def ml_normalisation(self, aniso=False):
# estimate number of residues per unit cell
mr = matthews.matthews_rupp(self.intensities.crystal_symmetry())
n_residues = mr.n_residues
# estimate B-factor and scale factors for normalisation
if aniso:
normalisation = absolute_scaling.ml_aniso_absolute_scaling(
self.intensities, n_residues=n_residues)
u_star = normalisation.u_star
else:
normalisation = absolute_scaling.ml_iso_absolute_scaling(
self.intensities, n_residues=n_residues)
u_star = adptbx.b_as_u(
adptbx.u_iso_as_u_star(
self.intensities.unit_cell(), normalisation.b_wilson))
# apply scales
self.intensities = self.intensities.customized_copy(
data=scaling.ml_normalise_aniso(
self.intensities.indices(), self.intensities.data(),
normalisation.p_scale, self.intensities.unit_cell(),
u_star),
sigmas=scaling.ml_normalise_aniso(
self.intensities.indices(), self.intensities.sigmas(),
normalisation.p_scale, self.intensities.unit_cell(),
u_star)).set_info(self.intensities.info())
# record output in log file
s = StringIO()
mr.show(out=s)
normalisation.show(out=s)
logger.info(s.getvalue())
def anisotropic_correction(cache_0,
p_scale,
u_star,
b_add=None,
must_be_greater_than=0.):
## Make sure that u_star is not rwgk scaled, i.e. like you get it from
## the ml_absolute_scale_aniso routine (!which is !!NOT!! scaled!)
work_array = None
try:
work_array = cache_0.input.select( cache_0.input.data() > must_be_greater_than)
except KeyboardInterrupt: raise
except Exception: pass
if work_array is None:
work_array = cache_0.select( cache_0.data() > must_be_greater_than)
change_back_to_intensity=False
if work_array.is_xray_intensity_array():
work_array = work_array.f_sq_as_f()
change_back_to_intensity=True
assert not work_array.is_xray_intensity_array()
if b_add is not None:
u_star_add = adptbx.b_iso_as_u_star( work_array.unit_cell(),
b_add )
u_star = u_star+u_star_add
corrected_amplitudes = scaling.ml_normalise_aniso( work_array.indices(),
work_array.data(),
p_scale,
work_array.unit_cell(),
u_star )
if work_array.sigmas() is not None:
corrected_sigmas = scaling.ml_normalise_aniso( work_array.indices(),
work_array.sigmas(),
p_scale,
work_array.unit_cell(),
u_star )
else:
corrected_sigmas = None
work_array = work_array.customized_copy(
data = corrected_amplitudes,
sigmas = corrected_sigmas ).set_observation_type(work_array)
if change_back_to_intensity:
# XXX check for floating-point overflows (which trigger the Boost trap
# and crash the interpreter). The only known case is 2q8o:IOBS2,SIGIOBS2
# which is missing nearly all acentric hkls but it clearly points to a bug
# in this routine when dealing with pathological data.
f_max = flex.max(work_array.data())
if (not f_max < math.sqrt(sys.float_info.max)) :
raise OverflowError("Amplitudes will exceed floating point limit if "+
"converted to intensities (max F = %e)." % f_max)
work_array = work_array.f_as_f_sq()
return work_array
def anisotropic_correction(cache_0,
p_scale,
u_star,
b_add=None,
must_be_greater_than=0.):
## Make sure that u_star is not rwgk scaled, i.e. like you get it from
## the ml_absolute_scale_aniso routine (!which is !!NOT!! scaled!)
work_array = None
try:
work_array = cache_0.input.select(
cache_0.input.data() > must_be_greater_than)
except KeyboardInterrupt:
raise
except Exception:
pass
if work_array is None:
work_array = cache_0.select(cache_0.data() > must_be_greater_than)
change_back_to_intensity = False
if work_array.is_xray_intensity_array():
work_array = work_array.f_sq_as_f()
change_back_to_intensity = True
assert not work_array.is_xray_intensity_array()
if b_add is not None:
u_star_add = adptbx.b_iso_as_u_star(work_array.unit_cell(), b_add)
u_star = u_star + u_star_add
corrected_amplitudes = scaling.ml_normalise_aniso(work_array.indices(),
work_array.data(),
p_scale,
work_array.unit_cell(),
u_star)
if work_array.sigmas() is not None:
corrected_sigmas = scaling.ml_normalise_aniso(work_array.indices(),
work_array.sigmas(),
p_scale,
work_array.unit_cell(),
u_star)
else:
corrected_sigmas = None
work_array = work_array.customized_copy(
data=corrected_amplitudes,
sigmas=corrected_sigmas).set_observation_type(work_array)
if change_back_to_intensity:
# XXX check for floating-point overflows (which trigger the Boost trap
# and crash the interpreter). The only known case is 2q8o:IOBS2,SIGIOBS2
# which is missing nearly all acentric hkls but it clearly points to a bug
# in this routine when dealing with pathological data.
f_max = flex.max(work_array.data())
if (not f_max < math.sqrt(sys.float_info.max)):
raise OverflowError(
"Amplitudes will exceed floating point limit if " +
"converted to intensities (max F = %e)." % f_max)
work_array = work_array.f_as_f_sq()
return work_array
def _ml_normalisation(intensities, aniso):
# estimate number of residues per unit cell
mr = matthews.matthews_rupp(intensities.crystal_symmetry())
n_residues = mr.n_residues
# estimate B-factor and scale factors for normalisation
if aniso:
normalisation = absolute_scaling.ml_aniso_absolute_scaling(
intensities, n_residues=n_residues
)
u_star = normalisation.u_star
else:
normalisation = absolute_scaling.ml_iso_absolute_scaling(
intensities, n_residues=n_residues
)
u_star = adptbx.b_as_u(
adptbx.u_iso_as_u_star(intensities.unit_cell(), normalisation.b_wilson)
)
# record output in log file
if aniso:
b_cart = normalisation.b_cart
logger.info("ML estimate of overall B_cart value:")
logger.info(
"""\
%5.2f, %5.2f, %5.2f
%12.2f, %5.2f
%19.2f"""
% (b_cart[0], b_cart[3], b_cart[4], b_cart[1], b_cart[5], b_cart[2])
)
else:
logger.info("ML estimate of overall B value:")
logger.info(" %5.2f A**2" % normalisation.b_wilson)
logger.info("ML estimate of -log of scale factor:")
logger.info(" %5.2f" % (normalisation.p_scale))
s = StringIO()
mr.show(out=s)
normalisation.show(out=s)
logger.debug(s.getvalue())
# apply scales
return intensities.customized_copy(
data=scaling.ml_normalise_aniso(
intensities.indices(),
intensities.data(),
normalisation.p_scale,
intensities.unit_cell(),
u_star,
),
sigmas=scaling.ml_normalise_aniso(
intensities.indices(),
intensities.sigmas(),
normalisation.p_scale,
intensities.unit_cell(),
u_star,
),
)