def test_from_array_with_scale_positive(self):
array = numpy.array([[0, 0, 0], [10, 0, 0], [0, 10, 0], [10, 10, 10]])
bounds = Bounds.from_array(array, scale=1.1)
assert (bounds.low == numpy.array([0, 0,
0])).all(), (bounds.low,
array.min(axis=0))
assert (bounds.high == numpy.array([11, 11,
11])).all(), (bounds.high,
array.max(axis=0))
assert bounds.shape == (3, )
array = numpy.array([[-10, 0, 0], [-10, 0, 0], [0, -10, 0],
[-10, -10, -10]])
bounds = Bounds.from_array(array, scale=1.1)
assert (bounds.high == numpy.array([0, 0,
0])).all(), (bounds.high,
array.max(axis=0))
assert (bounds.low == numpy.array([-11, -11,
-11])).all(), (bounds.low,
array.min(axis=0))
assert bounds.shape == (3, )
array = numpy.array([[10, 10, 10], [100, 10, 10], [10, 100, 10],
[100, 100, 100]])
bounds = Bounds.from_array(array, scale=1.1)
assert numpy.allclose(bounds.low, numpy.array([9.0, 9.0, 9])), (
bounds.low,
array.min(axis=0),
)
assert numpy.allclose(bounds.high, numpy.array([110, 110, 110])), (
bounds.high,
array.max(axis=0),
)
assert bounds.shape == (3, )
def test_from_array_with_scale_negative(self):
# high +, low +, scale > 1
array = numpy.array([[-10, 0, 0], [-10, 0, 0], [0, -10, 0],
[-10, -10, -10]])
bounds = Bounds.from_array(array, scale=0.9)
assert (bounds.high == numpy.array([0, 0,
0])).all(), (bounds.high,
array.max(axis=0))
assert (bounds.low == numpy.array([-9, -9,
-9])).all(), (bounds.low,
array.min(axis=0))
assert bounds.shape == (3, )
array = numpy.array([[0, 0, 0], [10, 0, 0], [0, 10, 0], [10, 10, 10]])
bounds = Bounds.from_array(array, scale=0.9)
assert (bounds.low == numpy.array([0, 0, 0])).all(), (bounds, array)
assert (bounds.high == numpy.array([9, 9, 9])).all()
assert bounds.shape == (3, )
# high +, low +, scale < 1
array = numpy.array([[10, 10, 10], [100, 10, 10], [10, 100, 10],
[100, 100, 100]])
bounds = Bounds.from_array(array, scale=0.9)
assert numpy.allclose(bounds.low,
numpy.array([9, 9, 9])), (bounds.low,
array.min(axis=0))
assert numpy.allclose(bounds.high, numpy.array([90, 90, 90])), (
bounds.high,
array.max(axis=0),
)
assert bounds.shape == (3, )
# high -, low -, scale > 1
array = numpy.array([[-100, -10, -10], [-100, -10, -10],
[-10, -100, -10], [-100, -100, -100]])
bounds = Bounds.from_array(array, scale=1.1)
assert numpy.allclose(bounds.high, numpy.array([-9, -9, -9])), (
bounds.high,
array.max(axis=0),
)
assert numpy.allclose(bounds.low, numpy.array([-110, -110, -110])), (
bounds.low,
array.min(axis=0),
)
assert bounds.shape == (3, )
# high -, low -, scale < 1
array = numpy.array([[-100, -10, -10], [-100, -10, -10],
[-10, -100, -10], [-100, -100, -100]])
bounds = Bounds.from_array(array, scale=0.9)
assert numpy.allclose(bounds.high, numpy.array([-11, -11, -11])), (
bounds.high,
array.max(axis=0),
)
assert numpy.allclose(bounds.low, numpy.array([-90, -90, -90])), (
bounds.low,
array.min(axis=0),
)
assert bounds.shape == (3, )
def test_safe_margin(self, bounds_fixture: Bounds):
new_bounds = bounds_fixture.safe_margin()
assert numpy.allclose(new_bounds.low, bounds_fixture.low)
assert numpy.allclose(new_bounds.high, bounds_fixture.high)
low = numpy.full_like(bounds_fixture.low, -10)
new_bounds = bounds_fixture.safe_margin(low=low)
assert numpy.allclose(new_bounds.high, bounds_fixture.high)
assert numpy.allclose(new_bounds.low, low)
new_bounds = bounds_fixture.safe_margin(low=low, scale=2)
assert numpy.allclose(new_bounds.high, bounds_fixture.high * 2)
assert numpy.allclose(new_bounds.low, low * 2)
def create_bounds(name):
if name == "scalars":
return lambda: Bounds(high=5, low=-5, shape=(3, ))
elif name == "high_array":
return lambda: Bounds(high=numpy.array([1, 2, 5]), low=-5)
elif name == "low_array":
return lambda: Bounds(low=numpy.array([-1, -5, -3]), high=5)
elif name == "both_array":
array = numpy.array([1, 2, 5])
return lambda: Bounds(high=array, low=-array)
elif name == "high_list":
return lambda: Bounds(low=numpy.array([-5, -2, -3]), high=[5, 5, 5])
def _update_lims(self, swarm, X: numpy.ndarray):
backup_bounds = swarm.env.bounds if swarm is not None else Bounds.from_array(
X)
bounds = (swarm.critic.bounds if has_embedding(swarm)
and self.use_embeddings else backup_bounds)
self.xlim, self.ylim = bounds.safe_margin(low=self.low,
high=self.high).to_lims()
def test_clip(self):
tup = ((-1, 10), (-3, 4), (2, 5))
array = numpy.array([[-10, 0, 0], [11, 0, 0], [0, 11, 0], [11, 11,
11]])
bounds = Bounds.from_tuples(tup)
clipped = bounds.clip(array)
target = numpy.array([[-1, 0, 2], [10, 0, 2], [0, 4, 2], [10, 4, 5]])
assert numpy.allclose(clipped, target), (clipped, target)
def function_env():
bounds = Bounds(shape=(2, ), high=1, low=1, dtype=int)
env = Function(function=lambda x: np.ones(N_WALKERS), bounds=bounds)
params = {
"actions": {
"dtype": np.int64,
"size": (2, )
},
"dt": {
"dtype": np.float32
}
}
states = States(state_dict=params, batch_size=N_WALKERS)
return env, states
def get_z_coords(self, swarm: Swarm, X: numpy.ndarray = None):
if swarm is None:
return numpy.ones(self.n_points**self.n_points)
if swarm.critic.bounds is None:
swarm.critic.bounds = Bounds.from_array(X, scale=1.1)
# target grid to interpolate to
xi = numpy.linspace(swarm.critic.bounds.low[0],
swarm.critic.bounds.high[0], self.n_points)
yi = numpy.linspace(swarm.critic.bounds.low[1],
swarm.critic.bounds.high[1], self.n_points)
xx, yy = numpy.meshgrid(xi, yi)
grid = numpy.c_[xx.ravel(), yy.ravel()]
if swarm.swarm.critic.warmed:
memory_values = swarm.swarm.critic.predict(grid)
memory_values = relativize(-memory_values)
else:
memory_values = numpy.arange(grid.shape[0])
return memory_values
def get_z_coords(self, swarm: Swarm, X: numpy.ndarray = None):
if swarm is None:
return numpy.ones(self.n_points**self.n_points)
if swarm.critic.bounds is None:
swarm.critic.bounds = Bounds.from_array(X, scale=1.1)
# target grid to interpolate to
xi = numpy.linspace(swarm.critic.bounds.low[0],
swarm.critic.bounds.high[0], self.n_points)
yi = numpy.linspace(swarm.critic.bounds.low[1],
swarm.critic.bounds.high[1], self.n_points)
xx, yy = numpy.meshgrid(xi, yi)
grid = numpy.c_[xx.ravel(), yy.ravel()]
if swarm.swarm.critic.warmed:
memory_values = swarm.swarm.critic.model.transform(grid)
memory_values = numpy.array([
swarm.swarm.critic.memory[ix[0], ix[1]].astype(numpy.float32)
for ix in memory_values.astype(int)
])
else:
memory_values = numpy.arange(grid.shape[0])
return memory_values
def test_from_array(self):
array = numpy.array([[0, 0, 0], [11, 0, 0], [0, 11, 0], [11, 11, 11]])
bounds = Bounds.from_array(array)
assert (bounds.low == numpy.array([0, 0, 0])).all()
assert (bounds.high == numpy.array([11, 11, 11])).all()
assert bounds.shape == (3, )
def test_from_tuples(self):
tup = ((-1, 2), (-3, 4), (2, 5))
bounds = Bounds.from_tuples(tup)
assert (bounds.low == numpy.array([-1, -3, 2])).all()
assert (bounds.high == numpy.array([2, 4, 5])).all()
def _update_lims(self, swarm, X: numpy.ndarray):
backup_bounds = swarm.env.bounds if swarm is not None else Bounds.from_array(
X)
bounds = (swarm.critic.bounds if has_embedding(swarm)
and self.use_embeddings else backup_bounds)
self.xlim, self.ylim = bounds.to_tuples()[:2]
