def _get_MCL_lambda(self, obj):
xp = obj.get_array_module()
k, v, o, d = self.residual_[1].shape
w = xp.reshape(self.residual_[1], (k, v, o * d))
w[w == 0] = xp.nan
w = w * 0 + 1
lambdaI = (
WeightedRegression(thru_orig=True, axis=-1, xp=xp)
.fit(
xp.reshape(self.q_resid_[1][..., :-1, :-1], (k, v, o * d)),
xp.reshape(self.residual_[1], (k, v, o * d)),
w,
)
.slope_
)
lambdaP = (
WeightedRegression(thru_orig=True, axis=-1, xp=xp)
.fit(
xp.reshape(self.q_resid_[0][..., :-1, :-1], (k, v, o * d)),
xp.reshape(self.residual_[0], (k, v, o * d)),
w,
)
.slope_
)
return self._p_to_i_concate(lambdaP, lambdaI, xp)[..., None]
def _get_MCL_lambda(self):
k, v, o, d = self.residual_[1].shape
w = np.reshape(self.residual_[1], (k, v, o * d))
w[w == 0] = np.nan
w = w * 0 + 1
lambdaI = WeightedRegression(thru_orig=True, axis=-1).fit(
np.reshape(self.q_resid_[1][..., :-1, :-1], (k, v, o * d)),
np.reshape(self.residual_[1], (k, v, o * d)), w).slope_
lambdaP = WeightedRegression(thru_orig=True, axis=-1).fit(
np.reshape(self.q_resid_[0][..., :-1, :-1], (k, v, o * d)),
np.reshape(self.residual_[0], (k, v, o * d)), w).slope_
return self._p_to_i_concate(lambdaP, lambdaI)[..., np.newaxis]
def _get_MCL_lambda(self):
k, v, o, d = self.residual_[1].shape
w = np.reshape(self.residual_[1], (k, v, o * d))
w[w == 0] = np.nan
w = w * 0 + 1
lambdaI = WeightedRegression(
w,
x=np.reshape(self.q_resid_[1][..., :-1, :-1], (k, v, o * d)),
y=np.reshape(self.residual_[1], (k, v, o * d)),
thru_orig=True,
axis=-1).fit().slope_
lambdaP = WeightedRegression(
w,
x=np.reshape(self.q_resid_[0][..., :-1, :-1], (k, v, o * d)),
y=np.reshape(self.residual_[0], (k, v, o * d)),
thru_orig=True,
axis=-1).fit().slope_
return np.expand_dims(self._p_to_i_concate(lambdaP, lambdaI), -1)
def _get_MCL_model(self, X):
p, i = self.p_to_i_X_[0], self.p_to_i_X_[1]
modelsP = WeightedRegression(axis=2, thru_orig=True)
modelsP = modelsP.fit(p, i, 1 / p).sigma_fill(X.sigma_interpolation)
modelsI = WeightedRegression(axis=2, thru_orig=True)
modelsI = modelsI.fit(i, p, 1 / i).sigma_fill(X.sigma_interpolation)
q_f = self._p_to_i_concate(modelsP.slope_, modelsI.slope_)
rho_sigma = self._p_to_i_concate(modelsP.sigma_, modelsI.sigma_)
return np.swapaxes(q_f, -1, -2), np.swapaxes(rho_sigma, -1, -2)