GPU accelerated stochastic calibration

nlesc-dirac / sagecal

SAGECal is a fast, memory efficient and GPU accelerated radio interferometric calibration program. It supports all source models including points, Gaussians and Shapelets. Distributed calibration using MPI and consensus optimization is enabled. Both spectral and spatial priors can be used as constraints. Tools to build/restore sky models are included.

http://sagecal.sourceforge.net

GNU General Public License v2.0

12 stars 8 forks source link

GPU accelerated stochastic calibration #116

Closed SarodYatawatta closed 4 years ago

SarodYatawatta commented 4 years ago

upgrade to version 0.7.0
sagecal-mpi GPU cmake build
demo for stochastic LBFGS
bandpass calibration with consensus ADMM
GPU accelerated stochastic LBFGS

SarodYatawatta commented 4 years ago

What is elastic net?

L1+L2 regularization is elastic net

SarodYatawatta commented 4 years ago

What is persistent_data_t needed for?

To store info from previous minibatches in LBFGS

SarodYatawatta commented 4 years ago

I will merge it after some more testing