def test_disconnected_paths(self):
# Test that taking gradient going through a disconnected
# path rasises an exception
a = np.asarray(self.rng.randn(5, 5), dtype=config.floatX)
x = matrix("x")
# This MUST raise a DisconnectedInputError error.
# This also rasies an additional warning from gradients.py.
with pytest.raises(DisconnectedInputError):
grad(disconnected_grad(x).sum(), x)
# This MUST NOT raise a DisconnectedInputError error.
y = grad((x + disconnected_grad(x)).sum(), x)
a = matrix("a")
b = matrix("b")
y = a + disconnected_grad(b)
# This MUST raise a DisconnectedInputError error.
# This also rasies an additional warning from gradients.py.
with pytest.raises(DisconnectedInputError):
grad(y.sum(), b)
# This MUST NOT raise a DisconnectedInputError error.
grad(y.sum(), a)
def test_connection_pattern(self):
T = aesara.tensor
x = T.matrix("x")
y = gradient.disconnected_grad(x)
connection_pattern = y.owner.op.connection_pattern(y.owner)
assert connection_pattern == [[False]]
def test_op_removed(self):
x = matrix("x")
y = x * disconnected_grad(x)
f = aesara.function([x], y)
# need to refer to aesara.disconnected_grad here,
# aesara.disconnected_grad is a wrapper function!
assert disconnected_grad_ not in [node.op for node in f.maker.fgraph.toposort()]
def test_grad(self):
a = np.asarray(self.rng.randn(5, 5), dtype=config.floatX)
x = matrix("x")
expressions_gradients = [
(x * disconnected_grad(x), x),
(x * disconnected_grad(exp(x)), exp(x)),
(x**2 * disconnected_grad(x), 2 * x**2),
]
for expr, expr_grad in expressions_gradients:
g = grad(expr.sum(), x)
# gradient according to aesara
f = aesara.function([x], g, on_unused_input="ignore")
# desired gradient
f2 = aesara.function([x], expr_grad, on_unused_input="ignore")
assert np.allclose(f(a), f2(a))
def incsubtensor_logp(op, var, rvs_to_values, indexed_rv_var, rv_values,
*indices, **kwargs):
index = indices_from_subtensor(getattr(op, "idx_list", None), indices)
_, (new_rv_var, ) = clone(
tuple(v for v in graph_inputs((indexed_rv_var, ))
if not isinstance(v, Constant)),
(indexed_rv_var, ),
copy_inputs=False,
copy_orphans=False,
)
new_values = at.set_subtensor(
disconnected_grad(new_rv_var)[index], rv_values)
logp_var = logpt(indexed_rv_var, new_values, **kwargs)
return logp_var
def test_connection_pattern(self):
x = matrix("x")
y = disconnected_grad(x)
connection_pattern = y.owner.op.connection_pattern(y.owner)
assert connection_pattern == [[False]]
