_test_value(node, node_out, s)
         assert self.symbolic_random not in set(pytensor.graph.graph_inputs(makeiter(node_out)))
         return node_out
 
@@ -1410,7 +1380,6 @@ def make_size_and_deterministic_replacements(self, s, d, more_replacements=None)
         flat2rand.update(more_replacements)
         return flat2rand
 
-    @pytensor.config.change_flags(compute_test_value="off")
     def set_size_and_deterministic(self, node, s, d, more_replacements=None):
         """*Dev* - after node is sampled via :func:`symbolic_sample_over_posterior` or :func:`symbolic_single_sample` new random generator can be allocated and applied to node.
 
@@ -1435,7 +1404,6 @@ def set_size_and_deterministic(self, node, s, d, more_replacements=None):
         node = graph_replace(node, optimizations, strict=False)
         node = graph_replace(node, flat2rand, strict=False)
         assert not (set(self.symbolic_randoms) & set(pytensor.graph.graph_inputs(makeiter(node))))
-        try_to_set_test_value(_node, node, s)
         return node
 
     def to_flat_input(self, node, more_replacements=None):
@@ -1487,7 +1455,6 @@ def get_optimization_replacements(self, s, d):
             repl[self.datalogp] = self.single_symbolic_datalogp
         return repl
 
-    @pytensor.config.change_flags(compute_test_value="off")
     def sample_node(self, node, size=None, deterministic=False, more_replacements=None):
         """Sample given node or nodes over shared posterior.
 
@@ -1519,7 +1486,6 @@ def sample_node(self, node, size=None, deterministic=False, more_replacements=No
         else:
             node_out = self.symbolic_sample_over_posterior(node)
         node_out = self.set_size_and_deterministic(node_out, size, deterministic)
-        try_to_set_test_value(node_in, node_out, size)
         return node_out
 
     def rslice(self, name):

tests/distributions/test_discrete.py

@@ -420,14 +420,12 @@ def test_categorical_logp_batch_dims(self, method):
         expected_p = [0.2, 0.5] if method is logp else [1.0, 1.0]
         np.testing.assert_allclose(expr.exp().eval(), expected_p)
 
-    @pytensor.config.change_flags(compute_test_value="raise")
     def test_categorical_bounds(self):
         with pm.Model():
             x = pm.Categorical("x", p=np.array([0.2, 0.3, 0.5]))
             assert np.isinf(logp(x, -1).eval())
             assert np.isinf(logp(x, 3).eval())
 
-    @pytensor.config.change_flags(compute_test_value="raise")
     @pytest.mark.parametrize(
         "p",
         [