- 2.2.1-r2:
- Corrected missing keywords in immultifit.
- Adapted to CASA 5.x (so far)
- 2.2.1-r1:
- Correction of small bug in the model-writting operation onto the measurement set (i.e., when write_model='model'
- 2.2.1:
- Faster data read and model fitting (more efficient code). In some cases,
the speed-up can even be of x2.
- Improved w-term treatment in mosaics.
- Added a "test suite" with some example cases.
- New model (GaussianRing) and new parameter (HankelOrder), to work with models with no explicit Hankel transform (in these cases, the Bessel Function J0 is approximated with a Taylor series of HankelOrder order).
- New keywords:
-ldfac: to fine-tune the proportionality between FWHM of the primary beam
and the dish diameter: FWHM = ldfac*lambda/D (default: 1.22)
-phase_gains and amp_gains: to define parameterized antenna-gain
functions, to be fitted simultaneously with the source parameters.
-HankelOrder: to set the approximation degree of J0 in the expansion
of source models with no explicit analytic form (new feature).
- 2.2.0-r4:
- Corrected bug in "expo" model. It was not returning "major" as defined in the documentation (i.e., as the FWHM).
- 2.2.0-r3:
- Added ichan and echan keywords to the "modelFromClean"
function.
- Changed "write_model" to "write", where "write" can be set to
either '', 'model', or 'residuals'. The new 'residuals' feature
saves the post-fit residuals into the CORRECTED column.
- 2.2.0-r1:
- Changed pow(x,0.5) by sqrt in w-term correction. Speed-up by
roughly a factor 3 !
- 2.2.0:
- Added "phase_center" keyword, for the user to specify the
reference point on the sky to locate the model components.
- Full support of W projection.
- Better support for mosaic observations. Phase rotation is
not needed anymore, but primary beams are still approximated
using circular Gaussians (hence, the flux densities fitted to
sources close to the edge of the primary beam(s) may be
biased).
In addition, the model components must be small, compared to the
primary beams. If very large sources are to be fitted, please use
the "immultifit" function, instead of "uvmultifit".
- PATCH 2.1.4-r3:
- More precise estimates of derivatives for the Hessian matrix, when using spectral-line mode.
- Re-fitting with fine tunning was broken. Now, it is back again.
- PATCH 2.1.4-r2:
- The fit is not aborted if all visibilities of a
frequency channel are flagged and OneFitPErChannel = True.
- Added a clarifying message about re-scaling of weigths after the fit.
- Corrected bug: the power index of the weights used in the fit was
twice the one specified in "wgt_power". The effect of this
bug on the fitted parameters is small (typically within 1 sigma). It is
similar to a re-tunning of the "robust" parameter when imaging with the
Briggs weighting.
- NEW IN 2.1.4:
1.- Changed two wrong explanations in the documentation
(about meaning of pre-scaled reduced Chi-Squared and about
relationship between axis-ratio definition and position angle)
2.- New keyword 'only_flux', to accelerate fits of many
components with only their flux density as free parameter
(e.g., for simple implementations of sparse-sampling
deconvolution techniques)
3.- New keyword 'proper_motion', to correct for proper
motions of fast moving sources (e.g., binary stars) during
the fit, on-the-fly.
4.- Corrected bug that broke the possibility of using MJDrange.
Now, if user wants to refit the dataset (e.g., with a new MJDrange)
the fit() function shall be called with "reinit_model=False", to
avoid memory leaks.
TODO: Implement possibility of FITTING the proper motions.
- NEW IN 2.1.3:
1.- Improved convergence of Levenberg-Marquardt
2.- Simulated full-pol data can now be fitted with no problems.
- NEW IN 2.1.2:
1.- Corrected bug that gave flipped coordinate offsets in IMMULTIFIT
- NEW IN 2.01:
1.- Improved convergence of LM.
2.- Corrected bug that didn't read the coordinates shifts of mosaic observations IN NON-SIMULATED DATA.
- NEW IN 2.0b:
1.- Better memory handling in C++ extension.
2.- Faster processing (less overhead with Python API)
3.- Own implementation of Levenberg-Marquardt, to avoid the huge memory usage of the standard Scipy library (although at the cost of speed if the number of fitting parameters is large).
4.- Full support for CASA 4.2
5.- Added keywords LMtune and SMPtune, to tune the parameters that control the Levenberg-Marquardt and the Simplex algorithm, respectively.
6.- New feature in immultifit. Now, it also saves an image (cube) with the best-fit unconvolved model. BEWARE, however, of the "window effect", that will indeed convolve the model by the FT of the image FFT coverage.
- REMOVED:
1.- Keyword clib is removed. Now, all processing is done with the C++ extension. User must compile it as explained in the README file.
2.- Keywords rstep and rxtol are removed (now use the SMPtune keyword)
3.- The GUI has been removed (PySide doesn't seem to work well with the current CASA release. We will try to add it back in a future version).
