def test_replace_nan(x):
x_clean = assessment.replace_nan(x)
assert isinstance(x_clean, type(x))
try:
assert np.nan not in x_clean
except:
assert not np.isnan(x_clean)
def measure(individual):
func = numpy_phenotype(individual)
data = load_boston()
yhat = func(*data.data.T)
if np.isscalar(yhat):
yhat = np.ones_like(data.target) * yhat
fit = nrmse(data.target, yhat), len(individual)
return replace_nan(fit)
def rmse(self, individual, *f_args):
y = self.trajectory(individual, *f_args)
# 2/NT * ∫y0^2 dt from 0 to N*T.
ampl_ = np.sqrt(scipy.trapz(y[0, :]**2, x=self.x) * 2.0 / self.NT)
# 2/NTA^2 * ∫y1^2 dt from 0 to N*T.
omega_ = np.sqrt(
scipy.trapz(y[1, :]**2, x=self.x) * 2.0 / (self.NT * ampl_**2))
rmse_ampl = utils.numeric.rmse(self.ampl, ampl_)
rmse_omega = utils.numeric.rmse(self.omega, omega_)
# Alternative measure.
# rmse_ampl = utils.numeric.rmse(self.ampl * np.sin(self.omega * self.x), y[0, :])
# rmse_omega = utils.numeric.rmse(self.ampl * self.omega * np.cos(self.omega * self.x), y[1, :])
return assessment.replace_nan((rmse_ampl, rmse_omega))
def rmse(self, individual, *f_args):
y = self.trajectory(individual, *f_args)
rmse_y_0 = utils.numeric.rmse(self.target, y[0, :])
rmse_y_1 = utils.numeric.rmse(self.target, y[1, :])
rmse_y_2 = utils.numeric.rmse(self.target, y[2, :])
return assessment.replace_nan((rmse_y_0, rmse_y_1, rmse_y_2))
def measure(self, individual):
y = self.trajectory(individual)
rmse_y_0 = utils.numeric.rmse(self.target, y[0, :])
return assessment.replace_nan((rmse_y_0, len(individual)))
