def test_par_compose():
x = BV.uatom(3, 'x').with_output('x')
y = BV.uatom(3, 'y').with_output('y')
pcirc_x = C.PCirc(circ=x)
pcirc_y = C.PCirc(circ=y, dist_map={'y': lambda _: 1 / 3})
pcirc_xy = pcirc_x | pcirc_y
assert pcirc_xy.inputs == {'x'}
assert pcirc_xy.outputs == {'x', 'y'}
assert pcirc_xy.dist_map['y'](0) == 1 / 3
def test_seq_compose():
x = BV.uatom(3, 'x').with_output('y')
y = BV.uatom(3, 'y').with_output('y')
pcirc = C.PCirc(circ=y)
pcirc2 = C.PCirc(circ=x, dist_map={'x': lambda _: 1 / 3}) >> pcirc
pcirc3 = pcirc << C.PCirc(circ=x, dist_map={'x': lambda _: 1 / 3})
assert pcirc2.outputs == pcirc3.outputs == {'y'}
assert pcirc2.inputs == pcirc3.inputs == set()
assert pcirc2.dist_map['x'](0) == pcirc3.dist_map['x'](0) == 1 / 3
assert 0 <= pcirc2({})[0]['y'] <= 7
pcirc4 = C.PCirc(circ=(x + 1).with_output('y')) >> pcirc
assert pcirc4({'x': 0})[0] == {'y': 1}
def test_loopback_unroll():
x = BV.uatom(3, 'x')
y = BV.uatom(3, 'y')
adder = (x + y).with_output('z')
pcirc = C.PCirc(adder, dist_map={'y': lambda _: 1/3}) \
.assume((y > 0) & (y < 4))
pcirc2 = pcirc.loopback({
'input': 'x',
'output': 'z',
'init': 4,
'keep_output': True,
})
assert pcirc2.inputs == set()
assert pcirc2.outputs == {'z'}
assert len(pcirc2.latches) == 1
assert 4 < pcirc2({})[0]['z'] < 8
pcirc3 = pcirc2.unroll(3)
assert pcirc3.inputs == set()
assert pcirc3.outputs == {'z##time_1', 'z##time_2', 'z##time_3'}
pcirc4 = pcirc2.unroll(3, only_last_outputs=True)
assert pcirc4.outputs == {'z##time_3'}
pcirc4({}) # Could technically be any value due to roll back.
def test_pcirc_relabel():
x = BV.uatom(3, 'x')
pcirc = C.PCirc(circ=x, dist_map={'x': lambda _: 1 / 3})
pcirc2 = pcirc['i', {'x': 'y'}]
assert pcirc2.inputs == set()
assert pcirc2.circ.inputs == {'y'}
assert pcirc2.dist_map['y'](0) == 1 / 3
def prob(self, guard, *, log=False):
assert all(self.is_random(i) for i in guard.inputs)
order = list(guard.inputs)
guard = guard.bundle_inputs('##coins', order=order)
coin_biases = fn.lmapcat(self.coin_biases, order)
pcirc = C.PCirc(
circ=guard.aigbv,
coin_biases=len(order) * coin_biases,
coins_id='##coins',
)
return C.infer.prob(pcirc, log=log)
def test_pcirc_smoke():
x = BV.uatom(3, 'x')
y = BV.uatom(3, 'y')
z = (x + y).with_output('z')
pcirc = C.PCirc(circ=z, dist_map={'y': lambda _: 1/3}) \
.assume(y <= 2)
rvar = C.RandomVarCirc(pcirc)
# Warning. May be flaky.
for i in range(3):
assert 3 <= rvar({'x': 3}) <= 5