Author: bugman Date: Fri Nov 7 16:51:22 2014 New Revision: 26451 URL: http://svn.gna.org/viewcvs/relax?rev=26451&view=rev Log: Converted the Sobol' rotation matrices to float32 in the frame order target function. This is to conserve huge amounts of memory to allow for more Sobol' points to be used. For example for the models which use 3D Sobol' points (isotropic cone and pseudo-ellipse), a maximum of 50000 Sobol' points requires 50000000 to be created, using about 15 Gb of RAM. 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=26451&r1=26450&r2=26451&view=diff ============================================================================== --- branches/frame_order_cleanup/target_functions/frame_order.py (original) +++ branches/frame_order_cleanup/target_functions/frame_order.py Fri Nov 7 16:51:22 2014 @@ -2031,8 +2031,8 @@ sobol_data.model = self.model sobol_data.total_num = total_num sobol_data.sobol_angles = zeros((m, total_num), float32) - sobol_data.Ri_prime = zeros((total_num, 3, 3), float64) - sobol_data.Ri2_prime = zeros((total_num, 3, 3), float64) + sobol_data.Ri_prime = zeros((total_num, 3, 3), float32) + sobol_data.Ri2_prime = zeros((total_num, 3, 3), float32) # The Sobol' points. points = i4_sobol_generate(m, total_num, 1000)