def __init__( self ):
self.gp = CppGaussianProcess()
class GaussianProcess( object ):
def __init__( self ):
self.gp = CppGaussianProcess()
def function_domain( self, func_domain ):
self.func_domain = func_domain
self.gp.domain( self.func_domain.ranges.flatten() )
def build( self, build_points, build_values ):
self.gp.build( build_points, build_values )
def evaluate_set( self, test_points ):
print test_points.shape
print self.gp.evaluate_set( test_points ).shape
return self.gp.evaluate_set( test_points )[:,0]
def evaluate_surface_variance( self, test_points ):
return self.gp.evaluate_surface_variance( test_points )[:,0]
def leave_one_out_cross_validation( self ):
return self.gp.leave_one_out_cross_validation()
def set_correlation_lengths( self, corr_lengths ):
self.gp.set_correlation_lengths( corr_lengths )
def get_correlation_lengths( self ):
return self.gp.get_correlation_lengths()
def get_correlation_thetas( self ):
return self.gp.get_correlation_thetas()
def set_optimization_method( self, method_enum ):
self.gp.set_optimization_method( method_enum )
def set_verbosity( self, verbosity ):
self.gp.set_verbosity( verbosity )
def get_correlation_matrix( self ):
return self.gp.get_correlation_matrix()
def clear( self ):
# delete the internal memory of the c++ object
self.gp.clear()
def parse_cross_validation_data( self, pts, vals, cv_params ):
return pts, vals
def initialise_cross_validation( self, pts, vals, cv_params,
use_fast_cross_validation ):
self.use_fast_cross_validation = use_fast_cross_validation
# build an initial approximation on the entire data set to get
# correlation lengths which can then be used as a good guess
# for local optimization of the maximum likelihood. The
# use of the local optimizer should speed up convergence
#self.set_optimization_method( 3 ) # global_local
self.set_optimization_method( 4 ) # sampling
self.build( pts, vals )
self.corr_lengths = self.get_correlation_lengths()
self.set_optimization_method( 1 ) # local
def compute_residuals( self, pts, vals, train_indices,
validation_indices, cv_params = None ):
pts_train = pts[:,train_indices]
vals_train = vals[train_indices]
pts_validation = pts[:,validation_indices]
vals_validation = vals[validation_indices]
self.set_correlation_lengths( self.corr_lengths )
self.build( pts_train, vals_train )
fn_vals_pred = self.evaluate_set( pts_validation )
# must clear predictor if predictor is a c++ wrapped class
# otherwise c++ internally created memory will not be deleted
self.clear()
print 'In gaussian_process.compute_residuals'
# cv_params has not be changed, e.g. new params added so return None
return vals_validation - fn_vals_pred, None
