Move optimizers into optimize package

CHANGES.rst

@@ -6,10 +6,11 @@ SCICO Release Notes
 Version 0.0.2   (unreleased)
 ----------------------------
 
-• Additional optimization algorithms Linearized ADMM and PDHG
+• Additional optimization algorithms: Linearized ADMM and PDHG.
+• Move optimization algorithms into ``optimize`` subpackage.
 
 
 Version 0.0.1   (2021-11-24)
 ----------------------------
 
-• Initial release
+• Initial release.

examples/scripts/ct_astra_tv_admm.py

@@ -9,7 +9,7 @@
 =====================================
 
 This example demonstrates the use of class
-[admm.ADMM](../_autosummary/scico.admm.rst#scico.admm.ADMM) to solve a
+[admm.ADMM](../_autosummary/scico.optimize.admm.rst#scico.optimize.admm.ADMM) to solve a
 few-view CT reconstruction problem with anisotropic total variation
 (TV) regularization.
 
@@ -30,8 +30,8 @@
 
 import scico.numpy as snp
 from scico import functional, linop, loss, metric, plot
-from scico.admm import ADMM, LinearSubproblemSolver
 from scico.linop.radon_astra import ParallelBeamProjector
+from scico.optimize.admm import ADMM, LinearSubproblemSolver
 from scico.util import device_info
 
 """

examples/scripts/ct_astra_weighted_tv_admm.py

@@ -9,7 +9,7 @@
 =====================================
 
 This example demonstrates the use of class
-[admm.ADMM](../_autosummary/scico.admm.rst#scico.admm.ADMM) to solve a
+[admm.ADMM](../_autosummary/scico.optimize.admm.rst#scico.optimize.admm.ADMM) to solve a
 low-dose CT reconstruction problem with anisotropic total variation
 (TV) regularization.
 
@@ -32,8 +32,8 @@
 
 import scico.numpy as snp
 from scico import functional, linop, loss, metric, plot
-from scico.admm import ADMM, LinearSubproblemSolver
 from scico.linop.radon_astra import ParallelBeamProjector
+from scico.optimize.admm import ADMM, LinearSubproblemSolver
 from scico.util import device_info
 
 """

examples/scripts/ct_svmbir_ppp_bm3d_admm_cg.py

@@ -9,7 +9,7 @@
 ================================================================
 
 This example demonstrates the use of class
-[admm.ADMM](../_autosummary/scico.admm.rst#scico.admm.ADMM) to solve a
+[admm.ADMM](../_autosummary/scico.optimize.admm.rst#scico.optimize.admm.ADMM) to solve a
 tomographic reconstruction problem using the Plug-and-Play Priors framework
 :cite:`venkatakrishnan-2013-plugandplay2`, using BM3D :cite:`dabov-2008-image`
 as a denoiser and SVMBIR :cite:`svmbir-2020` for tomographic projection.
@@ -29,10 +29,10 @@
 
 import scico.numpy as snp
 from scico import metric, plot
-from scico.admm import ADMM, LinearSubproblemSolver
 from scico.functional import BM3D, NonNegativeIndicator
 from scico.linop import Diagonal, Identity
 from scico.linop.radon_svmbir import ParallelBeamProjector, SVMBIRWeightedSquaredL2Loss
+from scico.optimize.admm import ADMM, LinearSubproblemSolver
 from scico.util import device_info
 
 """

examples/scripts/ct_svmbir_ppp_bm3d_admm_prox.py

@@ -9,7 +9,7 @@
 ==================================================================
 
 This example demonstrates the use of class
-[admm.ADMM](../_autosummary/scico.admm.rst#scico.admm.ADMM) to solve a
+[admm.ADMM](../_autosummary/scico.optimize.admm.rst#scico.optimize.admm.ADMM) to solve a
 tomographic reconstruction problem using the Plug-and-Play Priors framework
 :cite:`venkatakrishnan-2013-plugandplay2`, using BM3D :cite:`dabov-2008-image`
 as a denoiser and SVMBIR :cite:`svmbir-2020` for tomographic projection.
@@ -29,10 +29,10 @@
 
 import scico.numpy as snp
 from scico import metric, plot
-from scico.admm import ADMM, LinearSubproblemSolver
 from scico.functional import BM3D, NonNegativeIndicator
 from scico.linop import Diagonal, Identity
 from scico.linop.radon_svmbir import ParallelBeamProjector, SVMBIRWeightedSquaredL2Loss
+from scico.optimize.admm import ADMM, LinearSubproblemSolver
 from scico.util import device_info
 
 """

examples/scripts/ct_svmbir_tv_multi.py

@@ -23,11 +23,10 @@
 
 import scico.numpy as snp
 from scico import functional, linop, metric, plot
-from scico.admm import ADMM, LinearSubproblemSolver
-from scico.ladmm import LinearizedADMM
 from scico.linop import Diagonal
 from scico.linop.radon_svmbir import ParallelBeamProjector, SVMBIRWeightedSquaredL2Loss
-from scico.primaldual import PDHG
+from scico.optimize import PDHG, LinearizedADMM
+from scico.optimize.admm import ADMM, LinearSubproblemSolver
 from scico.util import device_info
 
 """

examples/scripts/deconv_circ_tv_admm.py

@@ -9,7 +9,7 @@
 ================================================================
 
 This example demonstrates the use of class
-[admm.ADMM](../_autosummary/scico.admm.rst#scico.admm.ADMM) to solve
+[admm.ADMM](../_autosummary/scico.optimize.admm.rst#scico.optimize.admm.ADMM) to solve
 an image deconvolution problem with isotropic total variation (TV)
 regularization.
 
@@ -29,7 +29,7 @@
 import scico.numpy as snp
 import scico.random
 from scico import functional, linop, loss, metric, plot
-from scico.admm import ADMM, CircularConvolveSolver
+from scico.optimize.admm import ADMM, CircularConvolveSolver
 from scico.util import device_info
 
 """

examples/scripts/deconv_microscopy_allchn_tv_admm.py

@@ -14,7 +14,7 @@
 provided by the EPFL Biomedical Imaging Group.
 
 The deconvolution problem is solved using class
-[admm.ADMM](../_autosummary/scico.admm.rst#scico.admm.ADMM) to solve
+[admm.ADMM](../_autosummary/scico.optimize.admm.rst#scico.optimize.admm.ADMM) to solve
 an image deconvolution problem with isotropic total variation (TV)
 regularization
 
@@ -43,7 +43,7 @@
 import ray
 import scico.numpy as snp
 from scico import functional, linop, loss, plot, util
-from scico.admm import ADMM, CircularConvolveSolver
+from scico.optimize.admm import ADMM, CircularConvolveSolver
 
 """
 Define helper functions.

examples/scripts/deconv_microscopy_tv_admm.py

@@ -14,7 +14,7 @@
 provided by the EPFL Biomedical Imaging Group.
 
 The deconvolution problem is solved using class
-[admm.ADMM](../_autosummary/scico.admm.rst#scico.admm.ADMM) to solve
+[admm.ADMM](../_autosummary/scico.optimize.admm.rst#scico.optimize.admm.ADMM) to solve
 an image deconvolution problem with isotropic total variation (TV)
 regularization
 
@@ -40,7 +40,7 @@
 
 import scico.numpy as snp
 from scico import functional, linop, loss, plot, util
-from scico.admm import ADMM, CircularConvolveSolver
+from scico.optimize.admm import ADMM, CircularConvolveSolver
 
 """
 Define helper functions.

examples/scripts/deconv_ppp_bm3d_admm.py

@@ -9,7 +9,7 @@
 =======================================================
 
 This example demonstrates the use of class
-[admm.ADMM](../_autosummary/scico.admm.rst#scico.admm.ADMM) to solve
+[admm.ADMM](../_autosummary/scico.optimize.admm.rst#scico.optimize.admm.ADMM) to solve
 an image deconvolution problem using the Plug-and-Play Priors
 framework :cite:`venkatakrishnan-2013-plugandplay2`, using BM3D
 :cite:`dabov-2008-image` as a denoiser.
@@ -23,7 +23,7 @@
 
 import scico.numpy as snp
 from scico import functional, linop, loss, metric, plot, random
-from scico.admm import ADMM, LinearSubproblemSolver
+from scico.optimize.admm import ADMM, LinearSubproblemSolver
 from scico.util import device_info
 
 """

examples/scripts/deconv_ppp_bm3d_pgm.py

@@ -9,7 +9,7 @@
 ======================================================
 
 This example demonstrates the use of class
-[pgm.AcceleratedPGM](../_autosummary/scico.pgm.rst#scico.pgm.AcceleratedPGM)
+[pgm.AcceleratedPGM](../_autosummary/scico.optimize.pgm.rst#scico.optimize.pgm.AcceleratedPGM)
 to solve an image deconvolution problem using the Plug-and-Play Priors
 framework :cite:`venkatakrishnan-2013-plugandplay2`
 :cite:`kamilov-2017-plugandplay`, using BM3D :cite:`dabov-2008-image`
@@ -25,7 +25,7 @@
 import scico.numpy as snp
 import scico.random
 from scico import functional, linop, loss, metric, plot
-from scico.pgm import AcceleratedPGM
+from scico.optimize.pgm import AcceleratedPGM
 from scico.util import device_info
 
 """

examples/scripts/deconv_ppp_dncnn_admm.py

@@ -9,7 +9,7 @@
 ========================================================
 
 This example demonstrates the use of class
-[admm.ADMM](../_autosummary/scico.admm.rst#scico.admm.ADMM) to solve
+[admm.ADMM](../_autosummary/scico.optimize.admm.rst#scico.optimize.admm.ADMM) to solve
 an image deconvolution problem using the Plug-and-Play Priors
 framework :cite:`venkatakrishnan-2013-plugandplay2`, using DnCNN
 :cite:`zhang-2017-dncnn` as a denoiser.
@@ -24,7 +24,7 @@
 import scico.numpy as snp
 import scico.random
 from scico import functional, linop, loss, metric, plot
-from scico.admm import ADMM, LinearSubproblemSolver
+from scico.optimize.admm import ADMM, LinearSubproblemSolver
 from scico.util import device_info
 
 """

examples/scripts/deconv_tv_admm.py

@@ -9,7 +9,7 @@
 =============================================
 
 This example demonstrates the use of class
-[admm.ADMM](../_autosummary/scico.admm.rst#scico.admm.ADMM) to solve
+[admm.ADMM](../_autosummary/scico.optimize.admm.rst#scico.optimize.admm.ADMM) to solve
 an image deconvolution problem with anisotropic total variation (TV)
 regularization.
 
@@ -28,7 +28,7 @@
 import scico.numpy as snp
 import scico.random
 from scico import functional, linop, loss, metric, plot
-from scico.admm import ADMM, LinearSubproblemSolver
+from scico.optimize.admm import ADMM, LinearSubproblemSolver
 from scico.util import device_info
 
 """

examples/scripts/demosaic_ppp_bm3d_admm.py

@@ -24,8 +24,8 @@
 import scico.numpy as snp
 import scico.random
 from scico import functional, linop, loss, metric, plot
-from scico.admm import ADMM, LinearSubproblemSolver
 from scico.data import kodim23
+from scico.optimize.admm import ADMM, LinearSubproblemSolver
 from scico.util import device_info
 
 """

examples/scripts/denoise_tv_iso_admm.py

@@ -31,7 +31,7 @@
 import scico.numpy as snp
 import scico.random
 from scico import functional, linop, loss, plot
-from scico.admm import ADMM, LinearSubproblemSolver
+from scico.optimize.admm import ADMM, LinearSubproblemSolver
 from scico.util import device_info
 
 """

examples/scripts/denoise_tv_iso_multi.py

@@ -25,9 +25,8 @@
 import scico.numpy as snp
 import scico.random
 from scico import functional, linop, loss, plot
-from scico.admm import ADMM, LinearSubproblemSolver
-from scico.ladmm import LinearizedADMM
-from scico.primaldual import PDHG
+from scico.optimize import PDHG, LinearizedADMM
+from scico.optimize.admm import ADMM, LinearSubproblemSolver
 from scico.util import device_info
 
 """

examples/scripts/denoise_tv_iso_pgm.py

@@ -9,7 +9,7 @@
 ===========================================
 
 This example demonstrates the use of class
-[pgm.AcceleratedPGM](../_autosummary/scico.pgm.rst#scico.pgm.AcceleratedPGM)
+[pgm.AcceleratedPGM](../_autosummary/scico.optimize.pgm.rst#scico.optimize.pgm.AcceleratedPGM)
 to solve isotropic total variation (TV) regularization. It solves the
 denoising problem
 
@@ -40,7 +40,7 @@
 import scico.random
 from scico import functional, linop, loss, operator, plot
 from scico.blockarray import BlockArray
-from scico.pgm import AcceleratedPGM, RobustLineSearchStepSize
+from scico.optimize.pgm import AcceleratedPGM, RobustLineSearchStepSize
 from scico.typing import JaxArray
 from scico.util import device_info, ensure_on_device
 

examples/scripts/sparsecode_admm.py

@@ -9,7 +9,7 @@
 ===========================================
 
 This example demonstrates the use of class
-[admm.ADMM](../_autosummary/scico.admm.rst#scico.admm.ADMM) to solve
+[admm.ADMM](../_autosummary/scico.optimize.admm.rst#scico.optimize.admm.ADMM) to solve
 the non-negative sparse coding problem problem
 
   $$\mathrm{argmin}_{\mathbf{x}} \; \| \mathbf{y} - D \mathbf{x} \|_2^2
@@ -25,7 +25,7 @@
 import jax
 
 from scico import functional, linop, loss, plot
-from scico.admm import ADMM, LinearSubproblemSolver
+from scico.optimize.admm import ADMM, LinearSubproblemSolver
 from scico.util import device_info
 
 """

examples/scripts/sparsecode_pgm.py

@@ -9,7 +9,7 @@
 =========================================
 
 This example demonstrates the use of class
-[pgm.AcceleratedPGM](../_autosummary/scico.pgm.rst#scico.pgm.AcceleratedPGM)
+[pgm.AcceleratedPGM](../_autosummary/scico.optimize.pgm.rst#scico.optimize.pgm.AcceleratedPGM)
 to solve the sparse coding problem problem
 
   $$\mathrm{argmin}_{\mathbf{x}} \; \| \mathbf{y} - D \mathbf{x} \|_2^2
@@ -24,7 +24,7 @@
 import jax
 
 from scico import functional, linop, loss, plot
-from scico.pgm import AcceleratedPGM
+from scico.optimize.pgm import AcceleratedPGM
 from scico.util import device_info
 
 """

examples/scripts/sparsecode_poisson_blkarr_pgm.py

@@ -9,7 +9,7 @@
 =======================================================================
 
 This example demonstrates the use of class
-[pgm.PGMStepSize](../_autosummary/scico.pgm.rst#scico.pgm.PGMStepSize)
+[pgm.PGMStepSize](../_autosummary/scico.optimize.pgm.rst#scico.optimize.pgm.PGMStepSize)
 to solve the non-negative reconstruction problem with Poisson negative
 log likelihood loss
 
@@ -33,7 +33,7 @@
 from scico import functional, loss, plot
 from scico.blockarray import BlockArray
 from scico.operator import Operator
-from scico.pgm import (
+from scico.optimize.pgm import (
     AcceleratedPGM,
     AdaptiveBBStepSize,
     BBStepSize,

examples/scripts/sparsecode_poisson_pgm.py

@@ -9,7 +9,7 @@
 =======================================================================
 
 This example demonstrates the use of class
-[pgm.PGMStepSize](../_autosummary/scico.pgm.rst#scico.pgm.PGMStepSize)
+[pgm.PGMStepSize](../_autosummary/scico.optimize.pgm.rst#scico.optimize.pgm.PGMStepSize)
 to solve the non-negative reconstruction problem with Poisson negative
 log likelihood loss
 
@@ -30,7 +30,7 @@
 import scico.numpy as snp
 import scico.random
 from scico import functional, linop, loss, plot
-from scico.pgm import (
+from scico.optimize.pgm import (
     AcceleratedPGM,
     AdaptiveBBStepSize,
     BBStepSize,

scico/optimize/__init__.py

@@ -0,0 +1,22 @@
+# -*- coding: utf-8 -*-
+# Copyright (C) 2021 by SCICO Developers
+# All rights reserved. BSD 3-clause License.
+# This file is part of the SCICO package. Details of the copyright and
+# user license can be found in the 'LICENSE' file distributed with the
+# package.
+
+"""Optimization algorithms."""
+
+import sys
+
+# isort: off
+from .admm import ADMM
+from ._ladmm import LinearizedADMM
+from .pgm import PGM, AcceleratedPGM
+from ._primaldual import PDHG
+
+__all__ = ["ADMM", "LinearizedADMM", "PGM", "AcceleratedPGM", "PDHG"]
+
+# Imported items in __all__ appear to originate in top-level linop module
+for name in __all__:
+    getattr(sys.modules[__name__], name).__module__ = __name__

scico/admm.py → scico/optimize/admm.py

@@ -283,7 +283,7 @@ def internal_init(self, admm):
                     "CircularConvolveSolver requires f to be a scico.loss.SquaredL2Loss; "
                     f"got {type(admm.f)}"
                 )
-            if not (isinstance(admm.f.A, CircularConvolve) or isinstance(admm.f.A, Identity)):
+            if not isinstance(admm.f.A, (CircularConvolve, Identity)):
                 raise ValueError(
                     "CircularConvolveSolver requires f.A to be a scico.linop.CircularConvolve "
                     f"or scico.linop.Identity; got {type(admm.f.A)}"