def test_zscore_shift():
time = pd.date_range(start='2018-01-01', end='2020-01-01')
data_X = np.zeros(len(time))
data_y = np.ones(len(time))
X = xr.DataArray(data_X,
name='foo',
dims=['index'],
coords={
'index': time
}).to_dataframe()
y = xr.DataArray(data_y,
name='foo',
dims=['index'],
coords={
'index': time
}).to_dataframe()
shift_expected = xr.DataArray(np.ones(364),
name='foo',
dims=['day'],
coords={
'day': np.arange(1, 365)
}).to_series()
zscore = ZScoreRegressor()
zscore.fit(X, y)
np.testing.assert_allclose(zscore.shift_, shift_expected)
def test_zscore_scale():
time = pd.date_range(start='2018-01-01', end='2020-01-01')
data_X = np.linspace(0, 1, len(time))
data_y = data_X * 2
X = xr.DataArray(data_X,
name='foo',
dims=['index'],
coords={
'index': time
}).to_dataframe()
y = xr.DataArray(data_y,
name='foo',
dims=['index'],
coords={
'index': time
}).to_dataframe()
data_scale_expected = [2 for i in np.zeros(364)]
scale_expected = xr.DataArray(data_scale_expected,
name='foo',
dims=['day'],
coords={
'day': np.arange(1, 365)
}).to_series()
zscore = ZScoreRegressor()
zscore.fit(X, y)
np.testing.assert_allclose(zscore.scale_, scale_expected)
def test_zscore_shift():
time = pd.date_range(start="2018-01-01", end="2020-01-01")
data_X = np.zeros(len(time))
data_y = np.ones(len(time))
X = xr.DataArray(data_X,
name="foo",
dims=["index"],
coords={
"index": time
}).to_dataframe()
y = xr.DataArray(data_y,
name="foo",
dims=["index"],
coords={
"index": time
}).to_dataframe()
shift_expected = xr.DataArray(np.ones(364),
name="foo",
dims=["day"],
coords={
"day": np.arange(1, 365)
}).to_series()
zscore = ZScoreRegressor()
zscore.fit(X, y)
np.testing.assert_allclose(zscore.shift_, shift_expected)
def test_zscore_predict():
time = pd.date_range(start='2018-01-01', end='2020-01-01')
data_X = np.linspace(0, 1, len(time))
X = xr.DataArray(data_X,
name='foo',
dims=['index'],
coords={
'index': time
}).to_dataframe()
shift = xr.DataArray(np.zeros(364),
name='foo',
dims=['day'],
coords={
'day': np.arange(1, 365)
}).to_series()
scale = xr.DataArray(np.ones(364),
name='foo',
dims=['day'],
coords={
'day': np.arange(1, 365)
}).to_series()
zscore = ZScoreRegressor()
zscore.shift_ = shift
zscore.scale_ = scale
i = int(zscore.window_width / 2)
expected = xr.DataArray(data_X,
name='foo',
dims=['index'],
coords={
'index': time
}).to_dataframe()
expected[0:i] = 'NaN'
expected[-i:] = 'NaN'
out = zscore.predict(X)
np.testing.assert_allclose(out.astype(float), expected.astype(float))
def test_zscore_predict():
time = pd.date_range(start="2018-01-01", end="2020-01-01")
data_X = np.linspace(0, 1, len(time))
X = xr.DataArray(data_X,
name="foo",
dims=["index"],
coords={
"index": time
}).to_dataframe()
shift = xr.DataArray(np.zeros(364),
name="foo",
dims=["day"],
coords={
"day": np.arange(1, 365)
}).to_series()
scale = xr.DataArray(np.ones(364),
name="foo",
dims=["day"],
coords={
"day": np.arange(1, 365)
}).to_series()
zscore = ZScoreRegressor()
zscore.shift_ = shift
zscore.scale_ = scale
i = int(zscore.window_width / 2)
expected = xr.DataArray(data_X,
name="foo",
dims=["index"],
coords={
"index": time
}).to_dataframe()
expected[0:i] = "NaN"
expected[-i:] = "NaN"
np.testing.assert_allclose(
zscore.predict(X).astype(float), expected.astype(float))
LinearTrendTransformer,
PaddedDOYGrouper,
PureAnalog,
QuantileMapper,
QuantileMappingReressor,
ZScoreRegressor,
)
@parametrize_with_checks([
# Regressors
AnalogRegression(),
BcsdPrecipitation(),
BcsdTemperature(),
PureAnalog(),
ZScoreRegressor(),
QuantileMappingReressor(n_endpoints=2),
# transformers
LinearTrendTransformer(),
QuantileMapper(),
])
def test_sklearn_compatible_estimator(estimator, check):
check(estimator)
def test_linear_trend_roundtrip():
# TODO: there is probably a better analytic test here
n = 100
trend = 1
yint = 15