No description
  • C++ 86.7%
  • C 11.4%
  • CMake 1.9%
Find a file
2026-10-13 22:31:51 +02:00
modellists add new model list 2025-04-07 16:29:26 +02:00
src Allow reusing the omega range of the input 2026-10-13 22:31:51 +02:00
CMakeLists.txt Add support for fitting model that lacks paramters 2026-10-02 13:53:49 +02:00
gpl-3.0.txt Add License 2025-01-21 10:19:48 +01:00
README.md Add support for generateing a synthetic dataset based on a real dataset and a residual dataset form a real dataset 2026-10-13 20:36:31 +02:00

KissDatasetGenerator

KissDatasetGenerator creates dataset tarballs for TorchKissAnn KissDatasetGenerator uses eisgenerator to create synthetic datasets and can also create datasets from real data of various types KissDatasetGenerator supports down-selecting data as well as applying transformations sutch as those from libeisnoise and libeisdrt.

Requirements

  • You must be running unix or a unix-like os such as linux
  • A c++17 compiler like gcc
  • cmake
  • pkg-config
  • libeisgenerator
  • libkisstype
  • libeisdrt
  • libeisnoise

Build and instructions

To build and install this tool do the following:

$ git clone REPO_URL dataformaters
$ cd dataformaters
$ mkdir build
$ cd build
$ cmake ..
$ make
# make install

Usage

Usage: kissdatasetgenerator [OPTION...] 
Application that checkes and conditions a dataset in an eis dir

  -a, --archive              save as a tar archive instead of a directory
  -c, --frequency-count=[NUMBER]   the number of frequencies to simmulate,
                             default: 100
  -d, --dataset=[PATH]       input dataset to use or the model string, in case
                             of the regression purpose
  -g, --no-negative          remove examples with negative labels from the
                             dataset
  -l, --select-labels[=[LABLE1,LABEL2,...]]
                             select thiese labels to appear in the output
                             dataset (requires them to be present in the
                             input)
  -n, --no-normalization     turn off normalization on dataset types that
                             ususally do this
  -o, --out-dir=[PATH]       directory where to export dataset
  -p, --test-percent=[NUMBER]   test dataset percentage
  -q, --quiet                Show only errors
  -r, --frequency-range=[RANGE]   Frequency range to simulate for simulated
                             datasets
  -s, --size=[NUMBER]        size the dataset should have
  -t, --type=[TYPE]          type of dataset to export valid types: gen,
                             passfail, regression, real, residual
  -v, --verbose              Show debug messages
  -x, --extra-inputs[=[INPUT1,INPUT2,...]]
                             select thiese labels to appear in the output
                             dataset as extra inputs (requires them to be
                             present in the input)
  -?, --help                 Give this help list
      --usage                Give a short usage message
  -V, --version              Print program version

Mandatory or optional arguments to long options are also mandatory or optional
for any corresponding short options.

Report bugs to <carl@uvos.xyz>.

For example to generate a synthetic parameter regression dataset one could use:

kissdatasetgenerator -t regression -d "r-rc" -a -o r_rc -p 10 -c 20

which will result in the files r_rc_test.tar and r_rc_train.tar