Miscellaneous changes (#475)

* Fix error in estimate_parameters call and improve penalty parameter value

* Add flexibility in iteration counting and display

* Update list of contributors

* Typo fix

* Update submodule

* Expand projector documentation

* Update submodule

---------

Co-authored-by: Michael-T-McCann <[email protected]>
Co-authored-by: Brendt Wohlberg <[email protected]>

docs/source/team.rst

@@ -27,3 +27,5 @@ Contributors
 - `Saurav Maheshkar <https://github.com/SauravMaheshkar>`_ (Improvements to pre-commit configuration)
 - `Yanpeng Yuan <https://github.com/yanpeng7>`_ (ASTRA interface improvements)
 - `Li-Ta (Ollie) Lo <https://github.com/ollielo>`_ (ASTRA interface improvements)
+- `Renat Sibgatulin <https://github.com/Sibgatulin>`_ (Docs corrections)
+- `Salman Naqvi <https://github.com/shnaqvi>`_ (Contributions to approximate TV norm prox and proximal average implementation)

examples/scripts/deconv_tv_padmm.py

@@ -101,9 +101,9 @@
 """
 Set up a proximal ADMM solver object.
 """
-ρ = 1.0e-1  # ADMM penalty parameter
+ρ = 5.0e-2  # ADMM penalty parameter
 maxiter = 50  # number of ADMM iterations
-mu, nu = ProximalADMM.estimate_parameters(D)
+mu, nu = ProximalADMM.estimate_parameters(A)
 
 solver = ProximalADMM(
     f=f,

examples/scripts/denoise_tv_admm.py

@@ -80,7 +80,7 @@
 """
 Denoise with anisotropic total variation for comparison.
 """
-# Tune the weight to give the same data fidelty as the isotropic case.
+# Tune the weight to give the same data fidelity as the isotropic case.
 λ_aniso = 1.2e0
 g_aniso = λ_aniso * functional.L1Norm()
 

examples/scripts/denoise_tv_apgm.py

@@ -166,7 +166,7 @@ def prox(self, v: Array, lam: float, **kwargs) -> Array:
 
 """
 Use RobustLineSearchStepSize object and set up AcceleratedPGM solver
-object. Weight was tuned to give the same data fidelty as the
+object. Weight was tuned to give the same data fidelity as the
 isotropic case. Run the solver.
 """
 

scico/diagnostics.py

@@ -26,6 +26,7 @@ def __init__(
         ident: Optional[dict] = None,
         display: bool = False,
         period: int = 1,
+        shift_cycles: bool = True,
         overwrite: bool = True,
         colsep: int = 2,
     ):
@@ -48,14 +49,17 @@ def __init__(
             fields: A dictionary associating field names with format
                 strings for displaying the corresponding values.
             ident: A dictionary associating field names.
-               with corresponding valid identifiers for use within the
-               namedtuple used to record results. Defaults to ``None``.
+                with corresponding valid identifiers for use within the
+                namedtuple used to record results. Defaults to ``None``.
             display: Flag indicating whether results should be printed
                 to stdout. Defaults to ``False``.
             period: Only display one result in every cycle of length
                 `period`.
+            shift_cycles: If ``True``, apply an offset to the iteration
+                count so that display cycles end at 0, `period` - 1, etc.
+                Otherwise, cycles end at `period`, 2 * `period`, etc.
             overwrite: If ``True``, display all results, but each one
-                 overwrites the next, except for one result per cycle.
+                overwrites the next, except for one result per cycle.
             colsep: Number of spaces seperating fields in displayed
                 tables. Defaults to 2.
 
@@ -69,6 +73,8 @@ def __init__(
             raise TypeError("Parameter fields must be an instance of dict.")
         # Subsampling rate of results that are to be displayed
         self.period: int = period
+        # Offset to iteration count for determining start of period
+        self.period_offset = 1 if shift_cycles else 0
         # Flag indicating whether to display and overwrite, or not display at all
         self.overwrite: bool = overwrite
         # Number of spaces seperating fields in displayed tables
@@ -159,13 +165,13 @@ def insert(self, values: Union[List, Tuple]):
                 print(self.disphdr)
                 self.disphdr = None
             if self.overwrite:
-                if (len(self.iterations) - 1) % self.period == 0:
+                if (len(self.iterations) - self.period_offset) % self.period == 0:
                     end = "\n"
                 else:
                     end = "\r"
                 print((" " * self.colsep).join(self.fieldformat) % values, end=end)
             else:
-                if (len(self.iterations) - 1) % self.period == 0:
+                if (len(self.iterations) - self.period_offset) % self.period == 0:
                     print((" " * self.colsep).join(self.fieldformat) % values)
 
     def end(self):
@@ -180,7 +186,7 @@ def end(self):
             self.display
             and self.overwrite
             and self.period > 1
-            and (len(self.iterations) - 1) % self.period
+            and (len(self.iterations) - self.period_offset) % self.period
         ):
             print()
 

scico/linop/xray/_xray.py

@@ -52,6 +52,19 @@ def __init__(self, projector):
 class Parallel2dProjector:
     """Parallel ray, single axis, 2D X-ray projector.
 
+    This implementation approximates the projection of each rectangular
+    pixel as a boxcar function (whereas the exact projection is a
+    trapezoid). Detector pixels are modeled as bins (rather than points)
+    and this approximation allows fast calculation of the contribution
+    of each pixel to each bin because the integral of the boxcar is
+    simple.
+
+    By requiring the side length of the pixels to be less than or equal
+    to the bin width (which is assumed to be 1.0), we ensure that each
+    pixel contributes to at most two bins, which accelerates the
+    accumulation of pixel values into bins (equivalently, makes the
+    linear operator sparse).
+
     `x0`, `dx`, and `y0` should be expressed in units such that the
     detector spacing `dy` is 1.0.
     """

scico/optimize/_common.py

@@ -82,12 +82,17 @@ def __init__(self, **kwargs: Any):
         Args:
             **kwargs: Optional parameter dict. Valid keys are:
 
+                iter0:
+                  Initial value of iteration counter. Default value is 0.
+
                 maxiter:
-                  Maximum iterations on call to :meth:`solve`.
+                  Maximum iterations on call to :meth:`solve`. Default
+                  value is 100.
 
                 nanstop:
                   If ``True``, stop iterations if a ``NaN`` or ``Inf``
                   value is encountered in a solver working variable.
+                  Default value is ``False``.
 
                 itstat_options:
                   A dict of named parameters to be passed to
@@ -102,14 +107,15 @@ def __init__(self, **kwargs: Any):
                   otherwise the default dict is updated with the dict
                   specified by this parameter.
         """
+        iter0 = kwargs.pop("iter0", 0)
         self.maxiter: int = kwargs.pop("maxiter", 100)
         self.nanstop: bool = kwargs.pop("nanstop", False)
         itstat_options = kwargs.pop("itstat_options", None)
 
         if kwargs:
             raise TypeError(f"Unrecognized keyword argument(s) {', '.join([k for k in kwargs])}")
 
-        self.itnum: int = 0
+        self.itnum: int = iter0
         self.timer: Timer = Timer()
 
         itstat_fields, itstat_attrib = self._itstat_default_fields()