Author: bugman
Date: Wed Sep 17 17:11:32 2014
New Revision: 25874
URL: http://svn.gna.org/viewcvs/relax?rev=25874&view=rev
Log:
Huge speed up for the generation of the Sobol' sequence data in the frame order
target function.
The new Sobol_data class has been created and is instantiated in the module
namespace as
target_function.frame_order.sobol_data. This is used to store all of the
Sobol' sequence associated
data, including the torsion-tilt angles and all corresponding rotation
matrices. When initialising
the target function, if the Sobol_data container holds the data for the same
model and same total
number of Sobol' points, then the pre-existing data will be used rather than
regenerating all the
data. This can save a huge amount of time.
Modified:
branches/frame_order_cleanup/target_functions/frame_order.py
Modified: branches/frame_order_cleanup/target_functions/frame_order.py
URL:
http://svn.gna.org/viewcvs/relax/branches/frame_order_cleanup/target_functions/frame_order.py?rev=25874&r1=25873&r2=25874&view=diff
==============================================================================
--- branches/frame_order_cleanup/target_functions/frame_order.py
(original)
+++ branches/frame_order_cleanup/target_functions/frame_order.py Wed Sep
17 17:11:32 2014
@@ -440,7 +440,7 @@
# PCS via numerical integration.
if self.pcs_flag:
# Numerical integration of the PCSs.
- pcs_numeric_int_double_rotor(points=self.sobol_angles,
max_points=self.sobol_max_points, sigma_max=sigma_max, sigma_max_2=sigma_max_2,
c=self.pcs_const, full_in_ref_frame=self.full_in_ref_frame,
r_pivot_atom=self.r_pivot_atom, r_pivot_atom_rev=self.r_pivot_atom_rev,
r_ln_pivot=self.r_ln_pivot, r_inter_pivot=self.r_inter_pivot, A=self.A_3D,
R_eigen=self.R_eigen, RT_eigen=RT_eigen, Ri_prime=self.Ri_prime,
Ri2_prime=self.Ri2_prime, pcs_theta=self.pcs_theta,
pcs_theta_err=self.pcs_theta_err, missing_pcs=self.missing_pcs)
+ pcs_numeric_int_double_rotor(points=sobol_data.sobol_angles,
max_points=self.sobol_max_points, sigma_max=sigma_max, sigma_max_2=sigma_max_2,
c=self.pcs_const, full_in_ref_frame=self.full_in_ref_frame,
r_pivot_atom=self.r_pivot_atom, r_pivot_atom_rev=self.r_pivot_atom_rev,
r_ln_pivot=self.r_ln_pivot, r_inter_pivot=self.r_inter_pivot, A=self.A_3D,
R_eigen=self.R_eigen, RT_eigen=RT_eigen, Ri_prime=sobol_data.Ri_prime,
Ri2_prime=sobol_data.Ri2_prime, pcs_theta=self.pcs_theta,
pcs_theta_err=self.pcs_theta_err, missing_pcs=self.missing_pcs)
# Calculate and sum the single alignment chi-squared value (for
the PCS).
for align_index in range(self.num_align):
@@ -518,7 +518,7 @@
# PCS via numerical integration.
if self.pcs_flag:
# Numerical integration of the PCSs.
- pcs_numeric_int_rotor_qrint(points=self.sobol_angles,
max_points=self.sobol_max_points, sigma_max=pi, c=self.pcs_const,
full_in_ref_frame=self.full_in_ref_frame, r_pivot_atom=self.r_pivot_atom,
r_pivot_atom_rev=self.r_pivot_atom_rev, r_ln_pivot=self.r_ln_pivot,
A=self.A_3D, R_eigen=self.R_eigen, RT_eigen=RT_eigen, Ri_prime=self.Ri_prime,
pcs_theta=self.pcs_theta, pcs_theta_err=self.pcs_theta_err,
missing_pcs=self.missing_pcs)
+ pcs_numeric_int_rotor_qrint(points=sobol_data.sobol_angles,
max_points=self.sobol_max_points, sigma_max=pi, c=self.pcs_const,
full_in_ref_frame=self.full_in_ref_frame, r_pivot_atom=self.r_pivot_atom,
r_pivot_atom_rev=self.r_pivot_atom_rev, r_ln_pivot=self.r_ln_pivot,
A=self.A_3D, R_eigen=self.R_eigen, RT_eigen=RT_eigen,
Ri_prime=sobol_data.Ri_prime, pcs_theta=self.pcs_theta,
pcs_theta_err=self.pcs_theta_err, missing_pcs=self.missing_pcs)
# Calculate and sum the single alignment chi-squared value (for
the PCS).
for align_index in range(self.num_align):
@@ -597,7 +597,7 @@
# PCS via numerical integration.
if self.pcs_flag:
# Numerical integration of the PCSs.
- pcs_numeric_int_iso_cone_qrint(points=self.sobol_angles,
max_points=self.sobol_max_points, theta_max=cone_theta, sigma_max=sigma_max,
c=self.pcs_const, full_in_ref_frame=self.full_in_ref_frame,
r_pivot_atom=self.r_pivot_atom, r_pivot_atom_rev=self.r_pivot_atom_rev,
r_ln_pivot=self.r_ln_pivot, A=self.A_3D, R_eigen=self.R_eigen,
RT_eigen=RT_eigen, Ri_prime=self.Ri_prime, pcs_theta=self.pcs_theta,
pcs_theta_err=self.pcs_theta_err, missing_pcs=self.missing_pcs)
+ pcs_numeric_int_iso_cone_qrint(points=sobol_data.sobol_angles,
max_points=self.sobol_max_points, theta_max=cone_theta, sigma_max=sigma_max,
c=self.pcs_const, full_in_ref_frame=self.full_in_ref_frame,
r_pivot_atom=self.r_pivot_atom, r_pivot_atom_rev=self.r_pivot_atom_rev,
r_ln_pivot=self.r_ln_pivot, A=self.A_3D, R_eigen=self.R_eigen,
RT_eigen=RT_eigen, Ri_prime=sobol_data.Ri_prime, pcs_theta=self.pcs_theta,
pcs_theta_err=self.pcs_theta_err, missing_pcs=self.missing_pcs)
# Calculate and sum the single alignment chi-squared value (for
the PCS).
for align_index in range(self.num_align):
@@ -678,7 +678,7 @@
# PCS via numerical integration.
if self.pcs_flag:
# Numerical integration of the PCSs.
- pcs_numeric_int_iso_cone_qrint(points=self.sobol_angles,
max_points=self.sobol_max_points, theta_max=theta_max, sigma_max=pi,
c=self.pcs_const, full_in_ref_frame=self.full_in_ref_frame,
r_pivot_atom=self.r_pivot_atom, r_pivot_atom_rev=self.r_pivot_atom_rev,
r_ln_pivot=self.r_ln_pivot, A=self.A_3D, R_eigen=self.R_eigen,
RT_eigen=RT_eigen, Ri_prime=self.Ri_prime, pcs_theta=self.pcs_theta,
pcs_theta_err=self.pcs_theta_err, missing_pcs=self.missing_pcs)
+ pcs_numeric_int_iso_cone_qrint(points=sobol_data.sobol_angles,
max_points=self.sobol_max_points, theta_max=theta_max, sigma_max=pi,
c=self.pcs_const, full_in_ref_frame=self.full_in_ref_frame,
r_pivot_atom=self.r_pivot_atom, r_pivot_atom_rev=self.r_pivot_atom_rev,
r_ln_pivot=self.r_ln_pivot, A=self.A_3D, R_eigen=self.R_eigen,
RT_eigen=RT_eigen, Ri_prime=sobol_data.Ri_prime, pcs_theta=self.pcs_theta,
pcs_theta_err=self.pcs_theta_err, missing_pcs=self.missing_pcs)
# Calculate and sum the single alignment chi-squared value (for
the PCS).
for align_index in range(self.num_align):
@@ -756,7 +756,7 @@
# PCS via numerical integration.
if self.pcs_flag:
# Numerical integration of the PCSs.
-
pcs_numeric_int_iso_cone_torsionless_qrint(points=self.sobol_angles,
max_points=self.sobol_max_points, theta_max=cone_theta, c=self.pcs_const,
full_in_ref_frame=self.full_in_ref_frame, r_pivot_atom=self.r_pivot_atom,
r_pivot_atom_rev=self.r_pivot_atom_rev, r_ln_pivot=self.r_ln_pivot,
A=self.A_3D, R_eigen=self.R_eigen, RT_eigen=RT_eigen, Ri_prime=self.Ri_prime,
pcs_theta=self.pcs_theta, pcs_theta_err=self.pcs_theta_err,
missing_pcs=self.missing_pcs)
+
pcs_numeric_int_iso_cone_torsionless_qrint(points=sobol_data.sobol_angles,
max_points=self.sobol_max_points, theta_max=cone_theta, c=self.pcs_const,
full_in_ref_frame=self.full_in_ref_frame, r_pivot_atom=self.r_pivot_atom,
r_pivot_atom_rev=self.r_pivot_atom_rev, r_ln_pivot=self.r_ln_pivot,
A=self.A_3D, R_eigen=self.R_eigen, RT_eigen=RT_eigen,
Ri_prime=sobol_data.Ri_prime, pcs_theta=self.pcs_theta,
pcs_theta_err=self.pcs_theta_err, missing_pcs=self.missing_pcs)
# Calculate and sum the single alignment chi-squared value (for
the PCS).
for align_index in range(self.num_align):
@@ -831,7 +831,7 @@
# PCS via numerical integration.
if self.pcs_flag:
# Numerical integration of the PCSs.
- pcs_numeric_int_pseudo_ellipse_qrint(points=self.sobol_angles,
max_points=self.sobol_max_points, theta_x=cone_theta_x, theta_y=cone_theta_y,
sigma_max=cone_sigma_max, c=self.pcs_const,
full_in_ref_frame=self.full_in_ref_frame, r_pivot_atom=self.r_pivot_atom,
r_pivot_atom_rev=self.r_pivot_atom_rev, r_ln_pivot=self.r_ln_pivot,
A=self.A_3D, R_eigen=self.R_eigen, RT_eigen=RT_eigen, Ri_prime=self.Ri_prime,
pcs_theta=self.pcs_theta, pcs_theta_err=self.pcs_theta_err,
missing_pcs=self.missing_pcs)
+
pcs_numeric_int_pseudo_ellipse_qrint(points=sobol_data.sobol_angles,
max_points=self.sobol_max_points, theta_x=cone_theta_x, theta_y=cone_theta_y,
sigma_max=cone_sigma_max, c=self.pcs_const,
full_in_ref_frame=self.full_in_ref_frame, r_pivot_atom=self.r_pivot_atom,
r_pivot_atom_rev=self.r_pivot_atom_rev, r_ln_pivot=self.r_ln_pivot,
A=self.A_3D, R_eigen=self.R_eigen, RT_eigen=RT_eigen,
Ri_prime=sobol_data.Ri_prime, pcs_theta=self.pcs_theta,
pcs_theta_err=self.pcs_theta_err, missing_pcs=self.missing_pcs)
# Calculate and sum the single alignment chi-squared value (for
the PCS).
for align_index in range(self.num_align):
@@ -906,7 +906,7 @@
# PCS via numerical integration.
if self.pcs_flag:
# Numerical integration of the PCSs.
- pcs_numeric_int_pseudo_ellipse_qrint(points=self.sobol_angles,
max_points=self.sobol_max_points, theta_x=cone_theta_x, theta_y=cone_theta_y,
sigma_max=pi, c=self.pcs_const, full_in_ref_frame=self.full_in_ref_frame,
r_pivot_atom=self.r_pivot_atom, r_pivot_atom_rev=self.r_pivot_atom_rev,
r_ln_pivot=self.r_ln_pivot, A=self.A_3D, R_eigen=self.R_eigen,
RT_eigen=RT_eigen, Ri_prime=self.Ri_prime, pcs_theta=self.pcs_theta,
pcs_theta_err=self.pcs_theta_err, missing_pcs=self.missing_pcs)
+
pcs_numeric_int_pseudo_ellipse_qrint(points=sobol_data.sobol_angles,
max_points=self.sobol_max_points, theta_x=cone_theta_x, theta_y=cone_theta_y,
sigma_max=pi, c=self.pcs_const, full_in_ref_frame=self.full_in_ref_frame,
r_pivot_atom=self.r_pivot_atom, r_pivot_atom_rev=self.r_pivot_atom_rev,
r_ln_pivot=self.r_ln_pivot, A=self.A_3D, R_eigen=self.R_eigen,
RT_eigen=RT_eigen, Ri_prime=sobol_data.Ri_prime, pcs_theta=self.pcs_theta,
pcs_theta_err=self.pcs_theta_err, missing_pcs=self.missing_pcs)
# Calculate and sum the single alignment chi-squared value (for
the PCS).
for align_index in range(self.num_align):
@@ -981,7 +981,7 @@
# PCS via numerical integration.
if self.pcs_flag:
# Numerical integration of the PCSs.
-
pcs_numeric_int_pseudo_ellipse_torsionless_qrint(points=self.sobol_angles,
max_points=self.sobol_max_points, theta_x=cone_theta_x, theta_y=cone_theta_y,
c=self.pcs_const, full_in_ref_frame=self.full_in_ref_frame,
r_pivot_atom=self.r_pivot_atom, r_pivot_atom_rev=self.r_pivot_atom_rev,
r_ln_pivot=self.r_ln_pivot, A=self.A_3D, R_eigen=self.R_eigen,
RT_eigen=RT_eigen, Ri_prime=self.Ri_prime, pcs_theta=self.pcs_theta,
pcs_theta_err=self.pcs_theta_err, missing_pcs=self.missing_pcs)
+
pcs_numeric_int_pseudo_ellipse_torsionless_qrint(points=sobol_data.sobol_angles,
max_points=self.sobol_max_points, theta_x=cone_theta_x, theta_y=cone_theta_y,
c=self.pcs_const, full_in_ref_frame=self.full_in_ref_frame,
r_pivot_atom=self.r_pivot_atom, r_pivot_atom_rev=self.r_pivot_atom_rev,
r_ln_pivot=self.r_ln_pivot, A=self.A_3D, R_eigen=self.R_eigen,
RT_eigen=RT_eigen, Ri_prime=sobol_data.Ri_prime, pcs_theta=self.pcs_theta,
pcs_theta_err=self.pcs_theta_err, missing_pcs=self.missing_pcs)
# Calculate and sum the single alignment chi-squared value (for
the PCS).
for align_index in range(self.num_align):
@@ -1138,7 +1138,7 @@
# PCS via numerical integration.
if self.pcs_flag:
# Numerical integration of the PCSs.
- pcs_numeric_int_rotor_qrint(points=self.sobol_angles,
max_points=self.sobol_max_points, sigma_max=sigma_max, c=self.pcs_const,
full_in_ref_frame=self.full_in_ref_frame, r_pivot_atom=self.r_pivot_atom,
r_pivot_atom_rev=self.r_pivot_atom_rev, r_ln_pivot=self.r_ln_pivot,
A=self.A_3D, R_eigen=self.R_eigen, RT_eigen=RT_eigen, Ri_prime=self.Ri_prime,
pcs_theta=self.pcs_theta, pcs_theta_err=self.pcs_theta_err,
missing_pcs=self.missing_pcs)
+ pcs_numeric_int_rotor_qrint(points=sobol_data.sobol_angles,
max_points=self.sobol_max_points, sigma_max=sigma_max, c=self.pcs_const,
full_in_ref_frame=self.full_in_ref_frame, r_pivot_atom=self.r_pivot_atom,
r_pivot_atom_rev=self.r_pivot_atom_rev, r_ln_pivot=self.r_ln_pivot,
A=self.A_3D, R_eigen=self.R_eigen, RT_eigen=RT_eigen,
Ri_prime=sobol_data.Ri_prime, pcs_theta=self.pcs_theta,
pcs_theta_err=self.pcs_theta_err, missing_pcs=self.missing_pcs)
# Calculate and sum the single alignment chi-squared value (for
the PCS).
for align_index in range(self.num_align):
@@ -1198,19 +1198,25 @@
@type dims: list of str
"""
+ # The number of dimensions.
+ m = len(dims)
+
+ # The total number of points.
+ total_num = int(self.sobol_max_points * self.sobol_oversample * 10**m)
+
+ # Reuse pre-created data if available.
+ if total_num == sobol_data.total_num and self.model ==
sobol_data.model:
+ return
+
# Printout (useful to see how long this takes!).
print("Generating the torsion-tilt angle sampling via the Sobol'
sequence for numerical PCS integration.")
- # The number of dimensions.
- m = len(dims)
-
- # The total number of points.
- total_num = int(self.sobol_max_points * self.sobol_oversample * 10**m)
-
# Initialise.
- self.sobol_angles = zeros((m, total_num), float64)
- self.Ri_prime = zeros((total_num, 3, 3), float64)
- self.Ri2_prime = zeros((total_num, 3, 3), float64)
+ sobol_data.model = self.model
+ sobol_data.total_num = total_num
+ sobol_data.sobol_angles = zeros((m, total_num), float64)
+ sobol_data.Ri_prime = zeros((total_num, 3, 3), float64)
+ sobol_data.Ri2_prime = zeros((total_num, 3, 3), float64)
# The Sobol' points.
points = i4_sobol_generate(m, total_num, 1000)
@@ -1225,46 +1231,46 @@
# The tilt angle - the angle of rotation about the x-y plane
rotation axis.
if dims[j] in ['theta']:
theta = acos(2.0*points[j, i] - 1.0)
- self.sobol_angles[j, i] = theta
+ sobol_data.sobol_angles[j, i] = theta
# The angle defining the x-y plane rotation axis.
if dims[j] in ['phi']:
phi = 2.0 * pi * points[j, i]
- self.sobol_angles[j, i] = phi
+ sobol_data.sobol_angles[j, i] = phi
# The 1st torsion angle - the angle of rotation about the z'
axis (or y' for the double motion models).
if dims[j] in ['sigma']:
sigma = 2.0 * pi * (points[j, i] - 0.5)
- self.sobol_angles[j, i] = sigma
+ sobol_data.sobol_angles[j, i] = sigma
# The 2nd torsion angle - the angle of rotation about the x'
axis.
if dims[j] in ['sigma2']:
sigma2 = 2.0 * pi * (points[j, i] - 0.5)
- self.sobol_angles[j, i] = sigma2
+ sobol_data.sobol_angles[j, i] = sigma2
# Pre-calculate the rotation matrices for the double motion models.
if 'sigma2' in dims:
# The 1st rotation about the y-axis.
c_sigma = cos(sigma)
s_sigma = sin(sigma)
- self.Ri_prime[i, 0, 0] = c_sigma
- self.Ri_prime[i, 0, 2] = s_sigma
- self.Ri_prime[i, 1, 1] = 1.0
- self.Ri_prime[i, 2, 0] = -s_sigma
- self.Ri_prime[i, 2, 2] = c_sigma
+ sobol_data.Ri_prime[i, 0, 0] = c_sigma
+ sobol_data.Ri_prime[i, 0, 2] = s_sigma
+ sobol_data.Ri_prime[i, 1, 1] = 1.0
+ sobol_data.Ri_prime[i, 2, 0] = -s_sigma
+ sobol_data.Ri_prime[i, 2, 2] = c_sigma
# The 2nd rotation about the x-axis.
c_sigma2 = cos(sigma2)
s_sigma2 = sin(sigma2)
- self.Ri2_prime[i, 0, 0] = 1.0
- self.Ri2_prime[i, 1, 1] = c_sigma2
- self.Ri2_prime[i, 1, 2] = -s_sigma2
- self.Ri2_prime[i, 2, 1] = s_sigma2
- self.Ri2_prime[i, 2, 2] = c_sigma2
+ sobol_data.Ri2_prime[i, 0, 0] = 1.0
+ sobol_data.Ri2_prime[i, 1, 1] = c_sigma2
+ sobol_data.Ri2_prime[i, 1, 2] = -s_sigma2
+ sobol_data.Ri2_prime[i, 2, 1] = s_sigma2
+ sobol_data.Ri2_prime[i, 2, 2] = c_sigma2
# Pre-calculate the rotation matrix for the full tilt-torsion.
elif theta != None and phi != None and sigma != None:
- tilt_torsion_to_R(phi, theta, sigma, self.Ri_prime[i])
+ tilt_torsion_to_R(phi, theta, sigma, sobol_data.Ri_prime[i])
# Pre-calculate the rotation matrix for the torsionless models.
elif sigma == None:
@@ -1274,25 +1280,25 @@
s_phi = sin(phi)
c_phi_c_theta = c_phi * c_theta
s_phi_c_theta = s_phi * c_theta
- self.Ri_prime[i, 0, 0] = c_phi_c_theta*c_phi + s_phi**2
- self.Ri_prime[i, 0, 1] = c_phi_c_theta*s_phi - c_phi*s_phi
- self.Ri_prime[i, 0, 2] = c_phi*s_theta
- self.Ri_prime[i, 1, 0] = s_phi_c_theta*c_phi - c_phi*s_phi
- self.Ri_prime[i, 1, 1] = s_phi_c_theta*s_phi + c_phi**2
- self.Ri_prime[i, 1, 2] = s_phi*s_theta
- self.Ri_prime[i, 2, 0] = -s_theta*c_phi
- self.Ri_prime[i, 2, 1] = -s_theta*s_phi
- self.Ri_prime[i, 2, 2] = c_theta
+ sobol_data.Ri_prime[i, 0, 0] = c_phi_c_theta*c_phi + s_phi**2
+ sobol_data.Ri_prime[i, 0, 1] = c_phi_c_theta*s_phi -
c_phi*s_phi
+ sobol_data.Ri_prime[i, 0, 2] = c_phi*s_theta
+ sobol_data.Ri_prime[i, 1, 0] = s_phi_c_theta*c_phi -
c_phi*s_phi
+ sobol_data.Ri_prime[i, 1, 1] = s_phi_c_theta*s_phi + c_phi**2
+ sobol_data.Ri_prime[i, 1, 2] = s_phi*s_theta
+ sobol_data.Ri_prime[i, 2, 0] = -s_theta*c_phi
+ sobol_data.Ri_prime[i, 2, 1] = -s_theta*s_phi
+ sobol_data.Ri_prime[i, 2, 2] = c_theta
# Pre-calculate the rotation matrix for the rotor models.
else:
c_sigma = cos(sigma)
s_sigma = sin(sigma)
- self.Ri_prime[i, 0, 0] = c_sigma
- self.Ri_prime[i, 0, 1] = -s_sigma
- self.Ri_prime[i, 1, 0] = s_sigma
- self.Ri_prime[i, 1, 1] = c_sigma
- self.Ri_prime[i, 2, 2] = 1.0
+ sobol_data.Ri_prime[i, 0, 0] = c_sigma
+ sobol_data.Ri_prime[i, 0, 1] = -s_sigma
+ sobol_data.Ri_prime[i, 1, 0] = s_sigma
+ sobol_data.Ri_prime[i, 1, 1] = c_sigma
+ sobol_data.Ri_prime[i, 2, 2] = 1.0
# Printout (useful to see how long this takes!).
print(" Oversampled to %s points." % total_num)
@@ -1343,3 +1349,22 @@
# Convert the tensor back to 5D, rank-1 form, as the
back-calculated reduced tensor.
to_5D(self.A_5D_bc[index1:index2], self.A_3D_bc[align_index])
+
+
+
+class Sobol_data:
+ """Temporary storage of the Sobol' data for speed."""
+
+ def __init__(self):
+ """Set up the object."""
+
+ # Initialise some variables.
+ self.model = None
+ self.Ri_prime = None
+ self.Ri2_prime = None
+ self.sobol_angles = None
+ self.total_num = None
+
+
+# Instantiate the Sobol' data container.
+sobol_data = Sobol_data()
_______________________________________________
relax (http://www.nmr-relax.com)
This is the relax-commits mailing list
[email protected]
To unsubscribe from this list, get a password
reminder, or change your subscription options,
visit the list information page at
https://mail.gna.org/listinfo/relax-commits
