def test_polynomial_extrapolator(self):
"""
Test extrapolation using a polynomial extrapolator with degree = 1 using all previous points
in the parameters for extrapolation. This test confirms that the extrapolation of the
parameters has a specified error relative to the actual parameter values.
NOTE: The polynomial fit may give a runtime warning if the data is poorly fitted.
This depends on degree and dataset and may need be tuned by the user to achieve
optimal results. This reasoning holds for any instance using an internal
polynomial extrapolator.
"""
points = 0.7
params = PolynomialExtrapolator(degree=3).extrapolate(
points=[points], param_dict=PARAM_DICT)
sq_diff = [
(actual - expected)**2
for actual, expected in zip(params[points], PARAM_DICT[points])
]
self.assertLess(sum(sq_diff), 1e-3)
def test_sieve_poly_window_extrapolator(self):
"""
Test extrapolation using an Sieve/Clustering Extrapolator using a data window/lookback of
3 points and an internal polynomial extrapolator with degree = 1.
This test confirms that no extrapolation is performed on points before the
data window, i.e, the first two points, and that the extrapolation of the parameters on the
last three points has some specified error relative to the actual parameter values.
"""
points_interspersed = [0.3, 0.5, 0.7, 0.8, 1.5]
sieve_win_extrapolator = SieveExtrapolator(
extrapolator=PolynomialExtrapolator(degree=1), window=3)
params = sieve_win_extrapolator.extrapolate(points=points_interspersed,
param_dict=PARAM_DICT)
self.assertFalse(params.get(0.3))
self.assertFalse(params.get(0.5))
sq_diff_1 = [(actual - expected)**2
for actual, expected in zip(params[0.7], PARAM_DICT[0.7])]
self.assertLess(sum(sq_diff_1), 1e-1)
sq_diff_2 = [(actual - expected)**2
for actual, expected in zip(params[0.8], PARAM_DICT[0.8])]
self.assertLess(sum(sq_diff_2), 1e-1)
sq_diff_3 = [(actual - expected)**2
for actual, expected in zip(params[1.5], PARAM_DICT[1.5])]
self.assertLess(sum(sq_diff_3), 1e-1)
def test_poly_window_extrapolator(self):
"""
Test extrapolation using an WindowExtrapolator using a data window/lookback of 3 points
and an internal polynomial extrapolator with degree = 1. This test confirms that no
extrapolation is performed on points before the data window, i.e, the first two points,
and that the extrapolation of the parameters on the last three points has a error below
a threshold when compared to the actual parameter values.
"""
points_interspersed = [0.3, 0.5, 0.7, 0.8, 1.5]
window_extrapolator = Extrapolator.factory(
"window", extrapolator=PolynomialExtrapolator(degree=1), window=3)
params = window_extrapolator.extrapolate(points=points_interspersed,
param_dict=PARAM_DICT)
self.assertFalse(params.get(0.3))
self.assertFalse(params.get(0.5))
sq_diff_1 = [(actual - expected)**2
for actual, expected in zip(params[0.7], PARAM_DICT[0.7])]
self.assertLess(sum(sq_diff_1), 1e-1)
sq_diff_2 = [(actual - expected)**2
for actual, expected in zip(params[0.8], PARAM_DICT[0.8])]
self.assertLess(sum(sq_diff_2), 1e-2)
sq_diff_3 = [(actual - expected)**2
for actual, expected in zip(params[1.5], PARAM_DICT[1.5])]
self.assertLess(sum(sq_diff_3), 1e-2)
def test_pca_polynomial_window_extrapolator(self):
"""
Test extrapolation using an PCAExtrapolator using a data window/lookback of 3 points
and an internal polynomial extrapolator with degree = 1 using regular PCA as default.
This test confirms that no extrapolation is performed on points before the
data window, i.e, the first two points, and that the extrapolation of the parameters on
last three points has a specified error relative to the actual parameter values.
"""
points_interspersed = [.3, .5, .7, .8, 1.5]
pca_poly_win_ext = PCAExtrapolator(
extrapolator=PolynomialExtrapolator(degree=1), window=3)
params = pca_poly_win_ext.extrapolate(points=points_interspersed,
param_dict=PARAM_DICT)
self.assertFalse(params.get(.3))
self.assertFalse(params.get(.5))
sq_diff_1 = [(actual - expected)**2
for actual, expected in zip(params[.7], PARAM_DICT[.7])]
self.assertLess(sum(sq_diff_1), 1e-2)
sq_diff_2 = [(actual - expected)**2
for actual, expected in zip(params[.8], PARAM_DICT[.8])]
self.assertLess(sum(sq_diff_2), 1e-2)
sq_diff_3 = [(actual - expected)**2
for actual, expected in zip(params[1.5], PARAM_DICT[1.5])]
self.assertLess(sum(sq_diff_3), 1e-2)