def optimise_targets(self):
"""If any target observations have been made, attempt to optimise them
all."""
if not self.should_optimise:
return
from hypothesis.internal.conjecture.optimiser import Optimiser
# We want to avoid running the optimiser for too long in case we hit
# an unbounded target score. We start this off fairly conservatively
# in case interesting examples are easy to find and then ramp it up
# on an exponential schedule so we don't hamper the optimiser too much
# if it needs a long time to find good enough improvements.
max_improvements = 10
while True:
prev_calls = self.call_count
any_improvements = False
for target, data in list(
self.best_examples_of_observed_targets.items()):
optimiser = Optimiser(self,
data,
target,
max_improvements=max_improvements)
optimiser.run()
if optimiser.improvements > 0:
any_improvements = True
if self.interesting_examples:
break
max_improvements *= 2
if any_improvements:
continue
self.pareto_optimise()
if prev_calls == self.call_count:
break
def new_optimiser(self, example, target):
from hypothesis.internal.conjecture.optimiser import Optimiser
return Optimiser(self, example, target)