def test_ifelse_lazy_c():
a = scalar()
b = generic()
c = generic()
notimpl = NotImplementedOp()
cloops = [True, False]
if aesara.config.cxx == "":
cloops = [False]
for use_cloop in cloops:
for lazy in [True, None]:
linker = aesara.link.vm.VMLinker(use_cloop=use_cloop, lazy=lazy)
f = function(
[a, b, c],
ifelse(a, notimpl(b), c),
mode=Mode(linker=linker, optimizer="fast_run"),
)
with pytest.raises(NotImplementedOpException):
f(1, "a", "b")
assert f(0, "a", "b") == "b"
def test_grad_lazy_if(self):
# Tests that we can compute the gradients through lazy if
x = vector("x", dtype=self.dtype)
y = vector("y", dtype=self.dtype)
c = iscalar("c")
z = ifelse(c, x, y)
gx, gy = aesara.grad(z.sum(), [x, y])
f = function(
[c, x, y], [self.cast_output(gx), self.cast_output(gy)], mode=self.mode
)
# There is only 2 of the 3 ifelse that are moved on the GPU.
# The one that stay on the CPU is for the shape.
self.assertFunctionContains(f, self.get_ifelse(1), min=2, max=3)
rng = np.random.RandomState(utt.fetch_seed())
xlen = rng.randint(200)
ylen = rng.randint(200)
vx = np.asarray(rng.uniform(size=(xlen,)), self.dtype)
vy = np.asarray(rng.uniform(size=(ylen,)), self.dtype)
gx0, gy0 = f(1, vx, vy)
assert np.allclose(gx0.shape, vx.shape)
assert np.allclose(gy0.shape, vy.shape)
assert np.all(np.asarray(gx0) == 1.0)
assert np.all(np.asarray(gy0) == 0.0)
gx0, gy0 = f(0, vx, vy)
assert np.allclose(gx0.shape, vx.shape)
assert np.allclose(gy0.shape, vy.shape)
assert np.all(np.asarray(gx0) == 0.0)
assert np.all(np.asarray(gy0) == 1.0)
def test_grad_test_values(self):
# Regression test for test values of `ifelse` gradient.
with aesara.config.change_flags(compute_test_value="raise"):
x = scalar("x")
x.tag.test_value = 1
# Used to crash due to undefined test value.
aesara.grad(ifelse(0, x, x), x)
def test_multiple_out_grad(self):
# Tests that we can compute the gradients through lazy if
x1 = vector("x1")
x2 = vector("x2")
y1 = vector("y1")
y2 = vector("y2")
c = iscalar("c")
z = ifelse(c, (x1, x2), (y1, y2))
grads = aesara.grad(z[0].sum() + z[1].sum(), [x1, x2, y1, y2])
f = function([c, x1, x2, y1, y2], grads)
rng = np.random.RandomState(utt.fetch_seed())
lens = [rng.randint(200) for i in range(4)]
values = [
np.asarray(rng.uniform(size=(l,)), aesara.config.floatX) for l in lens
]
outs_1 = f(1, *values)
assert all([x.shape[0] == y for x, y in zip(outs_1, lens)])
assert np.all(outs_1[0] == 1.0)
assert np.all(outs_1[1] == 1.0)
assert np.all(outs_1[2] == 0.0)
assert np.all(outs_1[3] == 0.0)
outs_0 = f(0, *values)
assert all([x.shape[0] == y for x, y in zip(outs_1, lens)])
assert np.all(outs_0[0] == 0.0)
assert np.all(outs_0[1] == 0.0)
assert np.all(outs_0[2] == 1.0)
assert np.all(outs_0[3] == 1.0)
def test_multiple_out(self):
x1 = vector("x1", dtype=self.dtype)
x2 = vector("x2", dtype=self.dtype)
y1 = vector("y1", dtype=self.dtype)
y2 = vector("y2", dtype=self.dtype)
c = iscalar("c")
z = ifelse(c, (x1, x2), (y1, y2))
f = function([c, x1, x2, y1, y2], z, mode=self.mode)
self.assertFunctionContains1(f, self.get_ifelse(2))
ifnode = [x for x in f.maker.fgraph.toposort() if isinstance(x.op, IfElse)][0]
assert len(ifnode.outputs) == 2
rng = np.random.RandomState(utt.fetch_seed())
x1len = rng.randint(200)
x2len = rng.randint(200)
y1len = rng.randint(200)
y2len = rng.randint(200)
vx1 = np.asarray(rng.uniform(size=(x1len,)), self.dtype)
vx2 = np.asarray(rng.uniform(size=(x2len,)), self.dtype)
vy1 = np.asarray(rng.uniform(size=(y1len,)), self.dtype)
vy2 = np.asarray(rng.uniform(size=(y2len,)), self.dtype)
ovx1, ovx2 = f(1, vx1, vx2, vy1, vy2)
ovy1, ovy2 = f(0, vx1, vx2, vy1, vy2)
assert np.allclose(vx1, ovx1)
assert np.allclose(vy1, ovy1)
assert np.allclose(vx2, ovx2)
assert np.allclose(vy2, ovy2)
def test_mixed_dtype(self):
x1 = vector("x1", dtype="int32")
x2 = vector("x2", dtype=self.dtype)
y1 = vector("y1", dtype="int32")
y2 = vector("y2", dtype=self.dtype)
c = iscalar("c")
f = function([c, x1, x2, y1, y2], ifelse(c, (x1, x2), (y1, y2)), mode=self.mode)
self.assertFunctionContains1(f, self.get_ifelse(2))
rng = np.random.RandomState(utt.fetch_seed())
xlen = rng.randint(200)
ylen = rng.randint(200)
vx1 = np.asarray(rng.uniform(size=(xlen,)) * 3, "int32")
vx2 = np.asarray(rng.uniform(size=(xlen,)), self.dtype)
vy1 = np.asarray(rng.uniform(size=(ylen,)) * 3, "int32")
vy2 = np.asarray(rng.uniform(size=(ylen,)), self.dtype)
o1, o2 = f(1, vx1, vx2, vy1, vy2)
assert np.allclose(vx1, o1)
assert np.allclose(vx2, o2)
o1, o2 = f(0, vx1, vx2, vy1, vy2)
assert np.allclose(vy1, o1)
assert np.allclose(vy2, o2)
def test_grad_cast_input(self):
x = vector("x", dtype=self.dtype)
y = vector("y", dtype=self.dtype)
c = iscalar("c")
z = ifelse(c, self.cast_output(x), self.cast_output(y))
gx, gy = aesara.grad(z.sum(), [x, y])
function([c, x, y], [gx, gy], mode=self.mode)
def test_str(self):
x = vector("x", dtype=self.dtype)
y = vector("y", dtype=self.dtype)
c = iscalar("c")
res = ifelse(c, x, y)
assert str(res.owner).startswith("if{}")
res.owner.op.name = "name"
res.owner.op.as_view = True
assert str(res.owner).startswith("if{name,inplace}")
def test_remove_useless_inputs1(self):
x = vector("x")
y = vector("y")
c = iscalar("c")
z = ifelse(c, (x, x), (y, y))
f = function([c, x, y], z)
ifnode = [n for n in f.maker.fgraph.toposort() if isinstance(n.op, IfElse)][0]
assert len(ifnode.inputs) == 3
def test_grad_cast_input(self):
# Tests the gradient when both inputs are on the GPU.
x = vector("x", dtype=self.dtype)
y = vector("y", dtype=self.dtype)
c = iscalar("c")
z = ifelse(c, self.cast_output(x), self.cast_output(y))
gx, gy = aesara.grad(z.sum(), [x, y])
function([c, x, y], [gx, gy], mode=self.mode)
def test_lazy_if_on_generics(self):
x = generic()
y = generic()
c = iscalar("c")
f = function([c, x, y], ifelse(c, x, y))
vx = ["testX"]
vy = ["testY"]
assert f(1, vx, vy) == vx
assert f(0, vx, vy) == vy
def test_remove_useless_inputs2(self):
x1 = vector("x1")
x2 = vector("x2")
y1 = vector("y1")
y2 = vector("y2")
c = iscalar("c")
z = ifelse(c, (x1, x1, x1, x2, x2), (y1, y1, y2, y2, y2))
f = function([c, x1, x2, y1, y2], z)
ifnode = [x for x in f.maker.fgraph.toposort() if isinstance(x.op, IfElse)][0]
assert len(ifnode.outputs) == 3
def test_pushout3(self):
x1 = scalar("x1")
y1 = scalar("x2")
y2 = scalar("y2")
c = iscalar("c")
two = np.asarray(2, dtype=aesara.config.floatX)
x, y = ifelse(c, (x1, y1), (two, y2), name="f1")
o3 = np.asarray(0.3, dtype=aesara.config.floatX)
o2 = np.asarray(0.2, dtype=aesara.config.floatX)
z = ifelse(c, o3, o2, name="f2")
out = x * z * y
f = function([x1, y1, y2, c], out, allow_input_downcast=True)
assert isinstance(f.maker.fgraph.toposort()[-1].op, IfElse)
rng = np.random.RandomState(utt.fetch_seed())
vx1 = rng.uniform()
vy1 = rng.uniform()
vy2 = rng.uniform()
assert np.allclose(f(vx1, vy1, vy2, 1), vx1 * vy1 * 0.3)
assert np.allclose(f(vx1, vy1, vy2, 0), 2 * vy2 * 0.2)
def test_grad_test_values(self):
# Regression test for test values of `ifelse` gradient.
backup = aesara.config.compute_test_value
aesara.config.compute_test_value = "raise"
try:
x = tensor.scalar("x")
x.tag.test_value = 1
# Used to crash due to undefined test value.
tensor.grad(ifelse(0, x, x), x)
finally:
aesara.config.compute_test_value = backup
def test_grad_int_value(self):
w = aesara.shared(np.random.rand(10))
b = aesara.shared(np.random.rand())
params = [w, b]
x = vector()
y = scalar()
score = w.dot(x) + b
correct = score * y > 0
loss = ifelse(correct, 0, 1)
[(param, param - 0.5 * aesara.grad(cost=loss, wrt=param)) for param in params]
def test_pushout1(self):
x1 = scalar("x1")
x2 = scalar("x2")
y1 = scalar("y1")
y2 = scalar("y2")
w1 = scalar("w1")
w2 = scalar("w2")
c = iscalar("c")
x, y = ifelse(c, (x1, y1), (x2, y2), name="f1")
z = ifelse(c, w1, w2, name="f2")
out = x * z * y
f = function([x1, x2, y1, y2, w1, w2, c], out, allow_input_downcast=True)
assert isinstance(f.maker.fgraph.toposort()[-1].op, IfElse)
rng = np.random.RandomState(utt.fetch_seed())
vx1 = rng.uniform()
vx2 = rng.uniform()
vy1 = rng.uniform()
vy2 = rng.uniform()
vw1 = rng.uniform()
vw2 = rng.uniform()
assert np.allclose(f(vx1, vx2, vy1, vy2, vw1, vw2, 1), vx1 * vy1 * vw1)
assert np.allclose(f(vx1, vx2, vy1, vy2, vw1, vw2, 0), vx2 * vy2 * vw2)
def test_lazy_if(self):
# Tests that lazy if works .. even if the two results have different
# shapes but the same type (i.e. both vectors, or matrices or
# whatnot of same dtype)
x = vector("x", dtype=self.dtype)
y = vector("y", dtype=self.dtype)
c = iscalar("c")
f = function([c, x, y], ifelse(c, x, y), mode=self.mode)
self.assertFunctionContains1(f, self.get_ifelse(1))
rng = np.random.RandomState(utt.fetch_seed())
xlen = rng.randint(200)
ylen = rng.randint(200)
vx = np.asarray(rng.uniform(size=(xlen,)), self.dtype)
vy = np.asarray(rng.uniform(size=(ylen,)), self.dtype)
assert np.allclose(vx, f(1, vx, vy))
assert np.allclose(vy, f(0, vx, vy))
def test_multiple_out_crash(self):
# This test failed up to commit 2faeb62c38
p0 = self.shared(np.asarray(np.random.random([4, 8]),
dtype=self.dtype))
p1 = self.shared(np.asarray(np.random.random(8), dtype=self.dtype))
p2 = self.shared(np.asarray(np.random.random([8, 3]),
dtype=self.dtype))
p3 = self.shared(np.asarray(np.random.random(3), dtype=self.dtype))
p = [p0, p1, p2, p3]
# in my code these vars are the result of applying scan
ften0 = tensor3("ft0", dtype=self.dtype)
fmat1 = matrix("fm1", dtype=self.dtype)
ften2 = tensor3("ft2", dtype=self.dtype)
fmat3 = matrix("fm3", dtype=self.dtype)
# then I keep only the last iteration
fsub0 = ften0[-1]
fsub1 = fmat1[-1]
fsub2 = ften2[-1]
fsub3 = fmat3[-1]
fsub = [fsub0, fsub1, fsub2, fsub3]
acc = at.constant(1, "int8") >= 0
new_positions = ifelse(acc, fsub, p)
new_updates = [(p[0], new_positions[0])]
f = function([ften0, fmat1, ften2, fmat3], [],
updates=new_updates,
mode=self.mode)
self.assertFunctionContains1(f, self.get_ifelse(4))
i1 = np.asarray(np.random.random([19, 4, 8]), dtype=self.dtype)
i2 = np.asarray(np.random.random([19, 8]), dtype=self.dtype)
i3 = np.asarray(np.random.random([19, 8, 3]), dtype=self.dtype)
i4 = np.asarray(np.random.random([19, 3]), dtype=self.dtype)
f(i1, i2, i3, i4)
def test_sparse_tensor_error(self):
pytest.importorskip("scipy", minversion="0.7.0")
import aesara.sparse
rng = np.random.RandomState(utt.fetch_seed())
data = rng.rand(2, 3).astype(self.dtype)
x = self.shared(data)
y = aesara.sparse.matrix("csc", dtype=self.dtype, name="y")
z = aesara.sparse.matrix("csr", dtype=self.dtype, name="z")
cond = iscalar("cond")
with pytest.raises(TypeError):
ifelse(cond, x, y)
with pytest.raises(TypeError):
ifelse(cond, y, x)
with pytest.raises(TypeError):
ifelse(cond, x, z)
with pytest.raises(TypeError):
ifelse(cond, z, x)
with pytest.raises(TypeError):
ifelse(cond, y, z)
with pytest.raises(TypeError):
ifelse(cond, z, y)
