The nnp-select tool extracts a small subset from a larger training data set, e.g. for testing purposes. The initial set input.data is expected to be located in the current working directory. Two different modes of operation are supported: selecting structures

  1. randomly with given seed, or

  2. at regular intervals.


nnp-select <mode> <arg1 <arg2>>
           <mode> ... Choose selection mode (random/interval).
           Arguments for mode "random":
             <arg1> ... Ratio of selected structures (1.0 equals 100 %).
             <arg2> ... Seed for random number generator (integer).
           Arguments for mode "interval":
             <arg1> ... Select structures in this interval (integer).
           Execute in directory with these NNP files present:
           - input.data (structure file)

Sample screen output:

*** NNP-SELECT ****************************************************************

Selecting every 3 structure.
Structure       1 selected.
Structure       4 selected.
Structure       7 selected.
Structure      10 selected.
Structure      13 selected.
Structure      16 selected.
Total    structures           :      16
Selected structures           :       6
Selected structures percentage: 37.500 %

File output:

  • output.data: The requested subset of training structures.

  • reject.data: The rejected configurations, i.e. all data minus output.data.

  • nnp-select.log : Log file (copy of screen output).


  • Select randomly 10% of the original set with random seed 123:

    nnp-select random 0.1 123
  • Select every 20th structure from original set (starting with structure 1):

    nnp-select interval 20