Sporco

----------------------------

• Fixed functional evaluation error in cbpdn.ConvBPDNMaskDcpl
• Fixed bug in cbpdn.ConvTwoBlockCnstrnt with multi-channel dictionary
• New class ccmod.ConvCnstrMODMaskDcpl for dictionary update with mask
decoupling
• New class cbpdndl.ConvBPDNMaskDcplDictLearn for dictionary learning
with mask decoupling
• Corrected serious error in demo_dictlrn_cbpdn_md.py
• Fixed bug causing non-deterministic 'AuxVarObj' option behaviour
• New functions util.transpose_ntpl_list, util.complex_randn,
util.idle_cpu_count
• In cmod and ccmod modules, renamed sparse representation variable from A
to Z
• Changed callback function mechanism in admm.ADMM.solve and
dictlrn.DictLearn.solve: callback function no longer takes iteration number
as an argument (it is not available as a class attribute), and can terminate
solve iterations by returning a boolean True value.
• New parameters in plot.plot for selecting marker size and style, and in
plot.imview for specifying matplotlib.colors.Normalize object
• Added L21Weight option for cbpdn.ConvBPDNJoint
• Fixed bug in cbpdn.AddMaskSim handling of multi-channel dictionaries

----------------------------

• Fixed serious bug in cbpdn.ConvBPDNGradReg.setdict and
cbpdn.ConvBPDNGradReg.xstep resulting in incorrect solution of
linear system
• Fixed bug in cbpdn.GenericConvBPDN.xstep (and same method in some
derived classes) affecting calculation of linear solver accuracy for
single-channel dictionaries
• Fixed bug in multi-channel data handling in cbpdn.AddMaskSim
• Fixed bug in util.netgetdata
• New functions linalg.solvedbd_sm, linalg.solvedbd_sm_c
• Improved documentation of admm.admm module
• Changed default line width in plot.plot and added parameter for
specifying label padding to plot.surf
• Improved capabilities of util.Timer class and modified admm.ADMM
class to use it
• New FastSolve option instructs admm.ADMM class to skip
non-essential calculations
• New AccurateDFid option for more accurate functional evaluation in
admm.BPDNDictLearn and admm.ConvBPDNDictLearn
• New IterTimer option to select timer used for admm.ADMM iteration
timing
• Introduced new inner product function linalg.inner and improved
speed of linalg.solvedbi_sm by using it instead of np.sum and
broadcast multiplication

----------------------------

• Bug fix release to correct error in Travis CI configuration
resulting in PyPI releases with broken plotting capabilities

----------------------------

• Major changes to policy of downloading required data on package
build: this functionality is now in script sporco_get_images, which
is not called during package build
• New function util.netgetdata
• Major changes to util.ExampleImages
• Bug fix for multi-channel images in bpdn.AddMaskSim
• Improved handling of floating point images in plot.imview

----------------------------

• New functions util.ntpl2array, util.array2ntpl, plot.close
• Modified util.rgb2gray to support array containing multiple images
• Modified scaling of return value of linalg.fl2norm2 to match docs
• In module linalg, moved functions mae, mse, snr, and psnr to new
module metric, and added new functions isnr, bsnr, pamse, and gmsd
in this module
• New methods admm.ADMM.getmin, cbpdn.AddMaskSim.setdict,
cbpdn.AddMaskSim.getcoef
• Modified classes in modules tvl1 and tvl2 to support Vector TV for
multi-channel images
• Added Jupyter Notebook versions of some example scripts
• Added some new example scripts

----------------------------

• Moved plotting functions from util to new module plot
• New function util.grid_search supporting parallel processing
evaluation of a function on a specified grid
• Extended capabilities of class util.ExampleImages
• New functions linalg.GradientFilters, linalg.promote16, linalg.roll,
linalg.blockcirculant, linalg.mae
• Modified admm.ADMM class so that objects of this type can be pickled
• Changes to interface of admm.ADMM.__init__,
admm.ADMM.iteration_stats, admm.ADMM.display_status,
admm.ADMMEqual.__init__, admm.ADMMTwoBlockCnstrnt.__init__
• New methods admm.ADMM.set_dtype, admm.ADMM.set_attr,
admm.ADMM.yinit, admm.ADMM.uinit, admm.ADMM.itstat_fields,
admm.ADMM.hdrtxt, admm.ADMM.hdrval, admm.ADMM.itstat_extra,
admm.ADMM.var_u
• In admm.ADMM and derived classes, major changes to object
initialisation and iteration stats calculation mechanisms, including
more careful initialisation of arrays to ensure consistent dtype
across all working variables
• In module bpdn, created new common base class GenericBPDN
• In module cbpdn, created new common base class GenericConvBPDN
• Improvements to docs