Map-maker for Herschel/PACS
This package provides tools to create maps from Herschel/PACS observations. It is massively parallel (MPI + OpenMP) and is written in Fortran (for the number crunching) and Python (for the abstraction easiness). Because of its high level of modularity, support for other astronomical instruments might be added in the future.
What’s new ?
- update for pyoperators 0.4 & pyfits 3.x
- bring back map-making with distributed maps
- comprehensive MPI testing of the naive, ls and rls mapper.
- PACS: API Change: change PACS’s get_tod default behaviour, by not applying the flatfield correction and subtracting the mean value. To do so, the appropriate keywords must be set to True
- PACS: make the PacsInstrument class be instanciated and used without a PacsObservation
- PACS: gather calibration files in PacsInfo class. Implement get_calibration methods, which centralise calibration file handling. Make Responsivity and BadPixelMask calibration files handled from Python. Add access to Absorption and FilterTransmission calibration files. Update calibration files to HCSS 8.
- PACS: fix masking of temporal jumps
- PACS: flag as ‘other’ all pointings before the first inscan one.
- datatypes: add fast median method in Quantity. (Tod takes mask into account)
- datatypes: getitem should also handle list of items.
- datatypes: Add astype method, fix wrapping of some ndarray’s methods in Tod and Map
- datatypes: implement rightward/leftward broadcasting in derived units.
- acquisitionmodels: add class PointingMatrix which contain the pointing matrix information and info about the input map. Make ProjectionOperator handle PointingMatrix inputs and the get_pointing_matrix routines return a PointingMatrix object. As a consequence, pointing matrices can be manipulated before being used to initialise a ProjectionOperator. Make get_mask work for MPI-distributed maps and TODs. Make get_ptp work for block column projections. Remove header, npixels_per_sample attributes
- acquisitionmodels: new methods intersects, apply_mask, get_pTx_pT1 (can significantly speed up naive maps) in ProjectionOperator. Remove _matrix attribute in projections.
- acquisitionmodels: remove 3rd dimension in pointing_matrix_mask/pack. During packing, reject PointingElements with a negative weight.
- acquisitionmodels: in InvNttOperator, fix filter_length if the operator has several blocks.
- acquisitionmodels: PackOperator and UnpackOperator: add unary and binary rules. Hence remove transpose methods. Rework Fortran routines to differentiate in-place and out-of-place operations. Add comprehensive test cases.
- acquisitionmodels: in DdTddOperator, if the the scalar argument is zero, return a ZeroOperator.
- acquisitionmodels: make DiscreteDifferenceOperator work even if there are more MPI processes than elements along the first dimension. Fix MPI outplace diffT case.
- acquisitionmodels: make the block_diagonal decorator handle implicit partitions, take the usual partitionin and axisin keywords instead of partition and axis.
- API change: remove_nonfinite → filter_nonfinite, add outplace operations, with and without a mask, Add processing Fortran interface.
- interpolate_linear, filter_median, filter_nonfinite, filter_polynomial: add ‘out’ keyword. Use partition instead of nsamples.
- inplace filter_polynomial
- implement median function, handling mask and NaN, parallelised along a given axis.
- observation.plot_scan: plot detector footprint on the first valid pointing
- force numeric locale to ‘POSIX’, to prevent gfortran using comma as fraction separator PR47007.
- new functions gather_fitsheader and scatter_fitsheader
- move all information messages to Python. Group info messages for block column projection operators.
- refactor mapper_ls and mapper_rls, by relying on algebraic rules and by removing the use of ReshapeOperator.
- in write_fits, fix MPI case. Do not rely on pyfits private methods and do the padding ourselves.
- add argument r0 in airy_disk function to specify first zero, as an alternative to fwhm.
- in imshow, discard outliers by using a percentile of the valid values.
- general: improve error messages
- rename numpyutils.py → utils.py and utils.py → datautils.py.
- wscript: preprocess var.py.in file to include configuration options and version name.
- wscript: make waf loc display Python/Fortran/Total lines of code. Remove loc-fortran and loc-python.
- wscript: run test-mpi test suite with np = 1 even if —enable-mpi has not been set
What’s new ?
- Update to PyOperators v0.3
- Update to waf 1.6.10
- Bring back MPI read/write capabilities for FitsArray/Map/Tod instances
- Bring back MPI-distributed discrete difference operator
- Pointing class: make constructor more generic and easy to use
- API changes: module quantity has been renamed to quantities. Although the old names will be kept for some time, operators now consistently end with
What’s new ?
- first release depending on the PyOperator package, which is the result of the merging of Tamasis’ new acquisition models and of the work by N. Barbey (linear_operators).
- the operator class is much more flexible, faster and the memory is better managed. In particular, operators can be combined in blocks. They can also be reduced, following algebraic rules.
- as a consequence, are available many more algorithms (FminBFGS, FminCG, FminCOBYLA, FminLBFGSB, FminNCG, FminPowell, FminSLSQP, FminTNC, Lanczos, Double Lookup Inference) and priors through an
Operator interface to the wavelet package pywt.
Observation classes are now populated with useful methods, so that Tamasis can now be used for other instruments than PACS.
- configuration stage now checks the version of the python modules (waf patch)
- dependencies to MPI and WCSLIB have been made optional.
- test suite converted to nose.
- API changes: in the Projection operator for the PACS instrument, use “
downsampling=True”, instead of “
oversampling=False”. The Pointing class constructor has now a more flexible single positional argument:
What’s new ?
- parallel write of FITS files using MPI-IO
- implement class
DistributedArray to ease manipulation of MPI-distributed maps
- new general purpose operators:
Rounding (9 different methods),
- new PACS operators: Jy-to-Volts and Volts-to-ADU converters
- linesearch with Armijo rule or strong Wolfe-Powell
- command line interface for map-making routines, contributed by Nicolas Barbey
What’s new ?
- API change:
InvNtt operator now works in direct space so it returns the actual time-time noise correlation matrix ala MADmap.
get_detector_stddev, which returns the detector standard deviation given an optional median filtering length.
plot_tod method useful to display a
Tod with the mask information, ex:
plot_tod(tod) will plot detector #100’s timeline
- parallel (OpenMP+MPI) non-linear conjugate gradient solver with preconditioning, with different descent methods and norms