def compute_functional_and_gradients(self):
f = scaling.wilson_total_nll(d_star_sq=self.d_star_sq,
f_obs=self.f_obs,
sigma_f_obs=self.sigma_f_obs,
epsilon=self.epsilon,
sigma_sq=self.sigma_prot_sq,
gamma_prot=self.gamma_prot,
centric=self.centric,
p_scale=self.x[0],
p_B_wilson=self.x[1])
g = flex.double(
scaling.wilson_total_nll_gradient(d_star_sq=self.d_star_sq,
f_obs=self.f_obs,
sigma_f_obs=self.sigma_f_obs,
epsilon=self.epsilon,
sigma_sq=self.sigma_prot_sq,
gamma_prot=self.gamma_prot,
centric=self.centric,
p_scale=self.x[0],
p_B_wilson=self.x[1]))
self.f = f
return f, g
def compute_functional_and_gradients(self):
f = scaling.wilson_total_nll(
d_star_sq=self.d_star_sq,
f_obs=self.f_obs,
sigma_f_obs=self.sigma_f_obs,
epsilon=self.epsilon,
sigma_sq=self.sigma_prot_sq,
gamma_prot=self.gamma_prot,
centric=self.centric,
p_scale=self.x[0],
p_B_wilson=self.x[1])
g = flex.double( scaling.wilson_total_nll_gradient(
d_star_sq=self.d_star_sq,
f_obs=self.f_obs,
sigma_f_obs=self.sigma_f_obs,
epsilon=self.epsilon,
sigma_sq=self.sigma_prot_sq,
gamma_prot=self.gamma_prot,
centric=self.centric,
p_scale=self.x[0],
p_B_wilson=self.x[1]) )
self.f = f
return f, g
def finite_diffs_iso(p_scale=0.0, p_B_wilson=0.0, centric=False, h=0.0001):
d_star_sq = flex.double(10, 0.25)
f_obs = flex.double(10, 1.0)
centric_array = flex.bool(10, centric)
sigma_f_obs = f_obs / 10.0
sigma_sq = flex.double(10, 1.0)
epsilon = flex.double(10, 1.0)
gamma = flex.double(10, 0.0)
stmp1 = scaling.wilson_total_nll(d_star_sq=d_star_sq,
f_obs=f_obs,
sigma_f_obs=sigma_f_obs,
epsilon=epsilon,
sigma_sq=sigma_sq,
gamma_prot=gamma,
centric=centric_array,
p_scale=p_scale - h,
p_B_wilson=p_B_wilson)
stmp2 = scaling.wilson_total_nll(d_star_sq=d_star_sq,
f_obs=f_obs,
sigma_f_obs=sigma_f_obs,
epsilon=epsilon,
sigma_sq=sigma_sq,
gamma_prot=gamma,
centric=centric_array,
p_scale=p_scale + h,
p_B_wilson=p_B_wilson)
s_grad_diff = (stmp1 - stmp2) / (-2.0 * h)
btmp1 = scaling.wilson_total_nll(d_star_sq=d_star_sq,
f_obs=f_obs,
sigma_f_obs=sigma_f_obs,
epsilon=epsilon,
sigma_sq=sigma_sq,
gamma_prot=gamma,
centric=centric_array,
p_scale=p_scale,
p_B_wilson=p_B_wilson - h)
btmp2 = scaling.wilson_total_nll(d_star_sq=d_star_sq,
f_obs=f_obs,
sigma_f_obs=sigma_f_obs,
epsilon=epsilon,
sigma_sq=sigma_sq,
gamma_prot=gamma,
centric=centric_array,
p_scale=p_scale,
p_B_wilson=p_B_wilson + h)
b_grad_diff = (btmp1 - btmp2) / (-2.0 * h)
grad = scaling.wilson_total_nll_gradient(d_star_sq=d_star_sq,
f_obs=f_obs,
sigma_f_obs=sigma_f_obs,
epsilon=epsilon,
sigma_sq=sigma_sq,
gamma_prot=gamma,
centric=centric_array,
p_scale=p_scale,
p_B_wilson=p_B_wilson)
assert approx_equal(s_grad_diff, grad[0])
assert approx_equal(b_grad_diff, grad[1])
def test_likelihood_iso():
d_star_sq = flex.double(10, 0.250)
f_obs = flex.double(10, 1.0)
sigma_f_obs = flex.double(10, 0.0000)
sigma_sq = flex.double(10, 1.0)
epsilon = flex.double(10, 1.0)
gamma = flex.double(10, 0.0)
centric = flex.bool(10, True)
acentric = flex.bool(10, False)
p_scale = 0.0
p_B_wilson = 0.0
centric_single_trans = scaling.wilson_single_nll(
d_star_sq=d_star_sq[0],
f_obs=f_obs[0],
sigma_f_obs=sigma_f_obs[0],
epsilon=epsilon[0],
sigma_sq=sigma_sq[0],
gamma_prot=gamma[0],
centric=centric[0],
p_scale=p_scale,
p_B_wilson=p_B_wilson,
transform=True)
centric_single_no_trans = scaling.wilson_single_nll(
d_star_sq=d_star_sq[0],
f_obs=f_obs[0],
sigma_f_obs=sigma_f_obs[0],
epsilon=epsilon[0],
sigma_sq=sigma_sq[0],
gamma_prot=gamma[0],
centric=centric[0],
p_scale=1.0,
p_B_wilson=p_B_wilson,
transform=False)
assert approx_equal(centric_single_trans, 1.072364) ## from Mathematica
assert approx_equal(centric_single_trans, centric_single_no_trans)
acentric_single_trans = scaling.wilson_single_nll(
d_star_sq=d_star_sq[0],
f_obs=f_obs[0],
sigma_f_obs=sigma_f_obs[0],
epsilon=epsilon[0],
sigma_sq=sigma_sq[0],
gamma_prot=gamma[0],
centric=acentric[0],
p_scale=p_scale,
p_B_wilson=p_B_wilson)
acentric_single_no_trans = scaling.wilson_single_nll(
d_star_sq=d_star_sq[0],
f_obs=f_obs[0],
sigma_f_obs=sigma_f_obs[0],
epsilon=epsilon[0],
sigma_sq=sigma_sq[0],
gamma_prot=gamma[0],
centric=acentric[0],
p_scale=1.0,
p_B_wilson=p_B_wilson,
transform=False)
assert approx_equal(acentric_single_trans, 0.306853) ## from Mathematica
assert approx_equal(acentric_single_trans, acentric_single_no_trans)
centric_total = scaling.wilson_total_nll(d_star_sq=d_star_sq,
f_obs=f_obs,
sigma_f_obs=sigma_f_obs,
epsilon=epsilon,
sigma_sq=sigma_sq,
gamma_prot=gamma,
centric=centric,
p_scale=p_scale,
p_B_wilson=p_B_wilson)
acentric_total = scaling.wilson_total_nll(d_star_sq=d_star_sq,
f_obs=f_obs,
sigma_f_obs=sigma_f_obs,
epsilon=epsilon,
sigma_sq=sigma_sq,
gamma_prot=gamma,
centric=acentric,
p_scale=p_scale,
p_B_wilson=p_B_wilson)
assert approx_equal(centric_total, centric_single_trans * 10.0)
assert approx_equal(acentric_total, acentric_single_trans * 10.0)
def finite_diffs_iso(p_scale=0.0,p_B_wilson=0.0,centric=False,h=0.0001):
d_star_sq = flex.double(10,0.25)
f_obs = flex.double(10,1.0)
centric_array = flex.bool(10,centric)
sigma_f_obs = f_obs/10.0
sigma_sq = flex.double(10,1.0)
epsilon = flex.double(10,1.0)
gamma =flex.double(10,0.0)
stmp1 = scaling.wilson_total_nll(d_star_sq = d_star_sq,
f_obs = f_obs,
sigma_f_obs = sigma_f_obs,
epsilon = epsilon,
sigma_sq = sigma_sq,
gamma_prot = gamma,
centric = centric_array,
p_scale = p_scale-h,
p_B_wilson = p_B_wilson )
stmp2 = scaling.wilson_total_nll(d_star_sq = d_star_sq,
f_obs = f_obs,
sigma_f_obs = sigma_f_obs,
epsilon = epsilon,
sigma_sq = sigma_sq,
gamma_prot = gamma,
centric = centric_array,
p_scale = p_scale+h,
p_B_wilson = p_B_wilson)
s_grad_diff = (stmp1-stmp2)/(-2.0*h)
btmp1 = scaling.wilson_total_nll(d_star_sq = d_star_sq,
f_obs = f_obs,
sigma_f_obs = sigma_f_obs,
epsilon = epsilon,
sigma_sq = sigma_sq,
gamma_prot = gamma,
centric = centric_array,
p_scale = p_scale,
p_B_wilson = p_B_wilson-h)
btmp2 = scaling.wilson_total_nll(d_star_sq = d_star_sq,
f_obs = f_obs,
sigma_f_obs = sigma_f_obs,
epsilon = epsilon,
sigma_sq = sigma_sq,
gamma_prot = gamma,
centric = centric_array,
p_scale = p_scale,
p_B_wilson = p_B_wilson+h)
b_grad_diff = (btmp1-btmp2)/(-2.0*h)
grad = scaling.wilson_total_nll_gradient(d_star_sq = d_star_sq,
f_obs = f_obs,
sigma_f_obs = sigma_f_obs,
epsilon = epsilon,
sigma_sq = sigma_sq,
gamma_prot = gamma,
centric = centric_array,
p_scale = p_scale,
p_B_wilson = p_B_wilson)
assert approx_equal(s_grad_diff, grad[0])
assert approx_equal(b_grad_diff, grad[1])
def test_likelihood_iso():
d_star_sq = flex.double(10,0.250)
f_obs = flex.double(10,1.0)
sigma_f_obs = flex.double(10,0.0000)
sigma_sq = flex.double(10,1.0)
epsilon = flex.double(10,1.0)
gamma = flex.double(10,0.0)
centric = flex.bool(10,True)
acentric = flex.bool(10,False)
p_scale = 0.0
p_B_wilson = 0.0
centric_single_trans = scaling.wilson_single_nll(
d_star_sq = d_star_sq[0],
f_obs = f_obs[0],
sigma_f_obs = sigma_f_obs[0],
epsilon = epsilon[0],
sigma_sq = sigma_sq[0],
gamma_prot = gamma[0],
centric = centric[0],
p_scale = p_scale,
p_B_wilson = p_B_wilson,
transform = True)
centric_single_no_trans = scaling.wilson_single_nll(
d_star_sq = d_star_sq[0],
f_obs = f_obs[0],
sigma_f_obs = sigma_f_obs[0],
epsilon = epsilon[0],
sigma_sq = sigma_sq[0],
gamma_prot = gamma[0],
centric = centric[0],
p_scale = 1.0,
p_B_wilson = p_B_wilson,
transform = False)
assert approx_equal(centric_single_trans, 1.072364 ) ## from Mathematica
assert approx_equal(centric_single_trans, centric_single_no_trans)
acentric_single_trans = scaling.wilson_single_nll(
d_star_sq = d_star_sq[0],
f_obs = f_obs[0],
sigma_f_obs = sigma_f_obs[0],
epsilon = epsilon[0],
sigma_sq = sigma_sq[0],
gamma_prot = gamma[0],
centric = acentric[0],
p_scale = p_scale,
p_B_wilson = p_B_wilson)
acentric_single_no_trans = scaling.wilson_single_nll(
d_star_sq = d_star_sq[0],
f_obs = f_obs[0],
sigma_f_obs = sigma_f_obs[0],
epsilon = epsilon[0],
sigma_sq = sigma_sq[0],
gamma_prot = gamma[0],
centric = acentric[0],
p_scale = 1.0,
p_B_wilson =p_B_wilson,
transform = False)
assert approx_equal(acentric_single_trans, 0.306853) ## from Mathematica
assert approx_equal(acentric_single_trans, acentric_single_no_trans)
centric_total = scaling.wilson_total_nll(
d_star_sq = d_star_sq,
f_obs = f_obs,
sigma_f_obs = sigma_f_obs,
epsilon = epsilon,
sigma_sq = sigma_sq,
gamma_prot = gamma,
centric = centric,
p_scale = p_scale,
p_B_wilson = p_B_wilson)
acentric_total = scaling.wilson_total_nll(
d_star_sq = d_star_sq,
f_obs = f_obs,
sigma_f_obs = sigma_f_obs,
epsilon = epsilon,
sigma_sq = sigma_sq,
gamma_prot = gamma,
centric = acentric,
p_scale = p_scale,
p_B_wilson = p_B_wilson)
assert approx_equal(centric_total, centric_single_trans*10.0)
assert approx_equal(acentric_total, acentric_single_trans*10.0)