Skip to content

Latest commit

 

History

History
35 lines (21 loc) · 1.12 KB

TODO.md

File metadata and controls

35 lines (21 loc) · 1.12 KB

Here is a "wish list" of dedispersers we'd like to compare:

  • Heimdall (in progress)

  • sigproc

  • presto single_pulse_search

  • FREDDA (not public)

  • AstroAccelerate

General:

  • Currently, the entire intensity array is held in memory, and dedispersed with a single call dedisperser.dedisperse(). For large timestreams, it would be better to simulate the array incrementally, and define an API for incremental dedispersion.

  • Currently, we implement Gaussian profiles (if the intrinsic_width parameter is nonzero). Is this the best choice? Would it make sense to have a boolean flag to switch to boxcar profiles?

  • Multiprocessing/MPI runs

  • Command-line tools for combining/appending

  • More plotting and postprocessing tools (for example, based on success fraction rather than reported SNR.)

Bonsai:

  • Currently, the analytic transfer matrix is computed from scratch whenever a bonsai_dedisperser is constructed, which is annoying! Should cache it in an HDF5 file (this is mostly implemented in bonsai already.)

  • Cleanup: less verbose output from bonsai_dedisperser.jsonize()