def setUp(self):
self.n = 60
self.names = ["obj.y"]
self.rng = np.random.default_rng(0)
self.y = self.rng.normal(size=self.n)
self.monitor = AttributeMonitor(self.names)
self.monitor.setup(self.n)
def test_monitor_history_is_same_when_chunk_hint_changes(self):
monitor = AttributeMonitor(["output_"])
self.pred.transform(self.x, self.y, chunk_hint=11, monitor=monitor)
monitor_alt = AttributeMonitor(["output_"])
self.pred.transform(self.x, self.y, chunk_hint=11, monitor=monitor_alt)
np.testing.assert_allclose(
monitor.history_.output_, monitor_alt.history_.output_
)
def setUp(self):
self.keys = [
"x", "v", "hi_d", "i", "i_float", "b", "s", "obj", "i_again"
]
self.monitor = AttributeMonitor(self.keys)
self.rng = np.random.default_rng(2)
self.x_data = [1.0, 2, 0.5, -1, 4, 0, 1, 2.1, -5.0, 3.2, 0.0, 0.5]
self.n = len(self.x_data)
self.v_data = self.rng.normal(size=(self.n, 5))
self.hi_d_data = self.rng.normal(size=(self.n, 5, 3, 2, 4))
self.i_data = np.arange(self.n)
self.i_float_data = np.arange(self.n, 0, -1).astype(float)
self.b_data = np.repeat([True, False], self.n // 2)
self.s_data = np.repeat(["foo", "bar", "goo"], self.n // 3)
self.obj_data = self.n * [object()]
self.i_again_data = [1.0] + [2] * (self.n - 1)
class ToTrack(object):
def __init__(self):
self.x = 0.0
self.v = 0
self.hi_d = np.array([])
self.i = 0
self.i_float = 0.0
self.b = False
self.s = ""
self.obj: object = None
self.i_again = 0
self.obj = ToTrack()
self.monitor.setup(self.n)
for x, v, hi_d, i, i_float, b, s, obj, i_again in zip(
self.x_data,
self.v_data,
self.hi_d_data,
self.i_data,
self.i_float_data,
self.b_data,
self.s_data,
self.obj_data,
self.i_again_data,
):
self.obj.x = x
self.obj.v = v
self.obj.hi_d = hi_d
self.obj.i = i
self.obj.i_float = i_float
self.obj.b = b
self.obj.s = s
self.obj.obj = obj
self.obj.i_again = i_again
self.monitor.record(self.obj)
def test_history_same_when_chunk_hint_changes(self):
names = ["coef_", "var_"]
monitor = AttributeMonitor(names)
self.xcorr.transform(self.x, self.y, monitor=monitor, chunk_hint=13)
xcorr_alt = OnlineCrosscorrelation(self.n_features, rate=self.rate)
monitor_alt = AttributeMonitor(names)
xcorr_alt.transform(self.x, self.y, monitor=monitor_alt, chunk_hint=2)
np.testing.assert_allclose(monitor.history_.coef_,
monitor_alt.history_.coef_)
np.testing.assert_allclose(monitor.history_.var_,
monitor_alt.history_.var_)
def test_chunk_hint_does_not_affect_monitoring(self):
names = ["input_", "output_"]
monitor1 = AttributeMonitor(names)
self.arma.transform(X=self.source_data,
monitor=monitor1,
chunk_hint=11)
monitor2 = AttributeMonitor(names)
arma2 = Arma(self.a, self.b)
arma2.transform(X=self.source_data, monitor=monitor2, chunk_hint=23)
np.testing.assert_allclose(monitor1.history_.input_,
monitor2.history_.input_)
np.testing.assert_allclose(monitor1.history_.output_,
monitor2.history_.output_)
def test_monitor_fit(self):
names = ["weights_", "lateral_", "output_"]
monitor = AttributeMonitor(names)
n_samples = 100
rng = np.random.default_rng(1)
x = rng.normal(size=(n_samples, self.n_features))
self.circuit.transform(x, monitor=monitor)
circuit1 = NonRecurrent(rng=self.seed, **self.kwargs)
weights = []
lateral = []
output = []
for crt_x in x:
weights.append(np.copy(circuit1.weights_))
lateral.append(np.copy(circuit1.lateral_))
output.append(np.copy(circuit1.output_))
circuit1.transform([crt_x])
np.testing.assert_allclose(monitor.history_.weights_, weights)
np.testing.assert_allclose(monitor.history_.lateral_, lateral)
np.testing.assert_allclose(monitor.history_.output_, output)
def test_monitor_history_is_returned_when_return_history_is_true(self):
monitor = AttributeMonitor(["output_"])
_, history = self.pred.transform(
self.x, self.y, monitor=monitor, return_history=True
)
self.assertIs(history, monitor.history_)
def test_history_returned_when_return_history_is_true(self):
monitor = AttributeMonitor(["xcorr.coef_"])
_, history = self.xcorr.transform(
self.x, self.y, monitor=monitor, return_history=True
)
self.assertIs(history, monitor.history_)
def test_var_history_matches_var(self):
self.xcorr.transform(self.x[:-1], self.y[:-1])
last_var = self.xcorr.var_
monitor = AttributeMonitor(["var_"])
self.xcorr.transform(self.x[[-1]], self.y[[-1]], monitor=monitor)
self.assertAlmostEqual(monitor.history_.var_[0], last_var)
def test_when_monitor_is_object_history_returned_is_its_attrib(self):
names = ["weights_"]
monitor = AttributeMonitor(names)
_, history = self.wta.transform(
self.predictors, self.dependent, monitor=monitor, return_history=True
)
self.assertIs(history, monitor.history_)
def test_monitor_dispatches_attributes_according_to_trafo_name(self):
monitor = AttributeMonitor(["trafo0.input_", "trafo1.input_"])
self.chain.transform(X=self.x0, monitor=monitor)
for trafo in self.chain._transformers[:-1]:
self.assertEqual(1, len(trafo.transform.call_args[1]["monitor"].names))
self.assertIn("input_", trafo.transform.call_args[1]["monitor"].names)
self.assertNotIn("monitor", self.chain._transformers[-1].transform.call_args[1])
def test_result_is_the_same_with_and_without_monitor(self):
monitor = AttributeMonitor(["xcorr.var_"])
res = self.xcorr.transform(self.x, self.y, monitor=monitor)
xcorr_alt = CrosscorrelationRegressor(
self.n_models, self.n_features, rng=self.seed
)
res_alt = xcorr_alt.transform(self.x, self.y)
np.testing.assert_allclose(res, res_alt)
def test_output_history_same_if_rate_is_constant_then_switches(self):
schedule = np.zeros(self.n_samples)
schedule[:self.n_partial] = self.rate
circuit = NonRecurrent(rate=schedule, **self.kwargs)
monitor = AttributeMonitor(["output_"])
circuit.transform(self.x, monitor=monitor)
np.testing.assert_allclose(
monitor.history_.output_[:self.n_partial],
self.monitor_partial.history_.output_,
)
def test_monitor_as_object(self):
names = ["weights_"]
monitor = AttributeMonitor(names)
self.wta.transform(self.predictors, self.dependent, monitor=monitor)
wta_alt = BioWTARegressor(self.n_models, self.n_features)
_, history_alt = wta_alt.transform(
self.predictors, self.dependent, monitor=names
)
np.testing.assert_allclose(monitor.history_.weights_, history_alt.weights_)
def test_history_same_when_chunk_hint_changes(self):
names = ["weights_", "lateral_", "output_"]
monitor = AttributeMonitor(names)
n_samples = 100
rng = np.random.default_rng(1)
x = rng.normal(size=(n_samples, self.n_features))
self.circuit.transform(x, monitor=monitor, chunk_hint=13)
circuit_alt = NonRecurrent(rng=self.seed, **self.kwargs)
monitor_alt = AttributeMonitor(names)
circuit_alt.transform(x, monitor=monitor_alt, chunk_hint=2)
np.testing.assert_allclose(monitor.history_.weights_,
monitor_alt.history_.weights_)
np.testing.assert_allclose(monitor.history_.lateral_,
monitor_alt.history_.lateral_)
np.testing.assert_allclose(monitor.history_.output_,
monitor_alt.history_.output_)
def test_monitor_var_correct(self):
monitor = AttributeMonitor(["var_"])
self.xcorr.transform(self.x, self.y, monitor=monitor)
acorr_alt = OnlineCrosscorrelation(self.n_features, rate=self.rate)
var_exp = np.zeros(self.n_samples)
for i in range(self.n_samples):
var_exp[i] = acorr_alt.var_
acorr_alt.transform(self.x[[i]], self.y[[i]])
np.testing.assert_allclose(monitor.history_.var_, var_exp)
class TestAttributeMonitorHierarchical(unittest.TestCase):
def setUp(self):
self.n = 60
self.names = ["obj.y"]
self.rng = np.random.default_rng(0)
self.y = self.rng.normal(size=self.n)
self.monitor = AttributeMonitor(self.names)
self.monitor.setup(self.n)
def test_record_works_correctly(self):
for i in range(self.n):
tracked = SimpleNamespace(obj=SimpleNamespace(y=self.y[i]))
self.monitor.record(tracked)
self.assertTrue(hasattr(self.monitor.history_, "obj"))
self.assertTrue(hasattr(self.monitor.history_.obj, "y"))
np.testing.assert_allclose(self.monitor.history_.obj.y, self.y)
def test_record_batch_works_correctly(self):
chunk_size = 13
for i in range(0, self.n, chunk_size):
tracked = SimpleNamespace(obj=SimpleNamespace(
y=self.y[i:i + chunk_size]))
self.monitor.record_batch(tracked)
self.assertTrue(hasattr(self.monitor.history_, "obj"))
self.assertTrue(hasattr(self.monitor.history_.obj, "y"))
np.testing.assert_allclose(self.monitor.history_.obj.y, self.y)
def test_raises_value_error_if_not_all_variables_are_the_same_length(self):
monitor_alt = AttributeMonitor(["x", "y"])
monitor_alt.setup(3)
obj = SimpleNamespace(x=np.zeros(3), y=[1])
with self.assertRaises(ValueError):
monitor_alt.record_batch(obj)
class TestAttributeMonitorStrAndRepr(unittest.TestCase):
def setUp(self):
self.names = ["foo", "bar"]
self.step = 3
self.monitor = AttributeMonitor(self.names, self.step)
def test_str(self):
s_exp = (f"AttributeMonitor(names={str(self.names)}, n_=None, " +
f"step={self.step}, t_=None)")
s = str(self.monitor)
self.assertEqual(s, s_exp)
def test_str_after_setup(self):
n = 12
self.monitor.setup(n)
s_exp = (f"AttributeMonitor(names={str(self.names)}, n_={n}, " +
f"step={self.step}, t_=0)")
s = str(self.monitor)
self.assertEqual(s, s_exp)
def test_repr(self):
r_exp = (f"AttributeMonitor(names={repr(self.names)}, n_=None, " +
f"step={self.step}, t_=None, i_=None, history_=None)")
r = repr(self.monitor)
self.assertEqual(r, r_exp)
def test_repr_after_setup(self):
n = 12
self.monitor.setup(n)
exp_hist = SimpleNamespace()
for name in self.names:
setattr(exp_hist, name, None)
r_exp = (f"AttributeMonitor(names={repr(self.names)}, n_={n}, " +
f"step={self.step}, t_=0, i_=0, history_={repr(exp_hist)})")
r = repr(self.monitor)
self.assertEqual(r, r_exp)
class TestAttributeMonitorInit(unittest.TestCase):
def setUp(self):
self.keys = ["foo", "bar", "x_y", "y.x"]
self.monitor = AttributeMonitor(self.keys)
def test_constructor_default_step_is_one(self):
self.assertEqual(self.monitor.step, 1)
def test_setup_makes_history_keys_for_all_attributes_without_dots(self):
self.monitor.setup(3)
is_in_history = [
key in self.monitor.history_.__dict__ for key in self.keys
if "." not in key
]
self.assertTrue(all(is_in_history))
def test_setup_makes_none_histories_for_all_attributes_without_dots(self):
self.monitor.setup(4)
is_none = [
self.monitor.history_.__dict__[key] is None for key in self.keys
if "." not in key
]
self.assertTrue(all(is_none))
def test_setup_makes_hierarchical(self):
self.monitor.setup(1)
self.assertTrue(hasattr(self.monitor.history_, "y"))
self.assertTrue(hasattr(self.monitor.history_.y, "x"))
def test_monitor_as_sequence(self):
names = ["trafo0.output_", "trafo2.input_"]
monitor = AttributeMonitor(names)
self.chain.transform(X=self.x0, monitor=monitor)
_, history = self.chain.transform(X=self.x0, monitor=names)
np.testing.assert_allclose(
monitor.history_.trafo0.output_, history.trafo0.output_
)
np.testing.assert_allclose(
monitor.history_.trafo2.input_, history.trafo2.input_
)
def test_monitor_nothing(self):
monitor = AttributeMonitor([])
monitor.setup(5)
obj = SimpleNamespace()
monitor.record_batch(obj)
self.assertEqual(len(monitor.history_.__dict__), 0)
def setUp(self):
self.n_features = 5
self.n_components = 3
self.seed = 3
self.kwargs = dict(n_features=self.n_features,
n_components=self.n_components,
rng=self.seed)
self.rng = np.random.default_rng(0)
self.n_samples = 53
self.x = self.rng.normal(size=(self.n_samples, self.n_features))
self.rate = 0.005
self.monitor_full = AttributeMonitor(["output_"])
self.circuit_full = NonRecurrent(rate=self.rate, **self.kwargs)
self.circuit_full.transform(self.x, monitor=self.monitor_full)
self.n_partial = self.n_samples // 2
self.monitor_partial = AttributeMonitor(["output_"])
self.circuit_partial = NonRecurrent(rate=self.rate, **self.kwargs)
self.circuit_partial.transform(self.x[:self.n_partial],
monitor=self.monitor_partial)
def test_monitor_nothing(self):
monitor = AttributeMonitor([])
n = 10
monitor.setup(n)
for i in range(n):
obj = SimpleNamespace()
monitor.record(obj)
self.assertEqual(len(monitor.history_.__dict__), 0)
def test_monitor_as_sequence(self):
names = ["xcorr.var_", "nsm.weights_"]
_, history = self.xcorr.transform(
self.x, self.y, monitor=names, return_history=True
)
monitor = AttributeMonitor(names)
xcorr_alt = CrosscorrelationRegressor(
self.n_models, self.n_features, rng=self.seed
)
_, history_alt = xcorr_alt.transform(
self.x, self.y, monitor=monitor, return_history=True
)
np.testing.assert_allclose(history.xcorr.var_, history_alt.xcorr.var_)
np.testing.assert_allclose(history.nsm.weights_, history_alt.nsm.weights_)
def test_correct(self):
x_data = [1.0, 2, 0.5, -1, 4, 0, 1, 2.1, -5.0, 3.2, 0.0]
step = 3
monitor = AttributeMonitor(["x"], step=step)
monitor.setup(len(x_data))
for x in x_data:
obj = SimpleNamespace(x=x)
monitor.record(obj)
np.testing.assert_allclose(monitor.history_.x, x_data[::step])
def test_store_copies_objects(self):
class ToTrack(object):
def __init__(self):
self.x = None
obj = ToTrack()
x_data = [None, [1, 2, 3]]
monitor = AttributeMonitor(["x"])
monitor.setup(len(x_data))
for x in x_data:
obj.x = x
monitor.record(obj)
x0 = list(x_data[1])
x_data[1][0] = 0
np.testing.assert_allclose(monitor.history_.x[1], x0)
def test_one_by_one_same_as_batch_when_step_is_one(self):
monitor_alt = AttributeMonitor(self.names)
monitor_alt.setup(self.n)
rng = np.random.default_rng(0)
v = rng.normal(size=self.n)
for i in range(self.n):
obj = SimpleNamespace(x=v[i])
self.monitor.record(obj)
i = 0
while i < self.n:
k = min(rng.integers(1, 10), self.n - i)
crt_v = v[i:i + k]
obj_batch = SimpleNamespace(x=crt_v)
monitor_alt.record_batch(obj_batch)
i += k
np.testing.assert_allclose(self.monitor.history_.x,
monitor_alt.history_.x)
def test_monitor_output(self):
monitor = AttributeMonitor(["output_"])
norms = self.variance.transform(X=self.x, monitor=monitor)
np.testing.assert_allclose(monitor.history_.output_, norms)
def test_monitor_coef_same_as_return_from_transform(self):
monitor = AttributeMonitor(["coef_"])
res = self.xcorr.transform(self.x, self.y, monitor=monitor)
np.testing.assert_allclose(res, monitor.history_.coef_)
