Sporco

• Added sporco.cuda interface to CUDA extension package sporco-cuda
• Completely restructured example scripts, which are now used to generate
example Jupyter notebooks and corresponding docs pages
• Modifications to the interfaces of the plot module functions that will
break existing code that uses the fgrf or axrf parameters
• Modifications to the interface of the plot.plot function that will
break existing code that uses the lwidth, lstyle, msize, or mstyle
parameters
• Replaced function util.imageblocks with util.extractblocks, and introduced
new functions util.averageblocks and util.combineblocks
• Added new module admm.parcbpdn implementing the parallel ADMM CBPDN
solver
• Fixed bugs in cbpdn.ConvBPDNProjL1 and cbpdn.ConvMinL1InL2Ball

• Fixed bugs in admm.ccmod module
• Test images required by usage examples are now included with the package
• Modifications to util.ExampleImages interface (these changes will break
code that uses the previous interface)
• Added call graph diagrams for many classes (see `Notes` in the package
documentation)
• New class admm.ADMMConsensus for ADMM consensus problems
• Major changes to ccmod module, including restructuring class hierarchy,
a new ADMM consensus solver, and moving of ccmod.ConvCnstrMODMaskDcpl to
a separate module
• Changes to cbpdndl module related to ccmod module restructuring
• New module ccmodmd supporting multiple algorithms for the dictionary
update with mask decoupling
• New module parcnsdl with a parallel processing implementation of
convolutional dictionary learning with the ADMM consensus dictionary
update
• Moved class ConvRepIndexing from cbpdn and ccmod modules to new module
cnvrep. Additional classes from ccmod module also moved into cnvrep.
• New module prox supporting evaluation of various norms and their proximal
and projection operators
• New classes bpdn.BPDNProjL1, bpdn.MinL1InL2Ball, cbpdn.ConvBPDNProjL1,
and cbpdn.ConvMinL1InL2Ball supporting constrained forms of the BPDN
and CBPDN problems

• Renamed prox.prox_l1l2 to prox_sl1l2
• Added new modules and example scripts for the Plug and Play Priors method
• New functions for nearest Kronecker product, Kronecker product SVD in linalg
module, and new functions for least absolute deviations linear regression,
and least maximum error linear regression in interp module
• Renamed linalg.GradientFilters to linalg.gradient_filters, linalg.Gax to
linalg.grad, linalg.GTax to linalg.gradT, util.extractblocks to
util.extract_blocks, util.averageblocks to util.average_blocks, and
util.combineblocks to util.combine_blocks
• Moved util.pca to linalg.pca

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

• No changes yet

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

• Added support for complex signals in admm.cbpdn, admm.ccmod, admm.tvl1,
and admm.tvl2 modules
• Fixed bug in admm.cbpdnin when used with multi-signal arrays (K > 1)

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

• Major support module restructuring: numerous functions from sporco.util
and sporco.linalg moved to new modules sporco.array, sporco.fft,
and sporco.signal, and functions nkp, kpsvd, tikhonov_filter,
gaussian, and local_contrast_normalise moved from sporco.util to
sporco.linalg
• Added new functions prox.norm_dl1l2 and prox.prox_dl1l2 for difference of
ℓ1 and ℓ2 norms and corresponding proximal operator
• Major restructuring of sporco.fista modules, now renamed to sporco.pgm
• Significant change to interface of fft.fftconv function
• New classes for sparse coding and dictionary learning with a weighted ℓ2
data fidelity term
• Functionality depending on use of fork in multiprocessing (modules
admm.parcbpdn and dictlrn.prlcnscdl, and parallel computation of
util.grid_search) no longer supported under MacOS