from asap.parameters import rcParams from asap.selector import selector from asap.scantable import scantable from asap.logging import asaplog, asaplog_post_dec import matplotlib.axes from matplotlib.font_manager import FontProperties from matplotlib.text import Text import re class asapplotter: """ The ASAP plotter. By default the plotter is set up to plot polarisations 'colour stacked' and scantables across panels. .. note:: Currenly it only plots 'spectra' not Tsys or other variables. """ def __init__(self, visible=None , **kwargs): self._visible = rcParams['plotter.gui'] if visible is not None: self._visible = visible self._plotter = self._newplotter(**kwargs) # additional tool bar self._plotter.figmgr.casabar=self._newcasabar() self._panelling = None self._stacking = None self.set_panelling() self.set_stacking() self._rows = None self._cols = None self._autoplot = False self._minmaxx = None self._minmaxy = None self._datamask = None self._data = None self._lmap = None self._title = None self._ordinate = None self._abcissa = None self._abcunit = None self._usermask = [] self._maskselection = None self._selection = selector() self._hist = rcParams['plotter.histogram'] self._fp = FontProperties() self._margins = self.set_margin(refresh=False) self._offset = None self._startrow = 0 self._ipanel = -1 self._panelrows = [] self._headtext={'string': None, 'textobj': None} def _translate(self, instr): keys = "s b i p t r".split() if isinstance(instr, str): for key in keys: if instr.lower().startswith(key): return key return None def _newplotter(self, **kwargs): backend=matplotlib.get_backend() if not self._visible: from asap.asaplot import asaplot elif backend == 'TkAgg': from asap.asaplotgui import asaplotgui as asaplot elif backend == 'Qt4Agg': from asap.asaplotgui_qt4 import asaplotgui as asaplot elif backend == 'GTkAgg': from asap.asaplotgui_gtk import asaplotgui as asaplot else: from asap.asaplot import asaplot return asaplot(**kwargs) def _newcasabar(self): backend=matplotlib.get_backend() if self._visible and backend == "TkAgg": from asap.casatoolbar import CustomToolbarTkAgg return CustomToolbarTkAgg(self) #from asap.casatoolbar import CustomFlagToolbarTkAgg #return CustomFlagToolbarTkAgg(self) return None @asaplog_post_dec def plot(self, scan=None): """ Plot a scantable. Parameters: scan: a scantable Note: If a scantable was specified in a previous call to plot, no argument has to be given to 'replot' NO checking is done that the abcissas of the scantable are consistent e.g. all 'channel' or all 'velocity' etc. """ self._startrow = 0 self._ipanel = -1 self._reset_header() if self._plotter.is_dead: if hasattr(self._plotter.figmgr,'casabar'): del self._plotter.figmgr.casabar self._plotter = self._newplotter() self._plotter.figmgr.casabar=self._newcasabar() if self._plotter.figmgr.casabar: self._plotter.figmgr.casabar.set_pagecounter(1) self._panelrows = [] self._plotter.hold() #self._plotter.clear() if not self._data and not scan: msg = "Input is not a scantable" raise TypeError(msg) if scan: self.set_data(scan, refresh=False) self._plot(self._data) if self._minmaxy is not None: self._plotter.set_limits(ylim=self._minmaxy) if self._plotter.figmgr.casabar: self._plotter.figmgr.casabar.enable_button() self._plotter.release() self._plotter.tidy() self._plotter.show(hardrefresh=False) return def gca(self): return self._plotter.figure.gca() def refresh(self): """Do a soft refresh""" self._plotter.figure.show() def create_mask(self, nwin=1, panel=0, color=None): """ Interactively define a mask. It retruns a mask that is equivalent to the one created manually with scantable.create_mask. Parameters: nwin: The number of mask windows to create interactively default is 1. panel: Which panel to use for mask selection. This is useful if different IFs are spread over panels (default 0) """ if self._data is None: return [] outmask = [] self._plotter.subplot(panel) xmin, xmax = self._plotter.axes.get_xlim() marg = 0.05*(xmax-xmin) self._plotter.axes.set_xlim(xmin-marg, xmax+marg) self.refresh() def cleanup(lines=False, texts=False, refresh=False): if lines: del self._plotter.axes.lines[-1] if texts: del self._plotter.axes.texts[-1] if refresh: self.refresh() for w in xrange(nwin): wpos = [] self.text(0.05,1.0, "Add start boundary", coords="relative", fontsize=10) point = self._plotter.get_point() cleanup(texts=True) if point is None: continue wpos.append(point[0]) self.axvline(wpos[0], color=color) self.text(0.05,1.0, "Add end boundary", coords="relative", fontsize=10) point = self._plotter.get_point() cleanup(texts=True, lines=True) if point is None: self.refresh() continue wpos.append(point[0]) self.axvspan(wpos[0], wpos[1], alpha=0.1, edgecolor=color, facecolor=color) ymin, ymax = self._plotter.axes.get_ylim() outmask.append(wpos) self._plotter.axes.set_xlim(xmin, xmax) self.refresh() if len(outmask) > 0: return self._data.create_mask(*outmask) return [] # forwards to matplotlib axes def text(self, *args, **kwargs): if kwargs.has_key("interactive"): if kwargs.pop("interactive"): pos = self._plotter.get_point() args = tuple(pos)+args self._axes_callback("text", *args, **kwargs) text.__doc__ = matplotlib.axes.Axes.text.__doc__ def arrow(self, *args, **kwargs): if kwargs.has_key("interactive"): if kwargs.pop("interactive"): pos = self._plotter.get_region() dpos = (pos[0][0], pos[0][1], pos[1][0]-pos[0][0], pos[1][1] - pos[0][1]) args = dpos + args self._axes_callback("arrow", *args, **kwargs) arrow.__doc__ = matplotlib.axes.Axes.arrow.__doc__ def annotate(self, text, xy=None, xytext=None, **kwargs): if kwargs.has_key("interactive"): if kwargs.pop("interactive"): xy = self._plotter.get_point() xytext = self._plotter.get_point() if not kwargs.has_key("arrowprops"): kwargs["arrowprops"] = dict(arrowstyle="->") self._axes_callback("annotate", text, xy, xytext, **kwargs) annotate.__doc__ = matplotlib.axes.Axes.annotate.__doc__ def axvline(self, *args, **kwargs): if kwargs.has_key("interactive"): if kwargs.pop("interactive"): pos = self._plotter.get_point() args = (pos[0],)+args self._axes_callback("axvline", *args, **kwargs) axvline.__doc__ = matplotlib.axes.Axes.axvline.__doc__ def axhline(self, *args, **kwargs): if kwargs.has_key("interactive"): if kwargs.pop("interactive"): pos = self._plotter.get_point() args = (pos[1],)+args self._axes_callback("axhline", *args, **kwargs) axhline.__doc__ = matplotlib.axes.Axes.axhline.__doc__ def axvspan(self, *args, **kwargs): if kwargs.has_key("interactive"): if kwargs.pop("interactive"): pos = self._plotter.get_region() dpos = (pos[0][0], pos[1][0]) args = dpos + args self._axes_callback("axvspan", *args, **kwargs) # hack to preventy mpl from redrawing the patch # it seem to convert the patch into lines on every draw. # This doesn't happen in a test script??? #del self._plotter.axes.patches[-1] axvspan.__doc__ = matplotlib.axes.Axes.axvspan.__doc__ def axhspan(self, *args, **kwargs): if kwargs.has_key("interactive"): if kwargs.pop("interactive"): pos = self._plotter.get_region() dpos = (pos[0][1], pos[1][1]) args = dpos + args self._axes_callback("axhspan", *args, **kwargs) # hack to preventy mpl from redrawing the patch # it seem to convert the patch into lines on every draw. # This doesn't happen in a test script??? #del self._plotter.axes.patches[-1] axhspan.__doc__ = matplotlib.axes.Axes.axhspan.__doc__ def _axes_callback(self, axesfunc, *args, **kwargs): panel = 0 if kwargs.has_key("panel"): panel = kwargs.pop("panel") coords = None if kwargs.has_key("coords"): coords = kwargs.pop("coords") if coords.lower() == 'world': kwargs["transform"] = self._plotter.axes.transData elif coords.lower() == 'relative': kwargs["transform"] = self._plotter.axes.transAxes self._plotter.subplot(panel) self._plotter.axes.set_autoscale_on(False) getattr(self._plotter.axes, axesfunc)(*args, **kwargs) self._plotter.show(False) self._plotter.axes.set_autoscale_on(True) # end matplotlib.axes fowarding functions @asaplog_post_dec def set_data(self, scan, refresh=True): """ Set a scantable to plot. Parameters: scan: a scantable refresh: True (default) or False. If True, the plot is replotted based on the new parameter setting(s). Otherwise,the parameter(s) are set without replotting. Note: The user specified masks and data selections will be reset if a new scantable is set. This method should be called before setting data selections (set_selection) and/or masks (set_mask). """ from asap import scantable if isinstance(scan, scantable): if self._data is not None: if scan != self._data: self._data = scan # reset self._reset() msg = "A new scantable is set to the plotter. "\ "The masks and data selections are reset." asaplog.push( msg ) else: self._data = scan self._reset() else: msg = "Input is not a scantable" raise TypeError(msg) # ranges become invalid when unit changes if self._abcunit and self._abcunit != self._data.get_unit(): self._minmaxx = None self._minmaxy = None self._abcunit = self._data.get_unit() self._datamask = None if refresh: self.plot() @asaplog_post_dec def set_mode(self, stacking=None, panelling=None, refresh=True): """ Set the plots look and feel, i.e. what you want to see on the plot. Parameters: stacking: tell the plotter which variable to plot as line colour overlays (default 'pol') panelling: tell the plotter which variable to plot across multiple panels (default 'scan' refresh: True (default) or False. If True, the plot is replotted based on the new parameter setting(s). Otherwise,the parameter(s) are set without replotting. Note: Valid modes are: 'beam' 'Beam' 'b': Beams 'if' 'IF' 'i': IFs 'pol' 'Pol' 'p': Polarisations 'scan' 'Scan' 's': Scans 'time' 'Time' 't': Times 'row' 'Row' 'r': Rows When either 'stacking' or 'panelling' is set to 'row', the other parameter setting is ignored. """ msg = "Invalid mode" if not self.set_panelling(panelling) or \ not self.set_stacking(stacking): raise TypeError(msg) #if self._panelling == 'r': # self._stacking = '_r' #if self._stacking == 'r': # self._panelling = '_r' if refresh and self._data: self.plot(self._data) return def set_panelling(self, what=None): """Set the 'panelling' mode i.e. which type of spectra should be spread across different panels. """ mode = what if mode is None: mode = rcParams['plotter.panelling'] md = self._translate(mode) if md: self._panelling = md self._title = None #if md == 'r': # self._stacking = '_r' # you need to reset counters for multi page plotting self._reset_counters() return True return False def set_layout(self,rows=None,cols=None,refresh=True): """ Set the multi-panel layout, i.e. how many rows and columns plots are visible. Parameters: rows: The number of rows of plots cols: The number of columns of plots refresh: True (default) or False. If True, the plot is replotted based on the new parameter setting(s). Otherwise,the parameter(s) are set without replotting. Note: If no argument is given, the potter reverts to its auto-plot behaviour. """ self._rows = rows self._cols = cols if refresh and self._data: self.plot(self._data) return def set_stacking(self, what=None): """Set the 'stacking' mode i.e. which type of spectra should be overlayed. """ mode = what if mode is None: mode = rcParams['plotter.stacking'] md = self._translate(mode) if md: self._stacking = md self._lmap = None #if md == 'r': # self._panelling = '_r' # you need to reset counters for multi page plotting self._reset_counters() return True return False def _reset_counters(self): self._startrow = 0 self._ipanel = -1 self._panelrows = [] def set_range(self,xstart=None,xend=None,ystart=None,yend=None,refresh=True, offset=None): """ Set the range of interest on the abcissa of the plot Parameters: [x,y]start,[x,y]end: The start and end points of the 'zoom' window refresh: True (default) or False. If True, the plot is replotted based on the new parameter setting(s). Otherwise,the parameter(s) are set without replotting. offset: shift the abcissa by the given amount. The abcissa label will have '(relative)' appended to it. Note: These become non-sensical when the unit changes. use plotter.set_range() without parameters to reset """ self._offset = offset if xstart is None and xend is None: self._minmaxx = None else: self._minmaxx = [xstart,xend] if ystart is None and yend is None: self._minmaxy = None else: self._minmaxy = [ystart,yend] if refresh and self._data: self.plot(self._data) return def set_legend(self, mp=None, fontsize = None, mode = 0, refresh=True): """ Specify a mapping for the legend instead of using the default indices: Parameters: mp: a list of 'strings'. This should have the same length as the number of elements on the legend and then maps to the indeces in order. It is possible to uses latex math expression. These have to be enclosed in r'', e.g. r'$x^{2}$' fontsize: The font size of the label (default None) mode: where to display the legend Any other value for loc else disables the legend: 0: auto 1: upper right 2: upper left 3: lower left 4: lower right 5: right 6: center left 7: center right 8: lower center 9: upper center 10: center refresh: True (default) or False. If True, the plot is replotted based on the new parameter setting(s). Otherwise,the parameter(s) are set without replotting. Example: If the data has two IFs/rest frequencies with index 0 and 1 for CO and SiO: plotter.set_stacking('i') plotter.set_legend(['CO','SiO']) plotter.plot() plotter.set_legend([r'$^{12}CO$', r'SiO']) """ self._lmap = mp self._plotter.legend(mode) if isinstance(fontsize, int): from matplotlib import rc as rcp rcp('legend', fontsize=fontsize) if refresh and self._data: self.plot(self._data) return def set_title(self, title=None, fontsize=None, refresh=True): """ Set the title of the plot. If multiple panels are plotted, multiple titles have to be specified. Parameters: refresh: True (default) or False. If True, the plot is replotted based on the new parameter setting(s). Otherwise,the parameter(s) are set without replotting. Example: # two panels are visible on the plotter plotter.set_title(["First Panel","Second Panel"]) """ self._title = title if isinstance(fontsize, int): from matplotlib import rc as rcp rcp('axes', titlesize=fontsize) if refresh and self._data: self.plot(self._data) return def set_ordinate(self, ordinate=None, fontsize=None, refresh=True): """ Set the y-axis label of the plot. If multiple panels are plotted, multiple labels have to be specified. Parameters: ordinate: a list of ordinate labels. None (default) let data determine the labels refresh: True (default) or False. If True, the plot is replotted based on the new parameter setting(s). Otherwise,the parameter(s) are set without replotting. Example: # two panels are visible on the plotter plotter.set_ordinate(["First Y-Axis","Second Y-Axis"]) """ self._ordinate = ordinate if isinstance(fontsize, int): from matplotlib import rc as rcp rcp('axes', labelsize=fontsize) rcp('ytick', labelsize=fontsize) if refresh and self._data: self.plot(self._data) return def set_abcissa(self, abcissa=None, fontsize=None, refresh=True): """ Set the x-axis label of the plot. If multiple panels are plotted, multiple labels have to be specified. Parameters: abcissa: a list of abcissa labels. None (default) let data determine the labels refresh: True (default) or False. If True, the plot is replotted based on the new parameter setting(s). Otherwise,the parameter(s) are set without replotting. Example: # two panels are visible on the plotter plotter.set_ordinate(["First X-Axis","Second X-Axis"]) """ self._abcissa = abcissa if isinstance(fontsize, int): from matplotlib import rc as rcp rcp('axes', labelsize=fontsize) rcp('xtick', labelsize=fontsize) if refresh and self._data: self.plot(self._data) return def set_colors(self, colmap, refresh=True): """ Set the colours to be used. The plotter will cycle through these colours when lines are overlaid (stacking mode). Parameters: colmap: a list of colour names refresh: True (default) or False. If True, the plot is replotted based on the new parameter setting(s). Otherwise,the parameter(s) are set without replotting. Example: plotter.set_colors("red green blue") # If for example four lines are overlaid e.g I Q U V # 'I' will be 'red', 'Q' will be 'green', U will be 'blue' # and 'V' will be 'red' again. """ if isinstance(colmap,str): colmap = colmap.split() self._plotter.palette(0, colormap=colmap) if refresh and self._data: self.plot(self._data) # alias for english speakers set_colours = set_colors def set_histogram(self, hist=True, linewidth=None, refresh=True): """ Enable/Disable histogram-like plotting. Parameters: hist: True (default) or False. The fisrt default is taken from the .asaprc setting plotter.histogram refresh: True (default) or False. If True, the plot is replotted based on the new parameter setting(s). Otherwise,the parameter(s) are set without replotting. """ self._hist = hist if isinstance(linewidth, float) or isinstance(linewidth, int): from matplotlib import rc as rcp rcp('lines', linewidth=linewidth) if refresh and self._data: self.plot(self._data) def set_linestyles(self, linestyles=None, linewidth=None, refresh=True): """ Set the linestyles to be used. The plotter will cycle through these linestyles when lines are overlaid (stacking mode) AND only one color has been set. Parameters: linestyles: a list of linestyles to use. 'line', 'dashed', 'dotted', 'dashdot', 'dashdotdot' and 'dashdashdot' are possible refresh: True (default) or False. If True, the plot is replotted based on the new parameter setting(s). Otherwise,the parameter(s) are set without replotting. Example: plotter.set_colors("black") plotter.set_linestyles("line dashed dotted dashdot") # If for example four lines are overlaid e.g I Q U V # 'I' will be 'solid', 'Q' will be 'dashed', # U will be 'dotted' and 'V' will be 'dashdot'. """ if isinstance(linestyles,str): linestyles = linestyles.split() self._plotter.palette(color=0,linestyle=0,linestyles=linestyles) if isinstance(linewidth, float) or isinstance(linewidth, int): from matplotlib import rc as rcp rcp('lines', linewidth=linewidth) if refresh and self._data: self.plot(self._data) def set_font(self, refresh=True,**kwargs): """ Set font properties. Parameters: family: one of 'sans-serif', 'serif', 'cursive', 'fantasy', 'monospace' style: one of 'normal' (or 'roman'), 'italic' or 'oblique' weight: one of 'normal or 'bold' size: the 'general' font size, individual elements can be adjusted seperately refresh: True (default) or False. If True, the plot is replotted based on the new parameter setting(s). Otherwise,the parameter(s) are set without replotting. """ from matplotlib import rc as rcp fdict = {} for k,v in kwargs.iteritems(): if v: fdict[k] = v self._fp = FontProperties(**fdict) if refresh and self._data: self.plot(self._data) def set_margin(self,margin=[],refresh=True): """ Set margins between subplots and plot edges. Parameters: margin: a list of margins in figure coordinate (0-1), i.e., fraction of the figure width or height. The order of elements should be: [left, bottom, right, top, horizontal space btw panels, vertical space btw panels]. refresh: True (default) or False. If True, the plot is replotted based on the new parameter setting(s). Otherwise,the parameter(s) are set without replotting. Note * When margin is not specified, the values are reset to the defaults of matplotlib. * If any element is set to be None, the current value is adopted. """ if margin == []: self._margins=self._reset_margin() else: self._margins=[None]*6 self._margins[0:len(margin)]=margin #print "panel margin set to ",self._margins if refresh and self._data: self.plot(self._data) def _reset_margin(self): ks=map(lambda x: 'figure.subplot.'+x, ['left','bottom','right','top','hspace','wspace']) return map(matplotlib.rcParams.get,ks) def plot_lines(self, linecat=None, doppler=0.0, deltachan=10, rotate=90.0, location=None): """ Plot a line catalog. Parameters: linecat: the linecatalog to plot doppler: the velocity shift to apply to the frequencies deltachan: the number of channels to include each side of the line to determine a local maximum/minimum rotate: the rotation (in degrees) for the text label (default 90.0) location: the location of the line annotation from the 'top', 'bottom' or alternate (None - the default) Notes: If the spectrum is flagged no line will be drawn in that location. """ if not self._data: raise RuntimeError("No scantable has been plotted yet.") from asap._asap import linecatalog if not isinstance(linecat, linecatalog): raise ValueError("'linecat' isn't of type linecatalog.") if not self._data.get_unit().endswith("Hz"): raise RuntimeError("Can only overlay linecatalogs when data is in frequency.") from numpy import ma for j in range(len(self._plotter.subplots)): self._plotter.subplot(j) lims = self._plotter.axes.get_xlim() for row in range(linecat.nrow()): # get_frequency returns MHz base = { "GHz": 1000.0, "MHz": 1.0, "Hz": 1.0e-6 } restf = linecat.get_frequency(row)/base[self._data.get_unit()] c = 299792.458 freq = restf*(1.0-doppler/c) if lims[0] < freq < lims[1]: if location is None: loc = 'bottom' if row%2: loc='top' else: loc = location maxys = [] for line in self._plotter.axes.lines: v = line._x asc = v[0] < v[-1] idx = None if not asc: if v[len(v)-1] <= freq <= v[0]: i = len(v)-1 while i>=0 and v[i] < freq: idx = i i-=1 else: if v[0] <= freq <= v[len(v)-1]: i = 0 while i len(v): upper = len(v) s = slice(lower, upper) y = line._y[s] maxy = ma.maximum(y) if isinstance( maxy, float): maxys.append(maxy) if len(maxys): peak = max(maxys) if peak > self._plotter.axes.get_ylim()[1]: loc = 'bottom' else: continue self._plotter.vline_with_label(freq, peak, linecat.get_name(row), location=loc, rotate=rotate) self._plotter.show(hardrefresh=False) def save(self, filename=None, orientation=None, dpi=None): """ Save the plot to a file. The known formats are 'png', 'ps', 'eps'. Parameters: filename: The name of the output file. This is optional and autodetects the image format from the file suffix. If non filename is specified a file called 'yyyymmdd_hhmmss.png' is created in the current directory. orientation: optional parameter for postscript only (not eps). 'landscape', 'portrait' or None (default) are valid. If None is choosen for 'ps' output, the plot is automatically oriented to fill the page. dpi: The dpi of the output non-ps plot """ self._plotter.save(filename,orientation,dpi) return @asaplog_post_dec def set_mask(self, mask=None, selection=None, refresh=True): """ Set a plotting mask for a specific polarization. This is useful for masking out "noise" Pangle outside a source. Parameters: mask: a mask from scantable.create_mask selection: the spectra to apply the mask to. refresh: True (default) or False. If True, the plot is replotted based on the new parameter setting(s). Otherwise,the parameter(s) are set without replotting. Example: select = selector() select.setpolstrings("Pangle") plotter.set_mask(mymask, select) """ if not self._data: msg = "Can only set mask after a first call to plot()" raise RuntimeError(msg) if len(mask): if isinstance(mask, list) or isinstance(mask, tuple): self._usermask = array(mask) else: self._usermask = mask if mask is None and selection is None: self._usermask = [] self._maskselection = None if isinstance(selection, selector): self._maskselection = {'b': selection.get_beams(), 's': selection.get_scans(), 'i': selection.get_ifs(), 'p': selection.get_pols(), 't': [] } else: self._maskselection = None if refresh: self.plot(self._data) def _slice_indeces(self, data): mn = self._minmaxx[0] mx = self._minmaxx[1] asc = data[0] < data[-1] start=0 end = len(data)-1 inc = 1 if not asc: start = len(data)-1 end = 0 inc = -1 # find min index #while start > 0 and data[start] < mn: # start+= inc minind=start for ind in xrange(start,end+inc,inc): if data[ind] > mn: break minind=ind # find max index #while end > 0 and data[end] > mx: # end-=inc #if end > 0: end +=1 maxind=end for ind in xrange(end,start-inc,-inc): if data[ind] < mx: break maxind=ind start=minind end=maxind if start > end: return end,start+1 elif start < end: return start,end+1 else: return start,end def _reset(self): self._usermask = [] self._usermaskspectra = None self._offset = None self.set_selection(None, False) self._reset_header() def _reset_header(self): self._headtext={'string': None, 'textobj': None} def _plot(self, scan): savesel = scan.get_selection() sel = savesel + self._selection order = self._get_sortstring([self._panelling,self._stacking]) if order: sel.set_order(order) scan.set_selection(sel) d = {'b': scan.getbeam, 's': scan.getscan, 'i': scan.getif, 'p': scan.getpol, 't': scan.get_time, 'r': int}#, '_r': int} polmodes = dict(zip(self._selection.get_pols(), self._selection.get_poltypes())) # this returns either a tuple of numbers or a length (ncycles) # convert this into lengths n0,nstack0 = self._get_selected_n(scan) if isinstance(n0, int): n = n0 else: n = len(n0) if isinstance(nstack0, int): nstack = nstack0 else: nstack = len(nstack0) # In case of row stacking rowstack = False titlemode = self._panelling if self._stacking == "r" and self._panelling != "r": rowstack = True titlemode = '_r' nptot = n maxpanel, maxstack = 16,16 if nstack > maxstack: msg ="Scan to be overlayed contains more than %d selections.\n" \ "Selecting first %d selections..." % (maxstack, maxstack) asaplog.push(msg) asaplog.post('WARN') nstack = min(nstack,maxstack) #n = min(n-self._ipanel-1,maxpanel) n = n-self._ipanel-1 ganged = False if n > 1: ganged = rcParams['plotter.ganged'] if self._panelling == 'i': ganged = False if self._rows and self._cols: n = min(n,self._rows*self._cols) self._plotter.set_panels(rows=self._rows,cols=self._cols, nplots=n,margin=self._margins,ganged=ganged) else: n = min(n,maxpanel) self._plotter.set_panels(rows=n,cols=0,nplots=n,margin=self._margins,ganged=ganged) else: self._plotter.set_panels(margin=self._margins) #r = 0 r = self._startrow nr = scan.nrow() a0,b0 = -1,-1 allxlim = [] allylim = [] #newpanel=True newpanel=False panelcount,stackcount = 0,0 # If this is not the first page if r > 0: # panelling value of the prev page a0 = d[self._panelling](r-1) # set the initial stackcount large not to plot # the start row automatically stackcount = nstack while r < nr: a = d[self._panelling](r) b = d[self._stacking](r) if a > a0 and panelcount < n: if n > 1: self._plotter.subplot(panelcount) self._plotter.palette(0) #title xlab = self._abcissa and self._abcissa[panelcount] \ or scan._getabcissalabel() if self._offset and not self._abcissa: xlab += " (relative)" ylab = self._ordinate and self._ordinate[panelcount] \ or scan._get_ordinate_label() self._plotter.set_axes('xlabel', xlab) self._plotter.set_axes('ylabel', ylab) #lbl = self._get_label(scan, r, self._panelling, self._title) lbl = self._get_label(scan, r, titlemode, self._title) if isinstance(lbl, list) or isinstance(lbl, tuple): if 0 <= panelcount < len(lbl): lbl = lbl[panelcount] else: # get default label #lbl = self._get_label(scan, r, self._panelling, None) lbl = self._get_label(scan, r, titlemode, None) self._plotter.set_axes('title',lbl) newpanel = True stackcount = 0 panelcount += 1 # save the start row to plot this panel for future revisit. if self._panelling != 'r' and \ len(self._panelrows) < self._ipanel+1+panelcount: self._panelrows += [r] #if (b > b0 or newpanel) and stackcount < nstack: if stackcount < nstack and (newpanel or rowstack or (a == a0 and b > b0)): y = [] if len(polmodes): y = scan._getspectrum(r, polmodes[scan.getpol(r)]) else: y = scan._getspectrum(r) # flag application mr = scan._getflagrow(r) from numpy import ma, array if mr: y = ma.masked_array(y,mask=mr) else: m = scan._getmask(r) from numpy import logical_not, logical_and if self._maskselection and len(self._usermask) == len(m): if d[self._stacking](r) in self._maskselection[self._stacking]: m = logical_and(m, self._usermask) y = ma.masked_array(y,mask=logical_not(array(m,copy=False))) x = array(scan._getabcissa(r)) if self._offset: x += self._offset if self._minmaxx is not None: s,e = self._slice_indeces(x) x = x[s:e] y = y[s:e] if len(x) > 1024 and rcParams['plotter.decimate']: fac = len(x)/1024 x = x[::fac] y = y[::fac] llbl = self._get_label(scan, r, self._stacking, self._lmap) if isinstance(llbl, list) or isinstance(llbl, tuple): if 0 <= stackcount < len(llbl): # use user label llbl = llbl[stackcount] else: # get default label llbl = self._get_label(scan, r, self._stacking, None) self._plotter.set_line(label=llbl) plotit = self._plotter.plot if self._hist: plotit = self._plotter.hist if len(x) > 0 and not mr: plotit(x,y) xlim= self._minmaxx or [min(x),max(x)] allxlim += xlim ylim= self._minmaxy or [ma.minimum(y),ma.maximum(y)] allylim += ylim else: xlim = self._minmaxx or [] allxlim += xlim ylim= self._minmaxy or [] allylim += ylim stackcount += 1 a0=a b0=b # last in colour stack -> autoscale x if stackcount == nstack and len(allxlim) > 0: allxlim.sort() self._plotter.subplots[panelcount-1]['axes'].set_xlim([allxlim[0],allxlim[-1]]) if ganged: allxlim = [allxlim[0],allxlim[-1]] else: # clear allxlim =[] newpanel = False #a0=a #b0=b # ignore following rows if (panelcount == n and stackcount == nstack) or (r == nr-1): # last panel -> autoscale y if ganged #if rcParams['plotter.ganged'] and len(allylim) > 0: if ganged and len(allylim) > 0: allylim.sort() self._plotter.set_limits(ylim=[allylim[0],allylim[-1]]) break r+=1 # next row # save the current counter for multi-page plotting self._startrow = r+1 self._ipanel += panelcount if self._plotter.figmgr.casabar: if self._ipanel >= nptot-1: self._plotter.figmgr.casabar.disable_next() else: self._plotter.figmgr.casabar.enable_next() if self._ipanel + 1 - panelcount > 0: self._plotter.figmgr.casabar.enable_prev() else: self._plotter.figmgr.casabar.disable_prev() #reset the selector to the scantable's original scan.set_selection(savesel) #temporary switch-off for older matplotlib #if self._fp is not None: if self._fp is not None and getattr(self._plotter.figure,'findobj',False): for o in self._plotter.figure.findobj(Text): o.set_fontproperties(self._fp) def _get_sortstring(self, lorders): d0 = {'s': 'SCANNO', 'b': 'BEAMNO', 'i':'IFNO', 'p': 'POLNO', 'c': 'CYCLENO', 't' : 'TIME', 'r':None, '_r':None } if not (type(lorders) == list) and not (type(lorders) == tuple): return None if len(lorders) > 0: lsorts = [] for order in lorders: if order == "r": # don't sort if row panelling/stacking return None ssort = d0[order] if ssort: lsorts.append(ssort) return lsorts return None def set_selection(self, selection=None, refresh=True, **kw): """ Parameters: selection: a selector object (default unset the selection) refresh: True (default) or False. If True, the plot is replotted based on the new parameter setting(s). Otherwise,the parameter(s) are set without replotting. """ if selection is None: # reset if len(kw) == 0: self._selection = selector() else: # try keywords for k in kw: if k not in selector.fields: raise KeyError("Invalid selection key '%s', valid keys are %s" % (k, selector.fields)) self._selection = selector(**kw) elif isinstance(selection, selector): self._selection = selection else: raise TypeError("'selection' is not of type selector") order = self._get_sortstring([self._panelling,self._stacking]) if order: self._selection.set_order(order) if refresh and self._data: self.plot(self._data) def _get_selected_n(self, scan): d1 = {'b': scan.getbeamnos, 's': scan.getscannos, 'i': scan.getifnos, 'p': scan.getpolnos, 't': scan.ncycle, 'r': scan.nrow}#, '_r': False} d2 = { 'b': self._selection.get_beams(), 's': self._selection.get_scans(), 'i': self._selection.get_ifs(), 'p': self._selection.get_pols(), 't': self._selection.get_cycles(), 'r': False}#, '_r': 1} n = d2[self._panelling] or d1[self._panelling]() nstack = d2[self._stacking] or d1[self._stacking]() # handle row panelling/stacking if self._panelling == 'r': nstack = 1 elif self._stacking == 'r': n = 1 return n,nstack def _get_label(self, scan, row, mode, userlabel=None): if isinstance(userlabel, list) and len(userlabel) == 0: userlabel = " " pms = dict(zip(self._selection.get_pols(),self._selection.get_poltypes())) if len(pms): poleval = scan._getpollabel(scan.getpol(row),pms[scan.getpol(row)]) else: poleval = scan._getpollabel(scan.getpol(row),scan.poltype()) d = {'b': "Beam "+str(scan.getbeam(row)), #'s': scan._getsourcename(row), 's': "Scan "+str(scan.getscan(row))+\ " ("+str(scan._getsourcename(row))+")", 'i': "IF"+str(scan.getif(row)), 'p': poleval, 't': str(scan.get_time(row)), 'r': "row "+str(row), #'_r': str(scan.get_time(row))+",\nIF"+str(scan.getif(row))+", "+poleval+", Beam"+str(scan.getbeam(row)) } '_r': "" } return userlabel or d[mode] def plotazel(self, scan=None, outfile=None): """ plot azimuth and elevation versus time of a scantable """ visible = rcParams['plotter.gui'] from matplotlib import pylab as PL from matplotlib.dates import DateFormatter, timezone from matplotlib.dates import HourLocator, MinuteLocator,SecondLocator, DayLocator from matplotlib.ticker import MultipleLocator from numpy import array, pi if not visible or not self._visible: PL.ioff() from matplotlib.backends.backend_agg import FigureCanvasAgg PL.gcf().canvas.switch_backends(FigureCanvasAgg) self._data = scan self._outfile = outfile dates = self._data.get_time(asdatetime=True) t = PL.date2num(dates) tz = timezone('UTC') PL.cla() PL.ioff() PL.clf() # Adjust subplot margins if len(self._margins) != 6: self.set_margin(refresh=False) lef, bot, rig, top, wsp, hsp = self._margins PL.gcf().subplots_adjust(left=lef,bottom=bot,right=rig,top=top, wspace=wsp,hspace=hsp) tdel = max(t) - min(t) ax = PL.subplot(2,1,1) el = array(self._data.get_elevation())*180./pi PL.ylabel('El [deg.]') dstr = dates[0].strftime('%Y/%m/%d') if tdel > 1.0: dstr2 = dates[len(dates)-1].strftime('%Y/%m/%d') dstr = dstr + " - " + dstr2 majloc = DayLocator() minloc = HourLocator(range(0,23,12)) timefmt = DateFormatter("%b%d") elif tdel > 24./60.: timefmt = DateFormatter('%H:%M') majloc = HourLocator() minloc = MinuteLocator(30) else: timefmt = DateFormatter('%H:%M') majloc = MinuteLocator(interval=5) minloc = SecondLocator(30) PL.title(dstr) if tdel == 0.0: th = (t - PL.floor(t))*24.0 PL.plot(th,el,'o',markersize=2, markerfacecolor='b', markeredgecolor='b') else: PL.plot_date(t,el,'o', markersize=2, markerfacecolor='b', markeredgecolor='b',tz=tz) #ax.grid(True) ax.xaxis.set_major_formatter(timefmt) ax.xaxis.set_major_locator(majloc) ax.xaxis.set_minor_locator(minloc) ax.yaxis.grid(True) yloc = MultipleLocator(30) ax.set_ylim(0,90) ax.yaxis.set_major_locator(yloc) if tdel > 1.0: labels = ax.get_xticklabels() # PL.setp(labels, fontsize=10, rotation=45) PL.setp(labels, fontsize=10) # Az plot az = array(self._data.get_azimuth())*180./pi if min(az) < 0: for irow in range(len(az)): if az[irow] < 0: az[irow] += 360.0 ax2 = PL.subplot(2,1,2) #PL.xlabel('Time (UT [hour])') PL.ylabel('Az [deg.]') if tdel == 0.0: PL.plot(th,az,'o',markersize=2, markeredgecolor='b',markerfacecolor='b') else: PL.plot_date(t,az,'o', markersize=2,markeredgecolor='b',markerfacecolor='b',tz=tz) ax2.xaxis.set_major_formatter(timefmt) ax2.xaxis.set_major_locator(majloc) ax2.xaxis.set_minor_locator(minloc) #ax2.grid(True) ax2.set_ylim(0,360) ax2.yaxis.grid(True) #hfmt = DateFormatter('%H') #hloc = HourLocator() yloc = MultipleLocator(60) ax2.yaxis.set_major_locator(yloc) if tdel > 1.0: labels = ax2.get_xticklabels() PL.setp(labels, fontsize=10) PL.xlabel('Time (UT [day])') else: PL.xlabel('Time (UT [hour])') PL.ion() PL.draw() if (self._outfile is not None): PL.savefig(self._outfile) def plotpointing(self, scan=None, outfile=None): """ plot telescope pointings """ visible = rcParams['plotter.gui'] from matplotlib import pylab as PL from numpy import array, pi if not visible or not self._visible: PL.ioff() from matplotlib.backends.backend_agg import FigureCanvasAgg PL.gcf().canvas.switch_backends(FigureCanvasAgg) self._data = scan self._outfile = outfile dir = array(self._data.get_directionval()).transpose() ra = dir[0]*180./pi dec = dir[1]*180./pi PL.cla() #PL.ioff() PL.clf() # Adjust subplot margins if len(self._margins) != 6: self.set_margin(refresh=False) lef, bot, rig, top, wsp, hsp = self._margins PL.gcf().subplots_adjust(left=lef,bottom=bot,right=rig,top=top, wspace=wsp,hspace=hsp) ax = PL.gca() #ax = PL.axes([0.1,0.1,0.8,0.8]) #ax = PL.axes([0.1,0.1,0.8,0.8]) ax.set_aspect('equal') PL.plot(ra, dec, 'b,') PL.xlabel('RA [deg.]') PL.ylabel('Declination [deg.]') PL.title('Telescope pointings') [xmin,xmax,ymin,ymax] = PL.axis() PL.axis([xmax,xmin,ymin,ymax]) #PL.ion() PL.draw() if (self._outfile is not None): PL.savefig(self._outfile) # plot total power data # plotting in time is not yet implemented.. @asaplog_post_dec def plottp(self, scan=None, outfile=None): if self._plotter.is_dead: if hasattr(self._plotter.figmgr,'casabar'): del self._plotter.figmgr.casabar self._plotter = self._newplotter() self._plotter.figmgr.casabar=self._newcasabar() self._plotter.hold() self._plotter.clear() from asap import scantable if not self._data and not scan: msg = "Input is not a scantable" raise TypeError(msg) if isinstance(scan, scantable): if self._data is not None: if scan != self._data: self._data = scan # reset self._reset() else: self._data = scan self._reset() # ranges become invalid when abcissa changes? #if self._abcunit and self._abcunit != self._data.get_unit(): # self._minmaxx = None # self._minmaxy = None # self._abcunit = self._data.get_unit() # self._datamask = None # Adjust subplot margins if len(self._margins) !=6: self.set_margin(refresh=False) lef, bot, rig, top, wsp, hsp = self._margins self._plotter.figure.subplots_adjust( left=lef,bottom=bot,right=rig,top=top,wspace=wsp,hspace=hsp) if self._plotter.figmgr.casabar: self._plotter.figmgr.casabar.disable_button() self._plottp(self._data) if self._minmaxy is not None: self._plotter.set_limits(ylim=self._minmaxy) self._plotter.release() self._plotter.tidy() self._plotter.show(hardrefresh=False) return def _plottp(self,scan): """ private method for plotting total power data """ from numpy import ma, array, arange, logical_not r=0 nr = scan.nrow() a0,b0 = -1,-1 allxlim = [] allylim = [] y=[] self._plotter.set_panels() self._plotter.palette(0) #title #xlab = self._abcissa and self._abcissa[panelcount] \ # or scan._getabcissalabel() #ylab = self._ordinate and self._ordinate[panelcount] \ # or scan._get_ordinate_label() xlab = self._abcissa or 'row number' #or Time ylab = self._ordinate or scan._get_ordinate_label() self._plotter.set_axes('xlabel',xlab) self._plotter.set_axes('ylabel',ylab) lbl = self._get_label(scan, r, 's', self._title) if isinstance(lbl, list) or isinstance(lbl, tuple): # if 0 <= panelcount < len(lbl): # lbl = lbl[panelcount] # else: # get default label lbl = self._get_label(scan, r, self._panelling, None) self._plotter.set_axes('title',lbl) y=array(scan._get_column(scan._getspectrum,-1)) m = array(scan._get_column(scan._getmask,-1)) y = ma.masked_array(y,mask=logical_not(array(m,copy=False))) x = arange(len(y)) # try to handle spectral data somewhat... l,m = y.shape if m > 1: y=y.mean(axis=1) plotit = self._plotter.plot llbl = self._get_label(scan, r, self._stacking, None) self._plotter.set_line(label=llbl) if len(x) > 0: plotit(x,y) # forwards to matplotlib.Figure.text def figtext(self, *args, **kwargs): """ Add text to figure at location x,y (relative 0-1 coords). This method forwards *args and **kwargs to a Matplotlib method, matplotlib.Figure.text. See the method help for detailed information. """ self._plotter.text(*args, **kwargs) # end matplotlib.Figure.text forwarding function # printing header information @asaplog_post_dec def print_header(self, plot=True, fontsize=9, logger=False, selstr='', extrastr=''): """ print data (scantable) header on the plot and/or logger. To plot the header on the plot, this method should be called after plotting spectra by the method, asapplotter.plot. Parameters: plot: whether or not print header info on the plot. fontsize: header font size (valid only plot=True) logger: whether or not print header info on the logger. selstr: additional selection string (not verified) extrastr: additional string to print at the beginning (not verified) """ if not plot and not logger: return if not self._data: raise RuntimeError("No scantable has been set yet.") # Now header will be printed on plot and/or logger. # Get header information and format it. ssum=self._data.__str__() # Print Observation header to the upper-left corner of plot headstr=[ssum[ssum.find('Observer:'):ssum.find('Flux Unit:')]] headstr.append(ssum[ssum.find('Beams:'):ssum.find('Observer:')] +ssum[ssum.find('Rest Freqs:'):ssum.find('Abcissa:')]) if extrastr != '': headstr[0]=extrastr+'\n'+headstr[0] self._headtext['extrastr'] = extrastr if selstr != '': self._headtext['selstr'] = selstr ssel=(selstr+self._data.get_selection().__str__()+self._selection.__str__() or 'none') headstr.append('***Selections***\n'+ssel) if plot: self._plotter.hold() self._header_plot(headstr,fontsize=fontsize) import time self._plotter.figure.text(0.99,0.01, time.strftime("%a %d %b %Y %H:%M:%S %Z"), horizontalalignment='right', verticalalignment='bottom',fontsize=8) self._plotter.release() if logger: selstr = "Selections: "+ssel asaplog.push("----------------\n Plot Summary\n----------------") asaplog.push(extrastr) asaplog.push(ssum[ssum.find('Beams:'):ssum.find('Selection:')]\ + selstr + ssum[ssum.find('Scan Source'):]) self._headtext['string'] = headstr del ssel, ssum, headstr def _header_plot(self, texts, fontsize=9): self._headtext['textobj']=[] nstcol=len(texts) for i in range(nstcol): self._headtext['textobj'].append( self._plotter.figure.text(0.03+float(i)/nstcol,0.98, texts[i], horizontalalignment='left', verticalalignment='top', fontsize=fontsize)) def clear_header(self): if not self._headtext['textobj']: asaplog.push("No header has been plotted. Exit without any operation") asaplog.post("WARN") else: self._plotter.hold() for textobj in self._headtext['textobj']: #if textobj.get_text() in self._headstring: try: textobj.remove() except NotImplementedError: self._plotter.figure.texts.pop(self._plotter.figure.texts.index(textobj)) self._plotter.release() self._reset_header()