frb - tool used to search Fast Radio Bursts in auto-spectral data

Up-to-date version will be released soon

numpy, scipy, matplotlib.pyplot (for plots)

user@host:~$ python search_file.py data.dat -nu_max NU_MAX -dnu DNU -dt DT -dm_min DM_MIN -dm_max DM_MAX [-batchsize BATCHSIZE] [-d_t D_T] [-d_dm D_DM] [-perc PERC] [-savefig_dyn fig.png] [-savefig_dedm fig.png] [-threads THREADS]

Parameters:

• data.dat - binary or text file with time sequence of dynamical spectra. If text file => # of columns = # of frequency channels and # of rows = # of time measurements. If binary then np.shape(np.load('data.dat')) = (# of freq. channels, # of time measurements,)

• -nu_max - frequency of highest frequency channel [MHz].

• -dnu - frequency width of single frequency channel [MHz].

• -dt - time step (resolution) [s].

• dm_min - minimal value of DM window to search [cm^3 / pc].

• dm_max - maximum value of DM window to search [cm^3 / pc].

• batchsize - size of image in t-direction, that will be searched for candidates in batches. Default: 100000

• d_t - width of feature [s] in (t, DM)-space along t-axis to treat it as candidate. Default: 0.003

• d_dm - width of feature [cm^3/pc] in (t, DM)-space along DM-axis to treat it as candidate. Default: 100.

• perc - percentile of image values that is used to blank image before searching for objects. Default: 99.5

• savefig_dyn - file name for saving picture of dynamical spectra.

• savefig_dedm - file name for saving picture of de-dispersed frequency averaged dynamical spectra.

• save_result - file name to save (t, DM)-coordinates of found candidates. [s, cm^3/pc]

• threads - number of threads used for parallelization of grid de-dispersion. Default is 1 (don't use parallelization).

Algorithm searches for extended regions in image of de-dispersed frequency averaged dynamical spectra (that is (t, DM)-plane). Currently there are 3 tunable parameters:

• perc. Current experience suggests values 99.9 - 99.95. Low value could bring many false features in (t, DM)-space, but as long as we can compare results using different telescopes it doesn't seem to be an issue. Value that is too high can split characteristic x-shaped dispersed signal features.

• d_t & d_dm. Some experience with fake FRB injected in real data have shown that 2 most informative features in classification of FRB candidates on (t, DM)-plane are their widths in t- and DM-directions. Currently it is the only method of classification that has been implemented. Nonetheless one can use any other features and their own algorithms of classification by overriding/extending TDMImageObjects._classify method that gets image and labelled array as first two positional arguments.

Current implementation allows parallelization of grid de-dispersion step using multiprocessing module.

Using text files requires additional RAM that can be a problem for large data sets.

Copyright 2015 Ilya Pashchenko.

frb is free software made available under the MIT License. For details see the LICENSE file.