plotastrodata 1.9.19.post1__tar.gz → 1.9.21__tar.gz

{plotastrodata-1.9.19.post1/plotastrodata.egg-info → plotastrodata-1.9.21}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: plotastrodata
-Version: 1.9.19.post1
+Version: 1.9.21
 Summary: plotastrodata is a tool for astronomers to create figures from FITS files and perform fundamental data analyses with ease.
 Home-page: https://github.com/yusukeaso-astron/plotastrodata
 Download-URL: https://github.com/yusukeaso-astron/plotastrodata
@@ -14,7 +14,7 @@ License-File: LICENSE
 Requires-Dist: astropy>=7.2
 Requires-Dist: corner
 Requires-Dist: dynesty
-Requires-Dist: emcee
+Requires-Dist: emcee>=3.0
 Requires-Dist: matplotlib
 Requires-Dist: numpy>=2.0
 Requires-Dist: pillow
@@ -84,7 +84,7 @@ plotastrodata can do the following things.
 * astropy >= 7.2
 * corner (only for fitting)
 * dynesty (only for fitting)
-* emcee (only for fitting)
+* emcee >= 3.0 (only for fitting)
 * ffmpeg (only for movie)
 * matplotlib
 * numpy >= 2.0

{plotastrodata-1.9.19.post1 → plotastrodata-1.9.21}/README.md

@@ -56,7 +56,7 @@ plotastrodata can do the following things.
 * astropy >= 7.2
 * corner (only for fitting)
 * dynesty (only for fitting)
-* emcee (only for fitting)
+* emcee >= 3.0 (only for fitting)
 * ffmpeg (only for movie)
 * matplotlib
 * numpy >= 2.0

{plotastrodata-1.9.19.post1 → plotastrodata-1.9.21}/plotastrodata/__init__.py

@@ -1,4 +1,4 @@
 import warnings
 
 warnings.simplefilter('ignore', FutureWarning)
-__version__ = '1.9.19-1'
+__version__ = '1.9.21'

{plotastrodata-1.9.19.post1 → plotastrodata-1.9.21}/plotastrodata/fitting_utils.py

@@ -2,9 +2,11 @@ import corner
 import emcee
 import matplotlib.pyplot as plt
 import numpy as np
+import pickle
 import ptemcee
 import warnings
 from dynesty import DynamicNestedSampler as DNS
+from functools import partial
 from multiprocessing import Pool
 from tqdm import tqdm
 from typing import Any, Callable
@@ -18,6 +20,25 @@ bar = None
 global_progressbar = True
 
 
+def _gaussian_log_likelihood(x: np.ndarray, model: Callable,
+                             xdata: np.ndarray, ydata: np.ndarray,
+                             sigma: float | np.ndarray) -> float:
+    """Return a Gaussian log likelihood using pickleable arguments."""
+    chi2 = np.sum((ydata - model(xdata, *x))**2 / sigma**2)
+    return chi2 / (-2)
+
+
+def _bounded_log_probability(x: np.ndarray, log_likelihood: Callable,
+                             bounds: np.ndarray,
+                             update_progress: bool = False) -> float:
+    """Combine a bounded uniform prior with a log likelihood."""
+    if update_progress:
+        bar.update(1)
+    if np.all((bounds[:, 0] < x) & (x < bounds[:, 1])):
+        return log_likelihood(x)
+    return -np.inf
+
+
 def logp(x: np.ndarray) -> float:
     """Log prior function made from the boundary (global_bounds) of fitting parameters.
 
@@ -38,9 +59,9 @@ def logp(x: np.ndarray) -> float:
 def _get_GR(samples: np.ndarray, nwalkers: int, ndata: int, dim: int
             ) -> np.ndarray:
     """Calculate the Gelman-Rubin statistics."""
-    B = np.std(np.mean(samples, axis=1), axis=0)
-    W = np.mean(np.std(samples, axis=1), axis=0)
-    V = (len(samples[0]) - 1) / len(samples[0]) * W \
+    B = np.std(np.mean(samples, axis=0), axis=0)
+    W = np.mean(np.std(samples, axis=0), axis=0)
+    V = (len(samples) - 1) / len(samples) * W \
         + (nwalkers + 1) / (nwalkers - 1) * B
     d = ndata - dim - 1
     GR = np.sqrt((d + 3) / (d + 1) * V / W)
@@ -66,7 +87,7 @@ class EmceeCorner():
 
     This class wraps ``emcee`` and ``ptemcee`` for simple bounded-parameter fitting.  The likelihood can be supplied directly through ``logl``, or it can be constructed from ``model``, ``xdata``, ``ydata``, and ``sigma``. Parameters are sampled with a uniform prior inside ``bounds`` and zero prior probability outside them.
 
-    After calling :meth:`fit`, the main results are stored as attributes:``samples`` for the post-burn-in chain, ``popt`` for the maximum-likelihood parameter set, and ``plow``, ``pmid``, and ``phigh`` for posterior percentiles.  The samples can be visualized with :meth:`plotcorner` and :meth:`plotchain`.
+    After calling :meth:`fit`, the main results are stored as attributes:``samples`` for the post-burn-in chain, ``popt`` for the maximum-likelihood parameter set, and ``plow``, ``pmid``, and ``phigh`` for posterior percentiles. ``samples`` has the shape ``(steps, walkers, dimensions)``. The samples can be visualized with :meth:`plotcorner` and :meth:`plotchain`.
 
     Args:
         bounds (np.ndarray): Parameter bounds with shape ``(dim, 2)``. logl (Callable, optional): Log-likelihood function. Defaults to None.
@@ -95,9 +116,8 @@ class EmceeCorner():
         if logl is None and (model is not None
                              and xdata is not None
                              and ydata is not None):
-            def logl(x: np.ndarray) -> float:
-                chi2 = np.sum((ydata - model(xdata, *x))**2 / sigma**2)
-                return chi2 / (-2)
+            logl = partial(_gaussian_log_likelihood, model=model,
+                           xdata=xdata, ydata=ydata, sigma=sigma)
         self.bounds = global_bounds
         self.dim = len(self.bounds)
         self.logl = logl
@@ -125,43 +145,50 @@ class EmceeCorner():
                               'logl': self.logl, 'logp': self.logp}
         else:
             if ncores > 1:
-                print('Use logl as log_prob_fn to avoid function-in-function.')
-                log_prob_fn = self.logl
-            else:
-                def log_prob_fn(x: np.ndarray) -> float:
-                    return self.logp(x) + self.logl(x)
+                try:
+                    pickle.dumps(self.logl)
+                except (pickle.PicklingError, AttributeError, TypeError) as exc:
+                    raise TypeError(
+                        'logl and model must be pickleable when ncores > 1. '
+                        'Define them at module scope instead of inside another '
+                        'function.') from exc
 
             sampler_cls = emcee.EnsembleSampler
             sampler_kwargs = {'nwalkers': nwalkers, 'ndim': self.dim,
-                              'log_prob_fn': log_prob_fn}
+                              'log_prob_fn': _bounded_log_probability,
+                              'args': (self.logl, self.bounds,
+                                       global_progressbar and ncores == 1)}
         if ncores > 1:
             with Pool(ncores) as pool:
                 sampler = sampler_cls(**sampler_kwargs, pool=pool)
+                # This run_mcmc is duplicated so sampling finishes before the pool closes.
+                sampler.run_mcmc(pos0, nsteps)
         else:
             sampler = sampler_cls(**sampler_kwargs, pool=None)
-        sampler.run_mcmc(pos0, nsteps)
+            sampler.run_mcmc(pos0, nsteps)
         return sampler
 
     def _get_samples(self, sampler: Any, nburnin: int,
                      pt: bool) -> np.ndarray:
         """Extract post-burn-in samples from sampler chain."""
         if pt:
-            return sampler.chain[0, :, nburnin:, :]  # temperatures, walkers, steps, dim
+            chain = sampler.chain[0]  # walkers, steps, dim
+            return np.swapaxes(chain, 0, 1)[nburnin:, :, :]  # steps, walkers, dim
         else:
-            return sampler.chain[:, nburnin:, :]  # walkers, steps, dim
+            return sampler.get_chain(discard=nburnin)  # steps, walkers, dim
 
     def _get_lnp_popt(self, sampler: Any, pt: bool, nburnin: int,
                       ) -> tuple[np.ndarray, np.ndarray]:
         """Get log probabilities and best-fit parameters from sampler."""
         if pt:
-            lnp = sampler.logprobability[0]  # 0th temperature chain
-            chain = sampler.chain[0]
+            lnp = np.swapaxes(sampler.logprobability[0], 0, 1)
+            chain = np.swapaxes(sampler.chain[0], 0, 1)
         else:
-            lnp = sampler.lnprobability
-            chain = sampler.chain
+            lnp = sampler.get_log_prob()  # steps, walkers
+            chain = sampler.get_chain()  # steps, walkers, dim
         idx_best = np.unravel_index(np.argmax(lnp), lnp.shape)
         popt = chain[idx_best]
-        lnp = lnp[:, nburnin:]
+        lnp = lnp[nburnin:, :]
         return lnp, popt
 
     def _get_percentiles(self, samples: np.ndarray
@@ -189,7 +216,7 @@ class EmceeCorner():
             ntry (int, optional): Number of trials for the Gelman-Rubin check. Defaults to 1.
             pos0 (np.nparray, optional): Initial parameter set in the shape of (ntemps, nwalkers, dim). Defaults to None.
             savechain (str, optional): File name of the chain in format of .npy. Existing files with the same name are overwritten by ``numpy.save``. Defaults to None.
-            ncores (int, optional): Number of cores for multiprocessing.Pool. ncores=1 does not use multiprocessing. Defaults to 1.
+            ncores (int, optional): Number of cores for multiprocessing.Pool. ncores=1 does not use multiprocessing. For ncores > 1, user-supplied logl and model functions must be pickleable, such as functions defined at module scope. Defaults to 1.
             grcheck (bool, optional): Whether to check Gelman-Rubin statistics. Defaults to False.
             pt (bool, optional): Whether to use ptemcee; otherwise, emcee is used. Defaults to False.
         """
@@ -242,13 +269,14 @@ class EmceeCorner():
             labels = [f'Par {i:d}' for i in range(self.dim)]
         if cornerrange is None:
             cornerrange = self.bounds
-        corner.corner(np.reshape(self.samples, (-1, self.dim)),
-                      truths=self.popt,
-                      quantiles=[self.percent[0] / 100,
-                                 0.5,
-                                 self.percent[1] / 100],
-                      show_titles=True, labels=labels, range=cornerrange)
-        close_figure(plt, savefig, show, tight=False)
+        fig = corner.corner(np.reshape(self.samples, (-1, self.dim)),
+                            truths=self.popt,
+                            quantiles=[self.percent[0] / 100,
+                                       0.5,
+                                       self.percent[1] / 100],
+                            show_titles=True, labels=labels,
+                            range=cornerrange)
+        close_figure(fig, savefig, show, tight=False)
 
     def plotchain(self, labels: list | None = None, ylim: list | None = None,
                   savefig: dict | str | None = None,
@@ -266,14 +294,14 @@ class EmceeCorner():
         if ylim is None:
             ylim = self.bounds
         fig = plt.figure(figsize=(4, 2 * self.dim))
-        x = np.arange(np.shape(self.samples)[1])
+        x = np.arange(np.shape(self.samples)[0])
         naverage = max(1, len(x) // 100)
         nend = len(x) - len(x) % 100 if naverage > 1 else len(x)
         x = x[:nend:naverage]
         for i in range(self.dim):
-            y = self.samples[:, :, i]  # walkers, steps, dim
+            y = self.samples[:, :, i]  # steps, walkers, dim
             plist = [self.percent[0], 50, self.percent[1]]
-            y = [np.percentile(y, p, axis=0) for p in plist]  # percent over the walkers, steps
+            y = [np.percentile(y, p, axis=1) for p in plist]  # percent over the walkers, steps
             y = [[np.percentile(np.reshape(yy[:nend], (naverage, -1)), p, axis=0)
                   for p in plist] for yy in y]  # percent over the walkers, percent over the steps
             ax = fig.add_subplot(self.dim, 1, i + 1)

{plotastrodata-1.9.19.post1 → plotastrodata-1.9.21}/plotastrodata/other_utils.py

@@ -296,7 +296,7 @@ def close_figure(fig: object, savefig: dict | str | None = None,
                  show: bool = False, tight: bool = True) -> None:
     """Save, show, and close the figure.
 
-    If ``savefig`` is provided, the figure is saved with Matplotlib ``Figure.savefig``. Existing files with the same name are overwritten by Matplotlib. After optional saving/showing, the figure is closed with ``plt.close()``.
+    If ``savefig`` is provided, the figure is saved with Matplotlib ``Figure.savefig``. Existing files with the same name are overwritten by Matplotlib. After optional saving/showing, the figure is closed with ``plt.close(fig)``.
 
     Default keyword values:
         Figure.savefig: ``bbox_inches='tight'`` and ``transparent=True``. Values in ``savefig`` override these defaults.
@@ -316,4 +316,4 @@ def close_figure(fig: object, savefig: dict | str | None = None,
         fig.savefig(**savefig0)
     if show:
         plt.show()
-    plt.close()
+    plt.close(fig)

{plotastrodata-1.9.19.post1 → plotastrodata-1.9.21}/plotastrodata/plot_utils.py

@@ -564,13 +564,18 @@ class PlotAstroData(AstroFrame):
                               figsize=figsize,
                               ncols=ncols, nrows=nrows, nchan=nchan)
         need_vlabel = nchan > 1 or animation
+        figs = []
         for ch in range(nchan):
             n, i, j = ch2nij(ch)
-            if internalfig and n not in plt.get_fignums():
-                fig = plt.figure(n, figsize=figsize)
+            if n == len(figs):
+                if internalfig:
+                    fig = plt.figure(figsize=figsize)
+                figs.append(fig)
                 if need_vlabel:
                     fig.subplots_adjust(hspace=0, wspace=0,
                                         right=0.87, top=0.87)
+            else:
+                fig = figs[n]
             if internalax:
                 sharex = ax[nij2ch(n, i - 1, j)] if i > 0 else None
                 sharey = ax[nij2ch(n, i, j - 1)] if j > 0 else None
@@ -582,6 +587,7 @@ class PlotAstroData(AstroFrame):
                             rf'${vlabel:.{veldigit}f}$', color='black',
                             backgroundcolor='white', zorder=20)
         self.fig = None if internalfig else fig
+        self.figs = figs
         self.ax = ax
         self.rowcol = nrows * ncols
         self.npages = npages
@@ -657,8 +663,6 @@ class PlotAstroData(AstroFrame):
             for ch, axnow in enumerate(self.ax):
                 if ch not in self._validchan(include_chan):
                     continue
-                if self.fig is None:
-                    plt.figure(ch // self.rowcol)
                 if patch == 'rectangle':
                     a = np.radians(angle)
                     xp = x - (width*np.cos(a) + height*np.sin(a)) / 2.
@@ -857,16 +861,13 @@ class PlotAstroData(AstroFrame):
         if not show_cbar:
             return
 
-        if self.fig is None:
-            fig = plt.figure(ch // self.rowcol)
-        else:
-            fig = self.fig
+        fig = self.figs[ch // self.rowcol]
         if len(self.ax) == 1:
             ax = self.ax[ch]
             cb = fig.colorbar(mappable[ch], ax=ax, label=cblabel,
                               format=cbformat, location=cblocation)
         else:
-            cax = plt.axes([0.88, 0.105, 0.015, 0.77])
+            cax = fig.add_axes([0.88, 0.105, 0.015, 0.77])
             cb = fig.colorbar(mappable[ch], cax=cax, label=cblabel,
                               format=cbformat)
         cb.ax.tick_params(labelsize=cbtickfontsize)
@@ -1104,14 +1105,13 @@ class PlotAstroData(AstroFrame):
                 axnow.set_ylabel('')
             if len(self.ax) == 1:
                 if self.fig is None:
-                    plt.figure(0).tight_layout()
+                    self.figs[0].tight_layout()
         if title is not None:
             if len(self.ax) > 1:
                 t = {'y': 0.9}
                 t_in = {'t': title} if isinstance(title, str) else title
                 t.update(t_in)
-                for i in range(self.npages):
-                    fig = plt.figure(i)
+                for fig in self.figs:
                     fig.suptitle(**t)
             else:
                 t = {'label': title} if isinstance(title, str) else title
@@ -1249,13 +1249,13 @@ class PlotAstroData(AstroFrame):
                 show: bool = False, **kwargs: Any) -> None:
         """Use savefig of matplotlib.
 
-        If ``filename`` is provided, existing files with the same name are overwritten by Matplotlib. This method closes all Matplotlib figures with ``plt.close('all')`` after optional saving/showing.
+        If ``filename`` is provided, existing files with the same name are overwritten by Matplotlib. After optional saving/showing, figures managed by this instance are closed.
 
         Default keyword values:
             Figure.savefig: ``transparent=True`` and ``bbox_inches='tight'``. User-supplied keyword arguments override these values.
 
         Args:
-            filename (str, optional): Output image file name. Existing files may be overwritten, and all Matplotlib figures are closed after saving/showing. Defaults to None.
+            filename (str, optional): Output image file name. Existing files may be overwritten, and figures managed by this instance are closed after saving/showing. Defaults to None.
             show (bool, optional): True means doing plt.show(). Defaults to False.
         """
         _kw = {'transparent': True, 'bbox_inches': 'tight'}
@@ -1265,15 +1265,15 @@ class PlotAstroData(AstroFrame):
             axnow.set_ylim(*self.Ylim)
         if isinstance(filename, str):
             ext = filename.split('.')[-1]
-            for i in range(self.npages):
+            for i, fig in enumerate(self.figs):
                 ver = '' if self.npages == 1 else f'_{i:d}'
-                fig = plt.figure(i)
                 fig.patch.set_alpha(0)
                 fname = filename.replace(f'.{ext}', f'{ver}.{ext}')
                 fig.savefig(fname, **_kw)
         if show:
             plt.show()
-        plt.close('all')
+        for fig in self.figs:
+            plt.close(fig)
 
     def get_figax(self) -> tuple[object, object] | None:
         """Output the external fig and ax after plotting.
@@ -1286,8 +1286,7 @@ class PlotAstroData(AstroFrame):
                   + ' with channel maps')
             return
 
-        fig = plt.figure(0) if self.fig is None else self.fig
-        return fig, self.ax[0]
+        return self.figs[0], self.ax[0]
 
 
 def _get_ylabel_profile(_kw: dict, Tb: bool, flux: bool, bunit: str

{plotastrodata-1.9.19.post1 → plotastrodata-1.9.21/plotastrodata.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: plotastrodata
-Version: 1.9.19.post1
+Version: 1.9.21
 Summary: plotastrodata is a tool for astronomers to create figures from FITS files and perform fundamental data analyses with ease.
 Home-page: https://github.com/yusukeaso-astron/plotastrodata
 Download-URL: https://github.com/yusukeaso-astron/plotastrodata
@@ -14,7 +14,7 @@ License-File: LICENSE
 Requires-Dist: astropy>=7.2
 Requires-Dist: corner
 Requires-Dist: dynesty
-Requires-Dist: emcee
+Requires-Dist: emcee>=3.0
 Requires-Dist: matplotlib
 Requires-Dist: numpy>=2.0
 Requires-Dist: pillow
@@ -84,7 +84,7 @@ plotastrodata can do the following things.
 * astropy >= 7.2
 * corner (only for fitting)
 * dynesty (only for fitting)
-* emcee (only for fitting)
+* emcee >= 3.0 (only for fitting)
 * ffmpeg (only for movie)
 * matplotlib
 * numpy >= 2.0

{plotastrodata-1.9.19.post1 → plotastrodata-1.9.21}/plotastrodata.egg-info/requires.txt

@@ -1,7 +1,7 @@
 astropy>=7.2
 corner
 dynesty
-emcee
+emcee>=3.0
 matplotlib
 numpy>=2.0
 pillow

{plotastrodata-1.9.19.post1 → plotastrodata-1.9.21}/setup.cfg

@@ -19,7 +19,7 @@ install_requires =
 	astropy >= 7.2
 	corner
 	dynesty
-	emcee
+	emcee >= 3.0
 	matplotlib
 	numpy >= 2.0
 	pillow