* fix markup error in changes.rst

* attempt to clean up some formatting warnings
* add new-style docstrings to emline_fit/utils.py

doc/changes.rst

@@ -17,7 +17,7 @@ Change Log
 .. _`PR #177`: https://github.com/desihub/fastspecfit/pull/177
 .. _`PR #179`: https://github.com/desihub/fastspecfit/pull/179
 .. _`PR #180`: https://github.com/desihub/fastspecfit/pull/180
-`` _`PR #181`: https://github.com/desihub/fastspecfit/pull/181
+.. _`PR #181`: https://github.com/desihub/fastspecfit/pull/181
 
 2.5.2 (2024-04-28)
 ------------------

py/fastspecfit/emline_fit/jacobian.py

@@ -49,21 +49,21 @@ def emline_model_jacobian(line_parameters,
     SQRT_2PI = np.sqrt(2*np.pi)
 
     nbins = len(log_obs_bin_edges) - 1
-    
+
     line_amplitudes, line_vshifts, line_sigmas = \
         np.split(line_parameters, 3)
-    
+
     # buffers for per-parameter calculations, sized large
     # enough for max possible range [s .. e), plus extra
     # padding as directed by caller.
     max_width = max_buffer_width(log_obs_bin_edges, line_sigmas, padding)
-    
+
     nlines = len(line_wavelengths)
     dd     = np.empty((3 * nlines, max_width), dtype=line_amplitudes.dtype)
     endpts = np.zeros((3 * nlines,         2), dtype=np.int32)
 
-    starts = endpts[:,0]
-    ends   = endpts[:,1]
+    starts = endpts[:, 0]
+    ends   = endpts[:, 1]
 
     # compute partial derivatives for avg values of all Gaussians
     # inside each bin. For each Gaussian, we only compute
@@ -94,7 +94,7 @@ def emline_model_jacobian(line_parameters,
         if hi == 0 or lo == len(log_obs_bin_edges): 
             continue
 
-        nedges = hi - lo + 2 # compute values at edges [lo - 1 ... hi]
+        nedges = hi - lo + 2  # compute values at edges [lo - 1 ... hi]
 
         # Compute contribs of each line to each partial derivative in place.
         # No sharing of params between peaks means that we never have to
@@ -180,7 +180,6 @@ def emline_model_jacobian(line_parameters,
     return (endpts, dd)    
 
 
-
 @staticmethod
 @jit(nopython=True, nogil=True)
 def patch_jacobian(obs_bin_centers,

py/fastspecfit/emline_fit/model.py

@@ -10,6 +10,7 @@
     norm_cdf
 )
 
+
 @jit(nopython=True, nogil=True)
 def emline_model(line_wavelengths,
                  line_parameters,
@@ -239,7 +240,7 @@ def emline_model_core(line_wavelength,
 
     # vals[i] --> edge i + lo - 1
 
-    vals[0] = 0. # edge lo - 1
+    vals[0] = 0.  # edge lo - 1
 
     for i in range(1, nedges-1):
 

py/fastspecfit/emline_fit/params_mapping.py

@@ -2,6 +2,7 @@
 
 from numba import jit
 
+
 class ParamsMapping(object):
     """Compute a mapping from the free (line) parameters of a spectrum
     fitting problem to the full set of parameters for the EMLine
@@ -99,11 +100,11 @@ def _mapFreeToFull(freeParms, nParms, sources, factors,
         for j, src_j_free in doubletPatches:
             factors[j] = freeParms[src_j_free] if patchDoublets else 1.
 
-        if fullParms == None:
+        if fullParms is None:
             fullParms = np.empty(nParms, dtype=freeParms.dtype)
 
         for j, src_j_free in enumerate(sources):
-            fullParms[j] = factors[j] # copy fixed zeros
+            fullParms[j] = factors[j]  # copy fixed zeros
             if src_j_free != -1:
                 fullParms[j] *= freeParms[src_j_free]
 
@@ -152,7 +153,7 @@ def _matvec(J_S, v, w):
 
         shape, elts, factors = J_S
 
-        for j in range(shape[0]): # total params
+        for j in range(shape[0]):  # total params
             w[j] = 0.
 
         for i, (dst, src) in enumerate(elts):
@@ -180,9 +181,10 @@ def _add_rmatvec(J_S, v, w):
         for i, (dst, src) in enumerate(elts):
             w[src] += factors[i] * v[dst]
 
-    
+
     ###########################################################
-
+
+
     def _precomputeMapping(self, isFree, pFree,
                            tiedSources, tiedFactors,
                            doubletTargets, doubletSources):
@@ -293,7 +295,7 @@ def _precomputeJacobian(self):
 
             # jacElts already has coefficient (j, p[j]).
             # its factor should be patched from v[ p[src_j] ]
-            jacDoubletPatches[2*i,  :] = (liveOffsets[j], p_src_j)
+            jacDoubletPatches[2*i, :] = (liveOffsets[j], p_src_j)
 
             # add a second coefficient (j, p[src_j]).
             # its factor should be patched from v[ p[j] ]

py/fastspecfit/emline_fit/sparse_rep.py

@@ -34,8 +34,8 @@ class EMLineJacobian(LinearOperator):
 
     """
 
-    def __init__(self, shape, nLineFreeParms, camerapix, jacs, J_S,
-                 J_P = None):       
+    def __init__(self, shape, nLineFreeParms, camerapix, jacs,
+                 J_S, J_P=None):       
         """
         Parameters
         ----------
@@ -63,7 +63,7 @@ def __init__(self, shape, nLineFreeParms, camerapix, jacs, J_S,
             nPatchParms = J_P[0].shape[0]
 
         dtype = jacs[0][1].dtype
-        nParms = jacs[0][1].shape[0] # num line params in full set
+        nParms = jacs[0][1].shape[0]  # num line params in full set
         self.vFull = np.empty(nParms, dtype=dtype)
 
         super().__init__(dtype, shape)
@@ -124,7 +124,7 @@ def _matmat(self, M):
 
         nBins = self.shape[0]
         nVecs = M.shape[1]
-        R = np.empty((nVecs, nBins), dtype=M.dtype) # transpose of result
+        R = np.empty((nVecs, nBins), dtype=M.dtype)  # transpose of result
 
         for i in range(nVecs):
             w = R[i,:]

py/fastspecfit/emline_fit/utils.py

@@ -5,36 +5,37 @@
 
 from fastspecfit.util import C_LIGHT
 
+
 # Do not bother computing normal PDF/CDF if more than this many 
 # standard deviations from mean.
 MAX_SDEV = 5.
 
-#
-# norm_pdf()
-# PDF of standard normal distribution at a point a
-#
-@jit(nopython=True, fastmath=False, nogil=True)
-def norm_pdf(a):
 
+@jit(nopython=True, nogil=True)
+def norm_pdf(a):
+    """
+    PDF of standard normal distribution at a point a
+    """
+
     SQRT_2PI = np.sqrt(2 * np.pi)
 
     return 1/SQRT_2PI * np.exp(-0.5 * a**2)
 
-#
-# norm_cdf()
-# Approximate the integral of a standard normal PDF from -infty to a.
-#
-# Optimization (currently disabled because it is not needed): If
-# |a| > MAX_SDEV, treat the value as extreme and return 0 or 1 as
-# appropriate.
-#
-@jit(nopython=True, fastmath=False, nogil=True)
+
+@jit(nopython=True, nogil=True)
 def norm_cdf(a):
+    """
+    Approximate the integral of a standard normal PDF from -infty to a.
+    """
 
     SQRT1_2 = 1.0 / np.sqrt(2)
 
     z = np.abs(a)
 
+    # Optimization (currently disabled because it is not needed): If
+    # |a| > MAX_SDEV, treat the value as extreme and return 0 or 1 as
+    # appropriate.
+
     #if z > MAX_SDEV: # short-circuit extreme values
     #    if a > 0:
     #        y = 1.
@@ -50,16 +51,25 @@ def norm_cdf(a):
     return y
 
 
-#
-# max_buffer_width()
-# Compute a safe estimate of the number of nonzero bin fluxes possible
-# for a line spanning a subrange of bins with edges log_obs_bin_edges,
-# assuming the line's width is one of the values in line_sigmas.
-# Optionally add 2*padding to allow future expansion to left and right.
-#
-@jit(nopython=True, fastmath=False, nogil=True)
+@jit(nopython=True, nogil=True)
 def max_buffer_width(log_obs_bin_edges, line_sigmas, padding=0):
+    """
+    Compute a safe estimate of the number of nonzero bin fluxes possible
+    for a line spanning a subrange of bins with edges log_obs_bin_edges,
+    assuming the line's width is one of the values in line_sigmas.
+    Optionally add 2*padding to allow future expansion to left and right.
 
+    Parameters
+    ----------
+    log_obs_bin_edges : :class:`np.ndarray` [# obs wavelength bins + 1]
+      log of wavelengths of all observed bin edges.
+    line_sigmas : :class:`np.ndarray`  [# nlines]
+      Gaussian widths of all spectral lines.
+    padding : :class:`int`
+      Padding parameter to add to width for future use.
+
+    """
+
     # Find the largest width sigma for any line, and
     # allocate enough space for twice that much width
     # in bins, given the smallest observed bin width.