def test_labeller_invalid_percentage():
labeller = Labeller(size=6,
func=np.max,
func_params={},
percentiles=[101],
n_steps_future=2)
with pytest.raises(ValueError):
labeller.fit_transform_resample(X, signal)
def test_labeller_shape():
size = 4
labeller = Labeller(size=size,
func=np.std,
func_params={},
percentiles=None,
n_steps_future=1)
signal_transformed = labeller.fit_transform(signal)
assert signal_transformed.shape == (20 - size + 1, )
def test_labeller_with_percentage():
size = 6
n_steps_future = 1
labeller = Labeller(size=size,
func=np.max,
func_params={},
percentiles=[100],
n_steps_future=n_steps_future)
labeller.fit(X)
assert np.max(X) == labeller.thresholds_[0]
def test_labeller_resampled():
size = 6
labeller = Labeller(size=size,
func=np.max,
func_params={},
percentiles=None,
n_steps_future=1)
x, y = labeller.fit_transform_resample(X, X)
assert_almost_equal(
y, np.array([5, 6, 7, 8, 9, 9, 9, 9, 9, 9, 5, 6, 7, 8, 9]))
assert len(x) == len(y)
# Test behaviour when n_steps_future = size - 1
labeller.set_params(n_steps_future=size - 1)
x, y = labeller.fit_transform_resample(X, X)
assert_almost_equal(
y, np.array([5, 6, 7, 8, 9, 9, 9, 9, 9, 9, 5, 6, 7, 8, 9]))
assert len(x) == len(y)
# Test behaviour when n_steps_future > size - 1
labeller.set_params(n_steps_future=size)
x, y = labeller.fit_transform_resample(X, X)
assert_almost_equal(y, np.array([6, 7, 8, 9, 9, 9, 9, 9, 9, 5, 6, 7, 8,
9]))
assert len(x) == len(y)
def test_labeller_transformed():
size = 6
n_steps_future = 1
labeller = Labeller(size=size,
func=np.max,
func_params={},
percentiles=None,
n_steps_future=n_steps_future)
x, y = labeller.fit_transform_resample(X, X)
assert_almost_equal(x, X[(size - 2):-n_steps_future])
assert len(x) == len(y)
# # Let us say that we simply wish to predict the future of a time series from itself. This is very common in the study of financial markets for example. ``giotto-tda`` provides convenience classes for target preparation from a time series. This notebook only shows a very simple example: many more options are described in ``Labeller``'s documentation. # # 我们希望预测一个时间序列的未来。例如，这在金融市场研究中非常普遍。 giotto-tda提供了方便的类，用于根据时间序列进行目标准备。该笔记本仅显示一个非常简单的示例：Labeller的文档中描述了更多选项。 # # If we wished to create a target ``y`` from ``X`` such that ``y[i]`` is equal to ``X[i + 1]``, while also modifying ``X`` and ``y`` so that they still have the same length, we could proceed as follows: # # 如果我们希望从“ X”创建目标“ y”，以使“ y [i]”等于“ X [i + 1]”，同时还要修改“ X”和``y''以便它们仍然具有相同的长度，我们可以按照以下步骤进行操作： # In[23]: from gtda.time_series import Labeller X = np.arange(10) Lab = Labeller(size=1, func=np.max) Xl, yl = Lab.fit_transform_resample(X, X) Xl, yl # 注意，在这种情况下，我们将X的两个副本提供给fit_transform_resample！ # 这就是使用端到端管道为将来的拓扑预测做准备的样子。同样，鼓励您在混合中包括自己的非拓扑功能！ # In[24]: SW = SlidingWindow(size=5) TE = TakensEmbedding(time_delay=1, dimension=2) VR = VietorisRipsPersistence() Ampl = Amplitude() RFR = RandomForestRegressor() # Full pipeline including the regressor
def test_labeller_params(size):
labeller = Labeller(size=size)
with pytest.raises(ValueError):
labeller.fit(signal)