PRIISM 0.14.0

Following contribution from @iganay958 has been incorporated.

• Installation guide for conda environment has been added
• Fixed bug related to comparison of floating point numbers

PRIISM 0.13.0

This release intends to support CASA 6.6.4 or higher. Requirements file has been separated into two: requirements.txt is for the latest CASA release (6.6.4+) while requirements-old.txt is for older CASA releases.

It is advisable that, if you already have PRIISM installation with older CASA version, you create new PRIISM installation from scratch based on CASA 6.6.4.

CASA 6.6.4 introduced new casadata management system. It no longer depends on casadata module. Instead, it will create data directory outside CASA installation (by default, ~/.casa/data) and the data directory will be shared among CASA releases. For this reason, you should create ~/.casa/data directory in advance.

PRIISM 0.12.0

PRIISM 0.12.0 supports Python 3.10 on Linux and macOS 12.

PRIISM 0.11.5

This release includes several bug fixes. Most important one is the fix for FFT normalization issue in sparseimaging reported by @shoshiay. The fix was provided by @ikeda46. Another bug was reported by @shoshiay regarding retrieval of rest frequency information from input measurementset. Finally, technical problem on GitHub Actions has been fixed.

Previous releases, 0.10.0, 0.11.0, 0.11.1, 0.11.2, 0.11.3, and 0.11.4 were not recommended because they are affected by FFT normalization issue so that converged solution could be wrong. Please update to 0.11.5 as soon as possible if you are using these versions.

PRIISM 0.11.4 (not recommended)

This release replaces GPyOpt, dependency for Bayesian optimization of cross-validation, with scikit-optimize (skopt) as GPyOpt is no longer maintained. Use of skopt fixes the issue that Baysian optimizer is not working with the latest version of numpy.

PRIISM 0.11.3 (not recommended)

This release includes a bug fix on rest frequency propagation from input MeasurementSets to output FITS image. The bug was reported to #35 by @shoshiay and has been fixed in #36. Also, installation instruction (in Japanese) for data reduction servers in Astronomy Data Center (NAOJ) has been updated to follow recent update of their system.

PRIISM 0.11.2 (not recommended)

This release includes a few changes for compatibility with numpy 1.24.0, which removed several data types.

PRIISM 0.11.1 (not recommended)

This release improves PRIISM build process so that it supports a build against monolithic (tar-ball) CASA release. Also, several warning messages during build process have been eliminated.

PRIISM 0.11.0 (not recommended)

This release offers new cross-validation option using Bayesian optimization. The cross validation method, crossvalidation, now has a new parameter optimizer to choose how to find the best pair of hyper-parameters (L1 and Ltsv). Setting optimizer='classical' (default) examines all combinations of given list of hyper-parameters while setting optimizer='bayesian' utilizes Bayesian optimization to find the solution. Example usage is described in cvrun.ipynb. Bayesian optimization enables users to search wide range of hyper-parameters with practical number of evaluations. In cvrun.ipynb, for example, we have shown that the algorithm is able to find the best solution only less than 15 evaluations from 12x12 (144 combinations) hyper-parameters.

In addition to the new feature described above, this release also includes bug fix on sparseimaging (by @ikeda46) as well as an improvement of build process (by @astropenguin).

Note that this release introduces new dependency to GPyOpt.

PRIISM 0.10.0 (not recommended)

This release incorporates sparseimaging update on NUFFT for proper computation of off-grid data. Also, new parameter scalehyperparam has been defined for solve and crossvalidation to absorb the change about scaling of hyper-parameter for regularization terms L1 and TSV. Default for scalehyperparam is True to keep backward compatibility. Please keep in mind that, if you set this parameter to False, you will obtain completely different image from the previous versions of PRIISM. Therefore, it is important that you must leave scalehyperparam default value when you want to compare the result with the ones obtained by older versions of PRIISM. When you start working on new dataset, scalehyperparam can be either True or False.