def sampler(self):
if self.__sampler is None:
assert self.bias is not None
assert 0 < self.bias < 1
self.__sampler = cu.Sampler(
[self.bias**i for i in range(len(self.element_strategies))])
return self.__sampler
def __init__(self, allow_infinity, allow_nan):
SearchStrategy.__init__(self)
assert isinstance(allow_infinity, bool)
assert isinstance(allow_nan, bool)
self.allow_infinity = allow_infinity
self.allow_nan = allow_nan
self.nasty_floats = [f for f in NASTY_FLOATS if self.permitted(f)]
weights = [0.2 * len(self.nasty_floats)] + [0.8] * len(self.nasty_floats)
self.sampler = d.Sampler(weights)
def __init__(self, strategies, bias=None):
SearchStrategy.__init__(self)
strategies = tuple(strategies)
self.original_strategies = list(strategies)
self.__element_strategies = None
self.bias = bias
self.__in_branches = False
if bias is not None:
assert 0 < bias < 1
self.sampler = cu.Sampler(
[bias ** i for i in range(len(strategies))])
else:
self.sampler = None
def __init__(self, allow_infinity, allow_nan, width):
SearchStrategy.__init__(self)
assert isinstance(allow_infinity, bool)
assert isinstance(allow_nan, bool)
assert width in (16, 32, 64)
self.allow_infinity = allow_infinity
self.allow_nan = allow_nan
self.width = width
self.nasty_floats = [
float_of(f, self.width) for f in NASTY_FLOATS if self.permitted(f)
]
weights = [0.2 * len(self.nasty_floats)] + [0.8] * len(self.nasty_floats)
self.sampler = d.Sampler(weights)
def test_sampler_distribution(weights):
total = sum(weights)
n = len(weights)
assume(total > 0)
probabilities = [w / total for w in weights]
sampler = cu.Sampler(weights)
calculated = [Fraction(0)] * n
for base, alternate, p_alternate in sampler.table:
calculated[base] += (1 - p_alternate) / n
calculated[alternate] += p_alternate / n
assert probabilities == calculated
def test_gives_the_correct_probabilities():
weights = [Fraction(1), Fraction(9)]
total = sum(weights)
probabilities = [w / total for w in weights]
sampler = cu.Sampler(probabilities)
assert cu.Sampler(weights).table == sampler.table
counts = [0] * len(weights)
i = 0
while i < 2**16:
data = ConjectureData.for_buffer(int_to_bytes(i, 2))
try:
c = sampler.sample(data)
counts[c] += 1
assert probabilities[c] >= Fraction(counts[c], 2**16)
except StopTest:
pass
if 1 in data.forced_indices:
i += 256
else:
i += 1
class WideRangeIntStrategy(SearchStrategy):
distribution = d.Sampler([4.0, 8.0, 1.0, 1.0, 0.5])
sizes = [8, 16, 32, 64, 128]
def __repr__(self):
return "WideRangeIntStrategy()"
def do_draw(self, data):
size = self.sizes[self.distribution.sample(data)]
r = data.draw_bits(size)
sign = r & 1
r >>= 1
if sign:
r = -r
return int(r)
def test_sampler_distribution(weights):
total = sum(weights)
n = len(weights)
assume(total > 0)
probabilities = [w / total for w in weights]
sampler = cu.Sampler(weights)
calculated = [Fraction(0)] * n
for base, alternate, p_alternate in sampler.table:
calculated[base] += (1 - p_alternate) / n
calculated[alternate] += p_alternate / n
for expected, actual in zip(probabilities, calculated):
if isinstance(actual, Fraction):
assert expected == actual
else:
assert abs(expected - actual) < 0.001
def test_sampler_shrinks():
sampler = cu.Sampler([4.0, 8.0, 1.0, 1.0, 0.5])
assert sampler.sample(ConjectureData.for_buffer([0] * 3)) == 0
def test_sampler_does_not_draw_minimum_if_zero():
sampler = cu.Sampler([0, 2, 47])
assert sampler.sample(ConjectureData.for_buffer([0, 0])) != 0
