Author: bugman
Date: Tue Sep 22 18:18:55 2009
New Revision: 9562
URL: http://svn.gna.org/viewcvs/relax?rev=9562&view=rev
Log:
Redesigned the model-free grid search setup around the new minfx interface.
Modified:
1.3/specific_fns/model_free/mf_minimise.py
Modified: 1.3/specific_fns/model_free/mf_minimise.py
URL:
http://svn.gna.org/viewcvs/relax/1.3/specific_fns/model_free/mf_minimise.py?rev=9562&r1=9561&r2=9562&view=diff
==============================================================================
--- 1.3/specific_fns/model_free/mf_minimise.py (original)
+++ 1.3/specific_fns/model_free/mf_minimise.py Tue Sep 22 18:18:55 2009
@@ -22,9 +22,13 @@
# Python module imports.
from math import pi
+from minfx.generic import generic_minimise
+from minfx.grid import grid
from numpy import float64, array, dot, zeros
from numpy.linalg import inv
-from re import match
+from re import match, search
+import sys
+from warnings import warn
# relax module imports.
from float import isNaN, isInf
@@ -32,9 +36,9 @@
from generic_fns.diffusion_tensor import diff_data_exists
from generic_fns.mol_res_spin import count_spins, exists_mol_res_spin_data,
return_spin_from_index, spin_loop
from maths_fns.mf import Mf
-from minfx.generic import generic_minimise
from physical_constants import h_bar, mu0, return_gyromagnetic_ratio
from relax_errors import RelaxError, RelaxInfError, RelaxLenError,
RelaxNaNError, RelaxNoModelError, RelaxNoPdbError, RelaxNoResError,
RelaxNoSequenceError, RelaxNoTensorError, RelaxNoValueError,
RelaxNoVectorsError, RelaxNucleusError, RelaxProtonTypeError,
RelaxSpinTypeError
+from relax_warnings import RelaxWarning
@@ -504,189 +508,165 @@
# If inc is a single int, convert it into an array of that value.
if isinstance(inc, int):
- temp = []
- for i in xrange(num_params):
- temp.append(inc)
- inc = temp
-
- # Minimisation options initialisation.
- min_options = []
- m = 0
-
- # Determine the model type.
- model_type = self.determine_model_type()
-
- # Minimisation options for diffusion tensor parameters.
- if model_type == 'diff' or model_type == 'all':
- # Get the diffusion tensor specific configuration.
- m = self.grid_search_config_diff(min_options, inc, m)
-
- # Model-free parameters (residue specific parameters).
- if model_type != 'diff':
- # The loop.
- if spin:
- loop = [spin]
- else:
- loop = spin_loop(spin_id)
-
- # Loop over the spins.
- for spin in loop:
- # Skip deselected residues.
- if not spin.select:
- continue
-
- # Get the spin specific configuration.
- m = self.grid_search_config_spin(min_options, spin, inc, m)
-
- # Test if the grid is too large.
- self.test_grid_size(min_options, verbosity=verbosity)
-
- # Complete the grid search configuration.
- self.grid_search_config_fin(min_options, lower, upper,
scaling_matrix)
+ inc = [inc]*num_params
+
+ # Set up the default bounds.
+ if not lower:
+ # Init.
+ lower = []
+ upper = []
+
+ # Determine the model type.
+ model_type = self.determine_model_type()
+
+ # Minimisation options for diffusion tensor parameters.
+ if model_type == 'diff' or model_type == 'all':
+ # Get the diffusion tensor specific configuration.
+ self.grid_search_diff_bounds(lower, upper)
+
+ # Model-free parameters (residue specific parameters).
+ if model_type != 'diff':
+ # The loop.
+ if spin:
+ loop = [spin]
+ else:
+ loop = spin_loop(spin_id)
+
+ # Loop over the spins.
+ for spin in loop:
+ # Skip deselected residues.
+ if not spin.select:
+ continue
+
+ # Get the spin specific configuration.
+ self.grid_search_spin_bounds(spin, lower, upper)
+
+ # Diagonal scaling of minimisation options.
+ for i in xrange(num_params):
+ lower[i] = lower[i] / scaling_matrix[i, i]
+ upper[i] = upper[i] / scaling_matrix[i, i]
# Return the minimisation options.
- return min_options
-
-
- def grid_search_config_diff(self, min_options, inc, m):
- """Set up of the grid search for the diffusion tensor.
-
- This method appends the grid search configuration details to
min_options list. These
- details are in the form of a list consisting of the number of
increments, lower bound, and
- upper bound for the corresponding residue.
-
- @param min_options: An array to append the grid search configuration
details to.
- @type min_options: list
- @param inc: The increments for each dimension of the space
for the grid search. The
- number of elements in the array must equal to
the number of parameters
- in the model.
- @type inc: array of int
- @param m: The parameter index for the complete model.
- @type m: int
- @return: The index of the last parameter encountered (m).
- @rtype: int
+ return inc, lower, upper
+
+
+ def grid_search_diff_bounds(self, lower, upper):
+ """Set up the default grid search bounds the diffusion tensor.
+
+ This method appends the default bounds to the lower and upper lists.
+
+ @param lower: The lower bound list to append to.
+ @type lower: list
+ @param upper: The upper bound list to append to.
+ @type upper: list
"""
# Spherical diffusion {tm}.
if cdp.diff_tensor.type == 'sphere':
- min_options.append([inc[0], 1.0 * 1e-9, 12.0 * 1e-9])
- m = m + 1
+ lower.append(1.0 * 1e-9)
+ upper.append(12.0 * 1e-9)
# Spheroidal diffusion {tm, Da, theta, phi}.
if cdp.diff_tensor.type == 'spheroid':
- min_options.append([inc[0], 1.0 * 1e-9, 12.0 * 1e-9])
+ # tm.
+ lower.append(1.0 * 1e-9)
+ upper.append(12.0 * 1e-9)
+
+ # Da.
if cdp.diff_tensor.spheroid_type == 'prolate':
- min_options.append([inc[1], 0.0, 1e7])
+ lower.append(0.0)
+ upper.append(1e7)
elif cdp.diff_tensor.spheroid_type == 'oblate':
- min_options.append([inc[1], -1e7, 0.0])
- else:
- min_options.append([inc[1], -1e7, 1e7])
- min_options.append([inc[2], 0.0, pi])
- min_options.append([inc[3], 0.0, pi])
- m = m + 4
+ lower.append(-1e7)
+ upper.append(0.0)
+ else:
+ lower.append(-1e7)
+ upper.append(1e7)
+
+ # theta.
+ lower.append(0.0)
+ upper.append(pi)
+
+ # phi.
+ lower.append(0.0)
+ upper.append(pi)
# Ellipsoidal diffusion {tm, Da, Dr, alpha, beta, gamma}.
elif cdp.diff_tensor.type == 'ellipsoid':
- min_options.append([inc[0], 1.0 * 1e-9, 12.0 * 1e-9])
- min_options.append([inc[1], 0.0, 1e7])
- min_options.append([inc[2], 0.0, 1.0])
- min_options.append([inc[3], 0.0, pi])
- min_options.append([inc[4], 0.0, pi])
- min_options.append([inc[5], 0.0, pi])
- m = m + 6
-
- # Return the parameter index.
- return m
-
-
- def grid_search_config_fin(self, min_options, lower, upper,
scaling_matrix):
- """Complete the grid search configuration.
-
- @param min_options: The grid search configuration details.
- @type min_options: list of lists (n, 3)
- @param lower: The lower bounds of the grid search which
must be equal to the
- number of parameters in the model.
- @type lower: array of numbers
- @param upper: The upper bounds of the grid search which
must be equal to the
- number of parameters in the model.
- @type upper: array of numbers
- @param scaling_matrix: The scaling matrix.
- @type scaling_matrix: numpy matrix
- """
-
- # Set the lower and upper bounds if these are supplied.
- if lower != None:
- for i in xrange(n):
- if lower[i] != None:
- min_options[i][1] = lower[i]
- if upper != None:
- for i in xrange(n):
- if upper[i] != None:
- min_options[i][2] = upper[i]
-
- # Diagonal scaling of minimisation options.
- for i in xrange(len(min_options)):
- min_options[i][1] = min_options[i][1] / scaling_matrix[i, i]
- min_options[i][2] = min_options[i][2] / scaling_matrix[i, i]
-
-
- def grid_search_config_spin(self, min_options, spin, inc, m):
- """Set up of the grid search for a single spin.
-
- This method appends the grid search configuration details to
min_options list. These
- details are in the form of a list consisting of the number of
increments, lower bound, and
- upper bound for the corresponding residue. The ordering of the
lists in min_options matches
- that of the params list in the spin container.
-
- @param min_options: An array to append the grid search configuration
details to.
- @type min_options: list
+ # tm.
+ lower.append(1.0 * 1e-9)
+ upper.append(12.0 * 1e-9)
+
+ # Da.
+ lower.append(0.0)
+ upper.append(1e7)
+
+ # Dr.
+ lower.append(0.0)
+ upper.append(1.0)
+
+ # alpha.
+ lower.append(0.0)
+ upper.append(pi)
+
+ # beta.
+ lower.append(0.0)
+ upper.append(pi)
+
+ # gamma.
+ lower.append(0.0)
+ upper.append(pi)
+
+
+ def grid_search_spin_bounds(self, spin, lower, upper):
+ """Set up the default grid search bounds for a single spin.
+
+ This method appends the default bounds to the lower and upper lists.
The ordering of the
+ lists in min_options matches that of the params list in the spin
container.
+
@param spin: A SpinContainer object.
@type spin: class instance
- @param inc: The increments for each dimension of the space
for the grid search. The
- number of elements in the array must equal to
the number of parameters
- in the model.
- @type inc: array of int
- @param m: The parameter index for the complete model.
- @type m: int
- @return: The index of the last parameter encountered (m).
- @rtype: int
+ @param lower: The lower bound list to append to.
+ @type lower: list
+ @param upper: The upper bound list to append to.
+ @type upper: list
"""
# Loop over the model-free parameters.
for i in xrange(len(spin.params)):
# Local tm.
if spin.params[i] == 'local_tm':
- min_options.append([inc[m], 1.0 * 1e-9, 12.0 * 1e-9])
+ lower.append(1.0 * 1e-9)
+ upper.append(12.0 * 1e-9)
# {S2, S2f, S2s}.
elif match('S2', spin.params[i]):
- min_options.append([inc[m], 0.0, 1.0])
+ lower.append(0.0)
+ upper.append(1.0)
# {te, tf, ts}.
elif match('t', spin.params[i]):
- min_options.append([inc[m], 0.0, 500.0 * 1e-12])
+ lower.append(0.0)
+ upper.append(500.0 * 1e-12)
# Rex.
elif spin.params[i] == 'Rex':
- min_options.append([inc[m], 0.0, 5.0 / (2.0 * pi *
spin.frq[0])**2])
+ lower.append(0.0)
+ upper.append(5.0 / (2.0 * pi * spin.frq[0])**2)
# Bond length.
elif spin.params[i] == 'r':
- min_options.append([inc[m], 1.0 * 1e-10, 1.05 * 1e-10])
+ lower.append(1.0 * 1e-10)
+ upper.append(1.05 * 1e-10)
# CSA.
elif spin.params[i] == 'CSA':
- min_options.append([inc[m], -120 * 1e-6, -200 * 1e-6])
+ lower.append(-120 * 1e-6)
+ upper.append(-200 * 1e-6)
# Unknown option.
else:
raise RelaxError("Unknown model-free parameter.")
-
- # Increment m.
- m = m + 1
-
- # Return the parameter index.
- return m
def minimise(self, min_algor=None, min_options=None, func_tol=None,
grad_tol=None, max_iterations=None, constraints=False, scaling=True,
verbosity=0, sim_index=None, lower=None, upper=None, inc=None):
@@ -904,7 +884,7 @@
# Configure the grid search.
if match('^[Gg]rid', min_algor):
- min_options = self.grid_search_config(num_params, spin=spin,
lower=lower, upper=upper, inc=inc, scaling_matrix=scaling_matrix)
+ inc, lower, upper = self.grid_search_config(num_params,
spin=spin, lower=lower, upper=upper, inc=inc, scaling_matrix=scaling_matrix)
# Scaling of values for the set function.
if match('^[Ss]et', min_algor):
@@ -913,6 +893,8 @@
# Linear constraints.
if constraints:
A, b = self.linear_constraints(num_params,
model_type=model_type, spin=spin, scaling_matrix=scaling_matrix)
+ else:
+ A, b = None, None
# Initialise the iteration counter and function, gradient, and
Hessian call counters.
iter_count = 0
@@ -968,18 +950,26 @@
# Minimisation.
###############
- # Constrained optimisation.
- if constraints:
+ # Grid search.
+ if search('^[Gg]rid', min_algor):
+ results = grid(func=self.mf.func, args=(), num_incs=inc,
lower=lower, upper=upper, A=A, b=b, verbosity=verbosity)
+
+ # Unpack the results.
+ param_vector, func, iter, warning = results
+ fc = iter
+ gc = 0.0
+ hc = 0.0
+
+ # Minimisation.
+ else:
results = generic_minimise(func=self.mf.func,
dfunc=self.mf.dfunc, d2func=self.mf.d2func, args=(), x0=param_vector,
min_algor=min_algor, min_options=min_options, func_tol=func_tol,
grad_tol=grad_tol, maxiter=max_iterations, A=A, b=b, full_output=1,
print_flag=verbosity)
- # Unconstrained optimisation.
- else:
- results = generic_minimise(func=self.mf.func,
dfunc=self.mf.dfunc, d2func=self.mf.d2func, args=(), x0=param_vector,
min_algor=min_algor, min_options=min_options, func_tol=func_tol,
grad_tol=grad_tol, maxiter=max_iterations, full_output=1,
print_flag=verbosity)
-
- # Disassemble the results.
- if results == None:
- continue
- param_vector, func, iter, fc, gc, hc, warning = results
+ # Unpack the results.
+ if results == None:
+ continue
+ param_vector, func, iter, fc, gc, hc, warning = results
+
+ # Update the stats.
iter_count = iter_count + iter
f_count = f_count + fc
g_count = g_count + gc
@@ -1274,28 +1264,3 @@
spin.g_count = None
spin.h_count = None
spin.warning = None
-
-
- def test_grid_size(self, min_options, verbosity=1):
- """Test the size of the grid search.
-
- @param min_options: The grid search configuration.
- @type min_options: list
- @keyword verbosity: A flag specifying the amount of information to
print. The higher the
- value, the greater the verbosity.
- @type verbosity: int
- @raises RelaxError: If the grid size corresponds to a long int.
- """
-
- # Determine the grid size.
- grid_size = 1
- for i in xrange(len(min_options)):
- grid_size = grid_size * min_options[i][0]
-
- # Print out.
- if verbosity >= 1:
- print(("Unconstrained grid search size: " + repr(grid_size) + "
(constraints may decrease this size).\n"))
-
- # Too big.
- if isinstance(grid_size, long):
- raise RelaxError("A grid search of size " + repr(grid_size) + "
is too large.")
r9562 - /1.3/specific_fns/model_free/mf_minimise.py