process_using.m

Applies the specified trained neural network file to the supplied DEER trace.

Syntax

    answers=process_using(deer_traces,net_file_name)

Parameters

  deer_traces  -  DEER trace(s) as a column vector or a 
                  matrix with multiple columns. The num-
                  ber of rows must match the input size 
                  of the neural network.

  net_file_name - the name of the file containing a neu-
                  ral network object, including full
                  path and the .mat extension.

Outputs

  answers       - neural network output(s) as a column 
                  vector or a matrix with multiple col-
                  umns. The number of rows matches the 
                  output size of the neural network.

Examples

In the example below, a batch of 100 distribution and DEER trace pairs are generated and processed using the generic distance distribution extraction netset:

      % Load default parameters
      parameters=library_dd; 

      % Number of traces
      parameters.ntraces=100;

      % Time and distance point counts
      parameters.np_time=512;
      parameters.np_dist=512;

      % PDS experiment type
      parameters.expt='deer';

      % Generate a test library
      library=deer_lib_gen([],parameters);

      % Run the analysis on the example set
      answers=process_using(deer_trace_lib,'net_set_any_peaks/1.mat');

Neural network answers can be plotted against the known right answers:

       % Plot an example
       example=randi(100); figure(); subplot(1,2,1); 
       plot(library.time_grid,library.deer_noisy_lib(:,example)); 
       kgrid; ktitle('Input DEER trace'); kxlabel('time, s');
       subplot(1,2,2); plot(library.dist_grid,[answers(:,example) ...
                                               library.dist_distr_lib(:,example)]); 
       kgrid; ktitle('Output distribution'); kxlabel('distance, $\rm{\AA}$');
       klegend('Network answer','True answer');

The resulting figure is shown below.

Notes

A CUDA capable GPU is not required by this function, it is generally fast enough even on minimal hardware.

See also

deer_lib_gen.m, deernet.m, dist_range.m, elexsys2deernet.m, train_one_net.m, Neural network module, Built-in_experiments

Version 2.8, authors: Ilya Kuprov, Steve Worswick, Jake Keeley