""" Plot panel. """ import logging import traceback import wx # Try a normal import first # If it fails, try specifying a version import matplotlib matplotlib.interactive(False) #Use the WxAgg back end. The Wx one takes too long to render matplotlib.use('WXAgg') from matplotlib.backends.backend_wxagg import FigureCanvasWxAgg from matplotlib.figure import Figure import os import transform from plottables import Data1D #TODO: make the plottables interactive from binder import BindArtist from matplotlib.font_manager import FontProperties DEBUG = False from plottables import Graph from plottables import Text from TextDialog import TextDialog from LabelDialog import LabelDialog import operator import math import pylab DEFAULT_CMAP = pylab.cm.jet import copy import numpy from sas.guiframe.events import StatusEvent def show_tree(obj, d=0): """Handy function for displaying a tree of graph objects""" print "%s%s" % ("-"*d, obj.__class__.__name__) if 'get_children' in dir(obj): for a in obj.get_children(): show_tree(a, d + 1) from convert_units import convert_unit def _rescale(lo, hi, step, pt=None, bal=None, scale='linear'): """ Rescale (lo,hi) by step, returning the new (lo,hi) The scaling is centered on pt, with positive values of step driving lo/hi away from pt and negative values pulling them in. If bal is given instead of point, it is already in [0,1] coordinates. This is a helper function for step-based zooming. """ # Convert values into the correct scale for a linear transformation # TODO: use proper scale transformers loprev = lo hiprev = hi if scale == 'log': assert lo > 0 if lo > 0: lo = math.log10(lo) if hi > 0: hi = math.log10(hi) if pt is not None: pt = math.log10(pt) # Compute delta from axis range * %, or 1-% if persent is negative if step > 0: delta = float(hi - lo) * step / 100 else: delta = float(hi - lo) * step / (100 - step) # Add scale factor proportionally to the lo and hi values, # preserving the # point under the mouse if bal is None: bal = float(pt - lo) / (hi - lo) lo = lo - (bal * delta) hi = hi + (1 - bal) * delta # Convert transformed values back to the original scale if scale == 'log': if (lo <= -250) or (hi >= 250): lo = loprev hi = hiprev else: lo, hi = math.pow(10., lo), math.pow(10., hi) return (lo, hi) def CopyImage(canvas): """ 0: matplotlib plot 1: wx.lib.plot 2: other """ bmp = wx.BitmapDataObject() bmp.SetBitmap(canvas.bitmap) wx.TheClipboard.Open() wx.TheClipboard.SetData(bmp) wx.TheClipboard.Close() class PlotPanel(wx.Panel): """ The PlotPanel has a Figure and a Canvas. OnSize events simply set a flag, and the actually redrawing of the figure is triggered by an Idle event. """ def __init__(self, parent, id=-1, xtransform=None, ytransform=None, scale='log_{10}', color=None, dpi=None, **kwargs): """ """ wx.Panel.__init__(self, parent, id=id, **kwargs) self.parent = parent if hasattr(parent, "parent"): self.parent = self.parent.parent self.dimension = 1 self.gotLegend = 0 # to begin, legend is not picked. self.legend_pos_loc = None self.legend = None self.line_collections_list = [] self.figure = Figure(None, dpi, linewidth=2.0) self.color = '#b3b3b3' from canvas import FigureCanvas self.canvas = FigureCanvas(self, -1, self.figure) self.SetColor(color) self._resizeflag = True self._SetSize() self.subplot = self.figure.add_subplot(111) self.figure.subplots_adjust(left=0.2, bottom=.2) self.yscale = 'linear' self.xscale = 'linear' self.sizer = wx.BoxSizer(wx.VERTICAL) self.sizer.Add(self.canvas, 1, wx.EXPAND) #add toolbar self.enable_toolbar = True self.toolbar = None self.add_toolbar() self.SetSizer(self.sizer) # Graph object to manage the plottables self.graph = Graph() #Boolean value to keep track of whether current legend is #visible or not self.legend_on = True self.grid_on = False #Location of legend, default is 0 or 'best' self.legendLoc = 0 self.position = None self._loc_labels = self.get_loc_label() self.Bind(wx.EVT_CONTEXT_MENU, self.onContextMenu) # Define some constants self.colorlist = ['b', 'g', 'r', 'c', 'm', 'y', 'k'] self.symbollist = ['o', 'x', '^', 'v', '<', '>', '+', 's', 'd', 'D', 'h', 'H', 'p', '-'] #List of texts currently on the plot self.textList = [] #User scale if xtransform != None: self.xLabel = xtransform else: self.xLabel = "log10(x)" if ytransform != None: self.yLabel = ytransform else: self.yLabel = "log10(y)" self.viewModel = "--" # keep track if the previous transformation of x # and y in Property dialog self.prevXtrans = "log10(x)" self.prevYtrans = "log10(y)" self.scroll_id = self.canvas.mpl_connect('scroll_event', self.onWheel) #taking care of dragging self.motion_id = self.canvas.mpl_connect('motion_notify_event', self.onMouseMotion) self.press_id = self.canvas.mpl_connect('button_press_event', self.onLeftDown) self.pick_id = self.canvas.mpl_connect('pick_event', self.onPick) self.release_id = self.canvas.mpl_connect('button_release_event', self.onLeftUp) wx.EVT_RIGHT_DOWN(self, self.onLeftDown) # to turn axis off whenn resizing the panel self.resizing = False self.leftdown = False self.leftup = False self.mousemotion = False self.axes = [self.subplot] ## Fit dialog self._fit_dialog = None # Interactor self.connect = BindArtist(self.subplot.figure) #self.selected_plottable = None # new data for the fit self.fit_result = Data1D(x=[], y=[], dy=None) self.fit_result.symbol = 13 #self.fit_result = Data1D(x=[], y=[],dx=None, dy=None) self.fit_result.name = "Fit" # For fit Dialog initial display self.xmin = 0.0 self.xmax = 0.0 self.xminView = 0.0 self.xmaxView = 0.0 self._scale_xlo = None self._scale_xhi = None self._scale_ylo = None self._scale_yhi = None self.Avalue = None self.Bvalue = None self.ErrAvalue = None self.ErrBvalue = None self.Chivalue = None # for 2D scale if scale != 'linear': scale = 'log_{10}' self.scale = scale self.data = None self.qx_data = None self.qy_data = None self.xmin_2D = None self.xmax_2D = None self.ymin_2D = None self.ymax_2D = None ## store reference to the current plotted vmin and vmax of plotted image ##z range in linear scale self.zmin_2D = None self.zmax_2D = None #index array self.index_x = None self.index_y = None #number of bins self.x_bins = None self.y_bins = None ## default color map self.cmap = DEFAULT_CMAP # Dragging info self.begDrag = False self.xInit = None self.yInit = None self.xFinal = None self.yFinal = None #axes properties self.xaxis_font = None self.xaxis_label = None self.xaxis_unit = None self.xaxis_color = 'black' self.xaxis_tick = None self.yaxis_font = None self.yaxis_label = None self.yaxis_unit = None self.yaxis_color = 'black' self.yaxis_tick = None # check if zoomed. self.is_zoomed = False # Plottables self.plots = {} # Default locations self._default_save_location = os.getcwd() # let canvas know about axes self.canvas.set_panel(self) self.ly = None self.q_ctrl = None #Bind focus to change the border color self.canvas.Bind(wx.EVT_SET_FOCUS, self.on_set_focus) self.canvas.Bind(wx.EVT_KILL_FOCUS, self.on_kill_focus) def _SetInitialSize(self,): """ """ pixels = self.parent.GetClientSize() self.canvas.SetSize(pixels) self.figure.set_size_inches(pixels[0] / self.figure.get_dpi(), pixels[1] / self.figure.get_dpi(), forward=True) def On_Paint(self, event): """ """ self.canvas.SetBackgroundColour(self.color) def on_set_focus(self, event): """ Send to the parenet the current panel on focus """ # light blue self.color = '#0099f7' self.figure.set_edgecolor(self.color) if self.parent and self.window_caption: self.parent.send_focus_to_datapanel(self.window_caption) self.draw() def on_kill_focus(self, event): """ Reset the panel color """ # light grey self.color = '#b3b3b3' self.figure.set_edgecolor(self.color) self.draw() def set_resizing(self, resizing=False): """ Set the resizing (True/False) """ pass # Not implemented def schedule_full_draw(self, func='append'): """ Put self in schedule to full redraw list """ pass # Not implemented def add_toolbar(self): """ add toolbar """ self.enable_toolbar = True from toolbar import NavigationToolBar self.toolbar = NavigationToolBar(parent=self, canvas=self.canvas) self.toolbar.Realize() ## The 'SetToolBar()' is not working on MAC: JHC #if IS_MAC: # Mac platform (OSX 10.3, MacPython) does not seem to cope with # having a toolbar in a sizer. This work-around gets the buttons # back, but at the expense of having the toolbar at the top #self.SetToolBar(self.toolbar) #else: # On Windows platform, default window size is incorrect, so set # toolbar width to figure width. tw, th = self.toolbar.GetSizeTuple() fw, fh = self.canvas.GetSizeTuple() # By adding toolbar in sizer, we are able to put it at the bottom # of the frame - so appearance is closer to GTK version. # As noted above, doesn't work for Mac. self.toolbar.SetSize(wx.Size(fw, th)) self.sizer.Add(self.toolbar, 0, wx.LEFT | wx.EXPAND) # update the axes menu on the toolbar self.toolbar.update() def onLeftDown(self, event): """ left button down and ready to drag """ # Check that the LEFT button was pressed if event.button == 1: self.leftdown = True ax = event.inaxes if ax != None: self.xInit, self.yInit = event.xdata, event.ydata try: pos_x = float(event.xdata) # / size_x pos_y = float(event.ydata) # / size_y pos_x = "%8.3g" % pos_x pos_y = "%8.3g" % pos_y self.position = str(pos_x), str(pos_y) wx.PostEvent(self.parent, StatusEvent(status=self.position)) except: self.position = None def onLeftUp(self, event): """ Dragging is done """ # Check that the LEFT button was released if event.button == 1: self.leftdown = False self.mousemotion = False self.leftup = True #release the legend if self.gotLegend == 1: self.gotLegend = 0 self.set_legend_alpha(1) def set_legend_alpha(self, alpha=1): """ Set legend alpha """ if self.legend != None: self.legend.legendPatch.set_alpha(alpha) def onPick(self, event): """ On pick legend """ legend = self.legend if event.artist == legend: #gets the box of the legend. bbox = self.legend.get_window_extent() #get mouse coordinates at time of pick. self.mouse_x = event.mouseevent.x self.mouse_y = event.mouseevent.y #get legend coordinates at time of pick. self.legend_x = bbox.xmin self.legend_y = bbox.ymin #indicates we picked up the legend. self.gotLegend = 1 self.set_legend_alpha(0.5) def _on_legend_motion(self, event): """ On legend in motion """ ax = event.inaxes if ax == None: return # Event occurred inside a plotting area lo_x, hi_x = ax.get_xlim() lo_y, hi_y = ax.get_ylim() # How much the mouse moved. x = mouse_diff_x = self.mouse_x - event.x y = mouse_diff_y = self.mouse_y - event.y # Put back inside if x < lo_x: x = lo_x if x > hi_x: x = hi_x if y < lo_y: y = lo_y if y > hi_y: y = hi_y # Move the legend from its previous location by that same amount loc_in_canvas = self.legend_x - mouse_diff_x, \ self.legend_y - mouse_diff_y # Transform into legend coordinate system trans_axes = self.legend.parent.transAxes.inverted() loc_in_norm_axes = trans_axes.transform_point(loc_in_canvas) self.legend_pos_loc = tuple(loc_in_norm_axes) self.legend._loc = self.legend_pos_loc self.resizing = True self.canvas.set_resizing(self.resizing) self.canvas.draw() def onMouseMotion(self, event): """ check if the left button is press and the mouse in moving. computer delta for x and y coordinates and then calls draghelper to perform the drag """ self.cusor_line(event) if self.gotLegend == 1: self._on_legend_motion(event) return if self.enable_toolbar: #Disable dragging without the toolbar to allow zooming with toolbar return self.mousemotion = True if self.leftdown == True and self.mousemotion == True: ax = event.inaxes if ax != None: # the dragging is perform inside the figure self.xFinal, self.yFinal = event.xdata, event.ydata # Check whether this is the first point if self.xInit == None: self.xInit = self.xFinal self.yInit = self.yFinal xdelta = self.xFinal - self.xInit ydelta = self.yFinal - self.yInit if self.xscale == 'log': xdelta = math.log10(self.xFinal) - math.log10(self.xInit) if self.yscale == 'log': ydelta = math.log10(self.yFinal) - math.log10(self.yInit) self._dragHelper(xdelta, ydelta) else: # no dragging is perform elsewhere self._dragHelper(0, 0) def cusor_line(self, event): """ """ pass def _offset_graph(self): """ Zoom and offset the graph to the last known settings """ for ax in self.axes: if self._scale_xhi is not None and self._scale_xlo is not None: ax.set_xlim(self._scale_xlo, self._scale_xhi) if self._scale_yhi is not None and self._scale_ylo is not None: ax.set_ylim(self._scale_ylo, self._scale_yhi) def _dragHelper(self, xdelta, ydelta): """ dragging occurs here """ # Event occurred inside a plotting area for ax in self.axes: lo, hi = ax.get_xlim() newlo, newhi = lo - xdelta, hi - xdelta if self.xscale == 'log': if lo > 0: newlo = math.log10(lo) - xdelta if hi > 0: newhi = math.log10(hi) - xdelta if self.xscale == 'log': self._scale_xlo = math.pow(10, newlo) self._scale_xhi = math.pow(10, newhi) ax.set_xlim(math.pow(10, newlo), math.pow(10, newhi)) else: self._scale_xlo = newlo self._scale_xhi = newhi ax.set_xlim(newlo, newhi) lo, hi = ax.get_ylim() newlo, newhi = lo - ydelta, hi - ydelta if self.yscale == 'log': if lo > 0: newlo = math.log10(lo) - ydelta if hi > 0: newhi = math.log10(hi) - ydelta if self.yscale == 'log': self._scale_ylo = math.pow(10, newlo) self._scale_yhi = math.pow(10, newhi) ax.set_ylim(math.pow(10, newlo), math.pow(10, newhi)) else: self._scale_ylo = newlo self._scale_yhi = newhi ax.set_ylim(newlo, newhi) self.canvas.draw_idle() def resetFitView(self): """ For fit Dialog initial display """ self.xmin = 0.0 self.xmax = 0.0 self.xminView = 0.0 self.xmaxView = 0.0 self._scale_xlo = None self._scale_xhi = None self._scale_ylo = None self._scale_yhi = None self.Avalue = None self.Bvalue = None self.ErrAvalue = None self.ErrBvalue = None self.Chivalue = None def onWheel(self, event): """ Process mouse wheel as zoom events :param event: Wheel event """ ax = event.inaxes step = event.step if ax != None: # Event occurred inside a plotting area lo, hi = ax.get_xlim() lo, hi = _rescale(lo, hi, step, pt=event.xdata, scale=ax.get_xscale()) if not self.xscale == 'log' or lo > 0: self._scale_xlo = lo self._scale_xhi = hi ax.set_xlim((lo, hi)) lo, hi = ax.get_ylim() lo, hi = _rescale(lo, hi, step, pt=event.ydata, scale=ax.get_yscale()) if not self.yscale == 'log' or lo > 0: self._scale_ylo = lo self._scale_yhi = hi ax.set_ylim((lo, hi)) else: # Check if zoom happens in the axes xdata, ydata = None, None x, y = event.x, event.y for ax in self.axes: insidex, _ = ax.xaxis.contains(event) if insidex: xdata, _ = ax.transAxes.inverted().transform_point((x, y)) insidey, _ = ax.yaxis.contains(event) if insidey: _, ydata = ax.transAxes.inverted().transform_point((x, y)) if xdata is not None: lo, hi = ax.get_xlim() lo, hi = _rescale(lo, hi, step, bal=xdata, scale=ax.get_xscale()) if not self.xscale == 'log' or lo > 0: self._scale_xlo = lo self._scale_xhi = hi ax.set_xlim((lo, hi)) if ydata is not None: lo, hi = ax.get_ylim() lo, hi = _rescale(lo, hi, step, bal=ydata, scale=ax.get_yscale()) if not self.yscale == 'log' or lo > 0: self._scale_ylo = lo self._scale_yhi = hi ax.set_ylim((lo, hi)) self.canvas.draw_idle() def returnTrans(self): """ Return values and labels used by Fit Dialog """ return self.xLabel, self.yLabel, self.Avalue, self.Bvalue, \ self.ErrAvalue, self.ErrBvalue, self.Chivalue def setTrans(self, xtrans, ytrans): """ :param xtrans: set x transformation on Property dialog :param ytrans: set y transformation on Property dialog """ self.prevXtrans = xtrans self.prevYtrans = ytrans def onFitting(self, event): """ when clicking on linear Fit on context menu , display Fitting Dialog """ list = {} menu = event.GetEventObject() id = event.GetId() self.set_selected_from_menu(menu, id) plotlist = self.graph.returnPlottable() if self.graph.selected_plottable is not None: for item in plotlist: if item.id == self.graph.selected_plottable: list[item] = plotlist[item] else: list = plotlist from fitDialog import LinearFit if len(list.keys()) > 0: first_item = list.keys()[0] dlg = LinearFit(parent=None, plottable=first_item, push_data=self.onFitDisplay, transform=self.returnTrans, title='Linear Fit') if (self.xmin != 0.0)and (self.xmax != 0.0)\ and(self.xminView != 0.0)and (self.xmaxView != 0.0): dlg.setFitRange(self.xminView, self.xmaxView, self.xmin, self.xmax) dlg.ShowModal() def set_selected_from_menu(self, menu, id): """ Set selected_plottable from context menu selection :param menu: context menu item :param id: menu item id """ if len(self.plots) < 1: return name = menu.GetHelpString(id) for plot in self.plots.values(): if plot.name == name: self.graph.selected_plottable = plot.id break def linear_plottable_fit(self, plot): """ when clicking on linear Fit on context menu, display Fitting Dialog :param plot: PlotPanel owning the graph """ from fitDialog import LinearFit if self._fit_dialog is not None: return self._fit_dialog = LinearFit(None, plot, self.onFitDisplay, self.returnTrans, -1, 'Linear Fit') # Set the zoom area if self._scale_xhi is not None and self._scale_xlo is not None: self._fit_dialog.set_fit_region(self._scale_xlo, self._scale_xhi) # Register the close event self._fit_dialog.register_close(self._linear_fit_close) # Show a non-model dialog self._fit_dialog.Show() def _linear_fit_close(self): """ A fit dialog was closed """ self._fit_dialog = None def _onProperties(self, event): """ when clicking on Properties on context menu , The Property dialog is displayed The user selects a transformation for x or y value and a new plot is displayed """ if self._fit_dialog is not None: self._fit_dialog.Destroy() self._fit_dialog = None plot_list = self.graph.returnPlottable() if len(plot_list.keys()) > 0: first_item = plot_list.keys()[0] if first_item.x != []: from PropertyDialog import Properties dial = Properties(self, -1, 'Properties') dial.setValues(self.prevXtrans, self.prevYtrans, self.viewModel) if dial.ShowModal() == wx.ID_OK: self.xLabel, self.yLabel, self.viewModel = dial.getValues() if self.viewModel == "Linear y vs x": self.xLabel = "x" self.yLabel = "y" self.viewModel = "--" dial.setValues(self.xLabel, self.yLabel, self.viewModel) if self.viewModel == "Guinier lny vs x^(2)": self.xLabel = "x^(2)" self.yLabel = "ln(y)" self.viewModel = "--" dial.setValues(self.xLabel, self.yLabel, self.viewModel) if self.viewModel == "XS Guinier ln(y*x) vs x^(2)": self.xLabel = "x^(2)" self.yLabel = "ln(y*x)" self.viewModel = "--" dial.setValues(self.xLabel, self.yLabel, self.viewModel) if self.viewModel == "Porod y*x^(4) vs x^(4)": self.xLabel = "x^(4)" self.yLabel = "y*x^(4)" self.viewModel = "--" dial.setValues(self.xLabel, self.yLabel, self.viewModel) self._onEVT_FUNC_PROPERTY() dial.Destroy() def set_yscale(self, scale='linear'): """ Set the scale on Y-axis :param scale: the scale of y-axis """ self.subplot.set_yscale(scale, nonposy='clip') self.yscale = scale def get_yscale(self): """ :return: Y-axis scale """ return self.yscale def set_xscale(self, scale='linear'): """ Set the scale on x-axis :param scale: the scale of x-axis """ self.subplot.set_xscale(scale) self.xscale = scale def get_xscale(self): """ :return: x-axis scale """ return self.xscale def SetColor(self, rgbtuple): """ Set figure and canvas colours to be the same """ if not rgbtuple: rgbtuple = wx.SystemSettings.GetColour(wx.SYS_COLOUR_BTNFACE).Get() col = [c / 255.0 for c in rgbtuple] self.figure.set_facecolor(col) self.figure.set_edgecolor(self.color) self.canvas.SetBackgroundColour(wx.Colour(*rgbtuple)) def _onSize(self, event): """ """ self._resizeflag = True def _onIdle(self, evt): """ """ if self._resizeflag: self._resizeflag = False self._SetSize() self.draw() def _SetSize(self, pixels=None): """ This method can be called to force the Plot to be a desired size, which defaults to the ClientSize of the panel """ if not pixels: pixels = tuple(self.GetClientSize()) self.canvas.SetSize(pixels) self.figure.set_size_inches(float(pixels[0]) / self.figure.get_dpi(), float(pixels[1]) / self.figure.get_dpi()) def draw(self): """ Where the actual drawing happens """ self.figure.canvas.draw_idle() def legend_picker(self, legend, event): """ Pick up the legend patch """ return self.legend.legendPatch.contains(event) def get_loc_label(self): """ Associates label to a specific legend location """ _labels = {} i = 0 _labels['best'] = i i += 1 _labels['upper right'] = i i += 1 _labels['upper left'] = i i += 1 _labels['lower left'] = i i += 1 _labels['lower right'] = i i += 1 _labels['right'] = i i += 1 _labels['center left'] = i i += 1 _labels['center right'] = i i += 1 _labels['lower center'] = i i += 1 _labels['upper center'] = i i += 1 _labels['center'] = i return _labels def onSaveImage(self, evt): """ Implement save image """ self.toolbar.save_figure(evt) def onContextMenu(self, event): """ Default context menu for a plot panel """ # Slicer plot popup menu id = wx.NewId() slicerpop = wx.Menu() slicerpop.Append(id, '&Save image', 'Save image as PNG') wx.EVT_MENU(self, id, self.onSaveImage) id = wx.NewId() slicerpop.Append(id, '&Printer setup', 'Set image size') wx.EVT_MENU(self, id, self.onPrinterSetup) id = wx.NewId() slicerpop.Append(id, '&Printer Preview', 'Set image size') wx.EVT_MENU(self, id, self.onPrinterPreview) id = wx.NewId() slicerpop.Append(id, '&Print image', 'Print image ') wx.EVT_MENU(self, id, self.onPrint) id = wx.NewId() slicerpop.Append(id, '&Copy', 'Copy to the clipboard') wx.EVT_MENU(self, id, self.OnCopyFigureMenu) #id = wx.NewId() #slicerpop.Append(id, '&Load 1D data file') #wx.EVT_MENU(self, id, self._onLoad1DData) id = wx.NewId() slicerpop.AppendSeparator() slicerpop.Append(id, '&Properties') wx.EVT_MENU(self, id, self._onProperties) id = wx.NewId() slicerpop.AppendSeparator() slicerpop.Append(id, '&Linear Fit') wx.EVT_MENU(self, id, self.onFitting) id = wx.NewId() slicerpop.AppendSeparator() slicerpop.Append(id, '&Toggle Legend On/Off', 'Toggle Legend On/Off') wx.EVT_MENU(self, id, self.onLegend) loc_menu = wx.Menu() for label in self._loc_labels: id = wx.NewId() loc_menu.Append(id, str(label), str(label)) wx.EVT_MENU(self, id, self.onChangeLegendLoc) id = wx.NewId() slicerpop.AppendMenu(id, '&Modify Legend Location', loc_menu) id = wx.NewId() slicerpop.Append(id, '&Modify Y Axis Label') wx.EVT_MENU(self, id, self._on_yaxis_label) id = wx.NewId() slicerpop.Append(id, '&Modify X Axis Label') wx.EVT_MENU(self, id, self._on_xaxis_label) try: # mouse event pos_evt = event.GetPosition() pos = self.ScreenToClient(pos_evt) except: # toolbar event pos_x, pos_y = self.toolbar.GetPositionTuple() pos = (pos_x, pos_y + 5) self.PopupMenu(slicerpop, pos) def onToolContextMenu(self, event): """ ContextMenu from toolbar :param event: toolbar event """ # reset postion self.position = None if self.graph.selected_plottable != None: self.graph.selected_plottable = None self.onContextMenu(event) # modified kieranrcampbell ILL june2012 def onLegend(self, legOnOff): """ Toggles whether legend is visible/not visible """ self.legend_on = legOnOff if not self.legend_on: for ax in self.axes: self.remove_legend(ax) else: # sort them by labels handles, labels = self.subplot.get_legend_handles_labels() hl = sorted(zip(handles, labels), key=operator.itemgetter(1)) handles2, labels2 = zip(*hl) self.line_collections_list = handles2 self.legend = self.subplot.legend(handles2, labels2, prop=FontProperties(size=10), loc=self.legendLoc) if self.legend != None: self.legend.set_picker(self.legend_picker) self.legend.set_axes(self.subplot) self.legend.set_zorder(20) self.subplot.figure.canvas.draw_idle() def onChangeLegendLoc(self, event): """ Changes legend loc based on user input """ menu = event.GetEventObject() label = menu.GetLabel(event.GetId()) self.ChangeLegendLoc(label) def ChangeLegendLoc(self, label): """ Changes legend loc based on user input """ self.legendLoc = label self.legend_pos_loc = None # sort them by labels handles, labels = self.subplot.get_legend_handles_labels() hl = sorted(zip(handles, labels), key=operator.itemgetter(1)) handles2, labels2 = zip(*hl) self.line_collections_list = handles2 self.legend = self.subplot.legend(handles2, labels2, prop=FontProperties(size=10), loc=self.legendLoc) if self.legend != None: self.legend.set_picker(self.legend_picker) self.legend.set_axes(self.subplot) self.legend.set_zorder(20) self.subplot.figure.canvas.draw_idle() def remove_legend(self, ax=None): """ Remove legend for ax or the current axes. """ from pylab import gca if ax is None: ax = gca() ax.legend_ = None def _on_addtext(self, event): """ Allows you to add text to the plot """ pos_x = 0 pos_y = 0 if self.position != None: pos_x, pos_y = self.position else: pos_x, pos_y = 0.01, 1.00 textdial = TextDialog(None, -1, 'Add Custom Text') if textdial.ShowModal() == wx.ID_OK: try: FONT = FontProperties() label = textdial.getText() xpos = pos_x ypos = pos_y font = FONT.copy() font.set_size(textdial.getSize()) font.set_family(textdial.getFamily()) font.set_style(textdial.getStyle()) font.set_weight(textdial.getWeight()) colour = textdial.getColor() if len(label) > 0 and xpos > 0 and ypos > 0: new_text = self.subplot.text(str(xpos), str(ypos), label, fontproperties=font, color=colour) self.textList.append(new_text) self.subplot.figure.canvas.draw_idle() except: if self.parent != None: msg = "Add Text: Error. Check your property values..." wx.PostEvent(self.parent, StatusEvent(status=msg)) else: raise textdial.Destroy() #Have a pop up box come up for user to type in the #text that they want to add...then create text Plottable #based on this and plot it at user designated coordinates def onGridOnOff(self, gridon_off): """ Allows ON/OFF Grid """ self.grid_on = gridon_off self.subplot.figure.canvas.draw_idle() def _on_xaxis_label(self, event): """ Allows you to add text to the plot """ xaxis_label, xaxis_unit, xaxis_font, xaxis_color, \ is_ok, is_tick = self._on_axis_label(axis='x') if not is_ok: return self.xaxis_label = xaxis_label self.xaxis_unit = xaxis_unit self.xaxis_font = xaxis_font self.xaxis_color = xaxis_color if is_tick: self.xaxis_tick = xaxis_font if self.data != None: # 2D self.xaxis(self.xaxis_label, self.xaxis_unit, \ self.xaxis_font, self.xaxis_color, self.xaxis_tick) self.subplot.figure.canvas.draw_idle() else: # 1D self._check_zoom_plot() def _check_zoom_plot(self): """ Check the zoom range and plot (1D only) """ xlo, xhi = self.subplot.get_xlim() ylo, yhi = self.subplot.get_ylim() ## Set the view scale for all plots self._onEVT_FUNC_PROPERTY(False) if self.is_zoomed: # Recover the x,y limits self.subplot.set_xlim((xlo, xhi)) self.subplot.set_ylim((ylo, yhi)) @property def is_zoomed(self): toolbar_zoomed = self.toolbar.GetToolEnabled(self.toolbar.wx_ids['Back']) return self._is_zoomed or toolbar_zoomed @is_zoomed.setter def is_zoomed(self, value): self._is_zoomed = value def _on_yaxis_label(self, event): """ Allows you to add text to the plot """ yaxis_label, yaxis_unit, yaxis_font, yaxis_color, \ is_ok, is_tick = self._on_axis_label(axis='y') if not is_ok: return self.yaxis_label = yaxis_label self.yaxis_unit = yaxis_unit self.yaxis_font = yaxis_font self.yaxis_color = yaxis_color if is_tick: self.yaxis_tick = yaxis_font if self.data != None: # 2D self.yaxis(self.yaxis_label, self.yaxis_unit, \ self.yaxis_font, self.yaxis_color, self.yaxis_tick) self.subplot.figure.canvas.draw_idle() else: # 1D self._check_zoom_plot() def _on_axis_label(self, axis='x'): """ Modify axes labels :param axis: x or y axis [string] """ is_ok = True title = 'Modify %s axis label' % axis font = 'serif' colour = 'black' if axis == 'x': label = self.xaxis_label unit = self.xaxis_unit else: label = self.yaxis_label unit = self.yaxis_unit textdial = TextDialog(None, -1, title, label, unit) if textdial.ShowModal() == wx.ID_OK: try: FONT = FontProperties() font = FONT.copy() font.set_size(textdial.getSize()) font.set_family(textdial.getFamily()) font.set_style(textdial.getStyle()) font.set_weight(textdial.getWeight()) unit = textdial.getUnit() colour = textdial.getColor() is_tick = textdial.getTickLabel() label_temp = textdial.getText() if label_temp.count("\%s" % "\\") > 0: if self.parent != None: msg = "Add Label: Error. Can not use double '\\' " msg += "characters..." wx.PostEvent(self.parent, StatusEvent(status=msg)) else: label = label_temp except: if self.parent != None: msg = "Add Label: Error. Check your property values..." wx.PostEvent(self.parent, StatusEvent(status=msg)) else: pass else: is_ok = False is_tick = True textdial.Destroy() return label, unit, font, colour, is_ok, is_tick def _on_removetext(self, event): """ Removes all text from the plot. Eventually, add option to remove specific text boxes """ num_text = len(self.textList) if num_text < 1: if self.parent != None: msg = "Remove Text: Nothing to remove. " wx.PostEvent(self.parent, StatusEvent(status=msg)) else: raise return txt = self.textList[num_text - 1] try: text_remove = txt.get_text() txt.remove() if self.parent != None: msg = "Removed Text: '%s'. " % text_remove wx.PostEvent(self.parent, StatusEvent(status=msg)) except: if self.parent != None: msg = "Remove Text: Error occurred. " wx.PostEvent(self.parent, StatusEvent(status=msg)) else: raise self.textList.remove(txt) self.subplot.figure.canvas.draw_idle() def properties(self, prop): """ Set some properties of the graph. The set of properties is not yet determined. """ # The particulars of how they are stored and manipulated (e.g., do # we want an inventory internally) is not settled. I've used a # property dictionary for now. # # How these properties interact with a user defined style file is # even less clear. # Properties defined by plot # Ricardo: # A empty label "$$" will prevent the panel from displaying! if prop["xlabel"]: self.subplot.set_xlabel(r"$%s$"%prop["xlabel"]) if prop["ylabel"]: self.subplot.set_ylabel(r"$%s$"%prop["ylabel"]) self.subplot.set_title(prop["title"]) def clear(self): """Reset the plot""" # TODO: Redraw is brutal. Render to a backing store and swap in # TODO: rather than redrawing on the fly. self.subplot.clear() self.subplot.hold(True) def render(self): """Commit the plot after all objects are drawn""" # TODO: this is when the backing store should be swapped in. if self.legend_on: ax = self.subplot ax.texts = self.textList try: handles, labels = ax.get_legend_handles_labels() # sort them by labels hl = sorted(zip(handles, labels), key=operator.itemgetter(1)) handles2, labels2 = zip(*hl) self.line_collections_list = handles2 self.legend = ax.legend(handles2, labels2, prop=FontProperties(size=10), loc=self.legendLoc) if self.legend != None: self.legend.set_picker(self.legend_picker) self.legend.set_axes(self.subplot) self.legend.set_zorder(20) except: self.legend = ax.legend(prop=FontProperties(size=10), loc=self.legendLoc) def xaxis(self, label, units, font=None, color='black', t_font=None): """xaxis label and units. Axis labels know about units. We need to do this so that we can detect when axes are not commesurate. Currently this is ignored other than for formatting purposes. """ self.xcolor = color if units.count("{") > 0 and units.count("$") < 2: units = '$' + units + '$' if label.count("{") > 0 and label.count("$") < 2: label = '$' + label + '$' if units != "": label = label + " (" + units + ")" if font: self.subplot.set_xlabel(label, fontproperties=font, color=color) if t_font != None: for tick in self.subplot.xaxis.get_major_ticks(): tick.label.set_fontproperties(t_font) for line in self.subplot.xaxis.get_ticklines(): size = t_font.get_size() line.set_markersize(size / 3) else: self.subplot.set_xlabel(label, color=color) pass def yaxis(self, label, units, font=None, color='black', t_font=None): """yaxis label and units.""" self.ycolor = color if units.count("{") > 0 and units.count("$") < 2: units = '$' + units + '$' if label.count("{") > 0 and label.count("$") < 2: label = '$' + label + '$' if units != "": label = label + " (" + units + ")" if font: self.subplot.set_ylabel(label, fontproperties=font, color=color) if t_font != None: for tick_label in self.subplot.get_yticklabels(): tick_label.set_fontproperties(t_font) for line in self.subplot.yaxis.get_ticklines(): size = t_font.get_size() line.set_markersize(size / 3) else: self.subplot.set_ylabel(label, color=color) pass def _connect_to_xlim(self, callback): """Bind the xlim change notification to the callback""" def process_xlim(axes): lo, hi = subplot.get_xlim() callback(lo, hi) self.subplot.callbacks.connect('xlim_changed', process_xlim) def interactive_points(self, x, y, dx=None, dy=None, name='', color=0, symbol=0, markersize=5, zorder=1, id=None, label=None, hide_error=False): """Draw markers with error bars""" self.subplot.set_yscale('linear') self.subplot.set_xscale('linear') if id is None: id = name from plottable_interactor import PointInteractor p = PointInteractor(self, self.subplot, zorder=zorder, id=id) if p.markersize != None: markersize = p.markersize p.points(x, y, dx=dx, dy=dy, color=color, symbol=symbol, zorder=zorder, markersize=markersize, label=label, hide_error=hide_error) self.subplot.set_yscale(self.yscale, nonposy='clip') self.subplot.set_xscale(self.xscale) def interactive_curve(self, x, y, dy=None, name='', color=0, symbol=0, zorder=1, id=None, label=None): """Draw markers with error bars""" self.subplot.set_yscale('linear') self.subplot.set_xscale('linear') if id is None: id = name from plottable_interactor import PointInteractor p = PointInteractor(self, self.subplot, zorder=zorder, id=id) p.curve(x, y, dy=dy, color=color, symbol=symbol, zorder=zorder, label=label) self.subplot.set_yscale(self.yscale, nonposy='clip') self.subplot.set_xscale(self.xscale) def plottable_selected(self, id): """ Called to register a plottable as selected """ #TODO: check that it really is in the list of plottables self.graph.selected_plottable = id def points(self, x, y, dx=None, dy=None, color=0, symbol=0, marker_size=5, label=None, id=None, hide_error=False): """Draw markers with error bars""" # Convert tuple (lo,hi) to array [(x-lo),(hi-x)] if dx != None and type(dx) == type(()): dx = nx.vstack((x - dx[0], dx[1] - x)).transpose() if dy != None and type(dy) == type(()): dy = nx.vstack((y - dy[0], dy[1] - y)).transpose() if dx == None and dy == None: self.subplot.plot(x, y, color=self._color(color), marker=self._symbol(symbol), markersize=marker_size, linestyle='', label=label) else: col = self._color(color) if hide_error: self.subplot.plot(x, y, color=col, marker=self._symbol(symbol), markersize=marker_size, linestyle='', label=label) else: self.subplot.errorbar(x, y, yerr=dy, xerr=None, ecolor=col, capsize=2, linestyle='', barsabove=False, mec=col, mfc=col, marker=self._symbol(symbol), markersize=marker_size, lolims=False, uplims=False, xlolims=False, xuplims=False, label=label) self.subplot.set_yscale(self.yscale, nonposy='clip') self.subplot.set_xscale(self.xscale) def _onToggleScale(self, event): """ toggle axis and replot image """ zmin_2D_temp = self.zmin_2D zmax_2D_temp = self.zmax_2D if self.scale == 'log_{10}': self.scale = 'linear' if not self.zmin_2D is None: zmin_2D_temp = math.pow(10, self.zmin_2D) if not self.zmax_2D is None: zmax_2D_temp = math.pow(10, self.zmax_2D) else: self.scale = 'log_{10}' if not self.zmin_2D is None: # min log value: no log(negative) if self.zmin_2D <= 0: zmin_2D_temp = -32 else: zmin_2D_temp = math.log10(self.zmin_2D) if not self.zmax_2D is None: zmax_2D_temp = math.log10(self.zmax_2D) self.image(data=self.data, qx_data=self.qx_data, qy_data=self.qy_data, xmin=self.xmin_2D, xmax=self.xmax_2D, ymin=self.ymin_2D, ymax=self.ymax_2D, cmap=self.cmap, zmin=zmin_2D_temp, zmax=zmax_2D_temp) def image(self, data, qx_data, qy_data, xmin, xmax, ymin, ymax, zmin, zmax, color=0, symbol=0, markersize=0, label='data2D', cmap=DEFAULT_CMAP): """ Render the current data """ self.data = data self.qx_data = qx_data self.qy_data = qy_data self.xmin_2D = xmin self.xmax_2D = xmax self.ymin_2D = ymin self.ymax_2D = ymax self.zmin_2D = zmin self.zmax_2D = zmax c = self._color(color) # If we don't have any data, skip. if self.data == None: return if self.data.ndim == 1: output = self._build_matrix() else: output = copy.deepcopy(self.data) # check scale if self.scale == 'log_{10}': try: if self.zmin_2D <= 0 and len(output[output > 0]) > 0: zmin_temp = self.zmin_2D output[output > 0] = numpy.log10(output[output > 0]) #In log scale Negative values are not correct in general #output[output<=0] = math.log(numpy.min(output[output>0])) elif self.zmin_2D <= 0: zmin_temp = self.zmin_2D output[output > 0] = numpy.zeros(len(output)) output[output <= 0] = -32 else: zmin_temp = self.zmin_2D output[output > 0] = numpy.log10(output[output > 0]) #In log scale Negative values are not correct in general #output[output<=0] = math.log(numpy.min(output[output>0])) except: #Too many problems in 2D plot with scale pass else: zmin_temp = self.zmin_2D self.cmap = cmap if self.dimension != 3: #Re-adjust colorbar self.subplot.figure.subplots_adjust(left=0.2, right=.8, bottom=.2) im = self.subplot.imshow(output, interpolation='nearest', origin='lower', vmin=zmin_temp, vmax=self.zmax_2D, cmap=self.cmap, extent=(self.xmin_2D, self.xmax_2D, self.ymin_2D, self.ymax_2D)) cbax = self.subplot.figure.add_axes([0.84, 0.2, 0.02, 0.7]) else: # clear the previous 2D from memory # mpl is not clf, so we do self.subplot.figure.clear() self.subplot.figure.subplots_adjust(left=0.1, right=.8, bottom=.1) X = self.x_bins[0:-1] Y = self.y_bins[0:-1] X, Y = numpy.meshgrid(X, Y) try: # mpl >= 1.0.0 ax = self.subplot.figure.gca(projection='3d') #ax.disable_mouse_rotation() cbax = self.subplot.figure.add_axes([0.84, 0.1, 0.02, 0.8]) if len(X) > 60: ax.disable_mouse_rotation() except: # mpl < 1.0.0 try: from mpl_toolkits.mplot3d import Axes3D except: logging.error("PlotPanel could not import Axes3D") self.subplot.figure.clear() ax = Axes3D(self.subplot.figure) if len(X) > 60: ax.cla() cbax = None self.subplot.figure.canvas.resizing = False im = ax.plot_surface(X, Y, output, rstride=1, cstride=1, cmap=cmap, linewidth=0, antialiased=False) self.subplot.set_axis_off() if cbax == None: ax.set_frame_on(False) cb = self.subplot.figure.colorbar(im, shrink=0.8, aspect=20) else: cb = self.subplot.figure.colorbar(im, cax=cbax) cb.update_bruteforce(im) cb.set_label('$' + self.scale + '$') if self.dimension != 3: self.figure.canvas.draw_idle() else: self.figure.canvas.draw() def _build_matrix(self): """ Build a matrix for 2d plot from a vector Returns a matrix (image) with ~ square binning Requirement: need 1d array formats of self.data, self.qx_data, and self.qy_data where each one corresponds to z, x, or y axis values """ # No qx or qy given in a vector format if self.qx_data == None or self.qy_data == None \ or self.qx_data.ndim != 1 or self.qy_data.ndim != 1: # do we need deepcopy here? return copy.deepcopy(self.data) # maximum # of loops to fillup_pixels # otherwise, loop could never stop depending on data max_loop = 1 # get the x and y_bin arrays. self._get_bins() # set zero to None #Note: Can not use scipy.interpolate.Rbf: # 'cause too many data points (>10000)<=JHC. # 1d array to use for weighting the data point averaging #when they fall into a same bin. weights_data = numpy.ones([self.data.size]) # get histogram of ones w/len(data); this will provide #the weights of data on each bins weights, xedges, yedges = numpy.histogram2d(x=self.qy_data, y=self.qx_data, bins=[self.y_bins, self.x_bins], weights=weights_data) # get histogram of data, all points into a bin in a way of summing image, xedges, yedges = numpy.histogram2d(x=self.qy_data, y=self.qx_data, bins=[self.y_bins, self.x_bins], weights=self.data) # Now, normalize the image by weights only for weights>1: # If weight == 1, there is only one data point in the bin so # that no normalization is required. image[weights > 1] = image[weights > 1] / weights[weights > 1] # Set image bins w/o a data point (weight==0) as None (was set to zero # by histogram2d.) image[weights == 0] = None # Fill empty bins with 8 nearest neighbors only when at least #one None point exists loop = 0 # do while loop until all vacant bins are filled up up #to loop = max_loop while not(numpy.isfinite(image[weights == 0])).all(): if loop >= max_loop: # this protects never-ending loop break image = self._fillup_pixels(image=image, weights=weights) loop += 1 return image def _get_bins(self): """ get bins set x_bins and y_bins into self, 1d arrays of the index with ~ square binning Requirement: need 1d array formats of self.qx_data, and self.qy_data where each one corresponds to x, or y axis values """ # No qx or qy given in a vector format if self.qx_data == None or self.qy_data == None \ or self.qx_data.ndim != 1 or self.qy_data.ndim != 1: # do we need deepcopy here? return copy.deepcopy(self.data) # find max and min values of qx and qy xmax = self.qx_data.max() xmin = self.qx_data.min() ymax = self.qy_data.max() ymin = self.qy_data.min() # calculate the range of qx and qy: this way, it is a little # more independent x_size = xmax - xmin y_size = ymax - ymin # estimate the # of pixels on each axes npix_y = int(math.floor(math.sqrt(len(self.qy_data)))) npix_x = int(math.floor(len(self.qy_data) / npix_y)) # bin size: x- & y-directions xstep = x_size / (npix_x - 1) ystep = y_size / (npix_y - 1) # max and min taking account of the bin sizes xmax = xmax + xstep / 2.0 xmin = xmin - xstep / 2.0 ymax = ymax + ystep / 2.0 ymin = ymin - ystep / 2.0 # store x and y bin centers in q space x_bins = numpy.linspace(xmin, xmax, npix_x) y_bins = numpy.linspace(ymin, ymax, npix_y) #set x_bins and y_bins self.x_bins = x_bins self.y_bins = y_bins def _fillup_pixels(self, image=None, weights=None): """ Fill z values of the empty cells of 2d image matrix with the average over up-to next nearest neighbor points :param image: (2d matrix with some zi = None) :return: image (2d array ) :TODO: Find better way to do for-loop below """ # No image matrix given if image == None or numpy.ndim(image) != 2 \ or numpy.isfinite(image).all() \ or weights == None: return image # Get bin size in y and x directions len_y = len(image) len_x = len(image[1]) temp_image = numpy.zeros([len_y, len_x]) weit = numpy.zeros([len_y, len_x]) # do for-loop for all pixels for n_y in range(len(image)): for n_x in range(len(image[1])): # find only null pixels if weights[n_y][n_x] > 0 or numpy.isfinite(image[n_y][n_x]): continue else: # find 4 nearest neighbors # check where or not it is at the corner if n_y != 0 and numpy.isfinite(image[n_y - 1][n_x]): temp_image[n_y][n_x] += image[n_y - 1][n_x] weit[n_y][n_x] += 1 if n_x != 0 and numpy.isfinite(image[n_y][n_x - 1]): temp_image[n_y][n_x] += image[n_y][n_x - 1] weit[n_y][n_x] += 1 if n_y != len_y - 1 and numpy.isfinite(image[n_y + 1][n_x]): temp_image[n_y][n_x] += image[n_y + 1][n_x] weit[n_y][n_x] += 1 if n_x != len_x - 1 and numpy.isfinite(image[n_y][n_x + 1]): temp_image[n_y][n_x] += image[n_y][n_x + 1] weit[n_y][n_x] += 1 # go 4 next nearest neighbors when no non-zero # neighbor exists if n_y != 0 and n_x != 0 and\ numpy.isfinite(image[n_y - 1][n_x - 1]): temp_image[n_y][n_x] += image[n_y - 1][n_x - 1] weit[n_y][n_x] += 1 if n_y != len_y - 1 and n_x != 0 and \ numpy.isfinite(image[n_y + 1][n_x - 1]): temp_image[n_y][n_x] += image[n_y + 1][n_x - 1] weit[n_y][n_x] += 1 if n_y != len_y and n_x != len_x - 1 and \ numpy.isfinite(image[n_y - 1][n_x + 1]): temp_image[n_y][n_x] += image[n_y - 1][n_x + 1] weit[n_y][n_x] += 1 if n_y != len_y - 1 and n_x != len_x - 1 and \ numpy.isfinite(image[n_y + 1][n_x + 1]): temp_image[n_y][n_x] += image[n_y + 1][n_x + 1] weit[n_y][n_x] += 1 # get it normalized ind = (weit > 0) image[ind] = temp_image[ind] / weit[ind] return image def curve(self, x, y, dy=None, color=0, symbol=0, label=None): """Draw a line on a graph, possibly with confidence intervals.""" c = self._color(color) self.subplot.set_yscale('linear') self.subplot.set_xscale('linear') self.subplot.plot(x, y, color=c, marker='', linestyle='-', label=label) self.subplot.set_yscale(self.yscale) self.subplot.set_xscale(self.xscale) def _color(self, c): """Return a particular colour""" return self.colorlist[c % len(self.colorlist)] def _symbol(self, s): """Return a particular symbol""" return self.symbollist[s % len(self.symbollist)] def _replot(self, remove_fit=False): """ Rescale the plottables according to the latest user selection and update the plot :param remove_fit: Fit line will be removed if True """ self.graph.reset_scale() self._onEVT_FUNC_PROPERTY(remove_fit=remove_fit) #TODO: Why do we have to have the following line? self.fit_result.reset_view() self.graph.render(self) self.subplot.figure.canvas.draw_idle() def _onEVT_FUNC_PROPERTY(self, remove_fit=True, show=True): """ Receive the x and y transformation from myDialog, Transforms x and y in View and set the scale """ # The logic should be in the right order # Delete first, and then get the whole list... if remove_fit: self.graph.delete(self.fit_result) self.ly = None self.q_ctrl = None list = self.graph.returnPlottable() # Changing the scale might be incompatible with # currently displayed data (for instance, going # from ln to log when all plotted values have # negative natural logs). # Go linear and only change the scale at the end. self.set_xscale("linear") self.set_yscale("linear") _xscale = 'linear' _yscale = 'linear' for item in list: item.setLabel(self.xLabel, self.yLabel) # control axis labels from the panel itself yname, yunits = item.get_yaxis() if self.yaxis_label != None: yname = self.yaxis_label yunits = self.yaxis_unit else: self.yaxis_label = yname self.yaxis_unit = yunits xname, xunits = item.get_xaxis() if self.xaxis_label != None: xname = self.xaxis_label xunits = self.xaxis_unit else: self.xaxis_label = xname self.xaxis_unit = xunits # Goes through all possible scales if self.xLabel == "x": item.transformX(transform.toX, transform.errToX) self.graph._xaxis_transformed("%s" % xname, "%s" % xunits) if self.xLabel == "x^(2)": item.transformX(transform.toX2, transform.errToX2) xunits = convert_unit(2, xunits) self.graph._xaxis_transformed("%s^{2}" % xname, "%s" % xunits) if self.xLabel == "x^(4)": item.transformX(transform.toX4, transform.errToX4) xunits = convert_unit(4, xunits) self.graph._xaxis_transformed("%s^{4}" % xname, "%s" % xunits) if self.xLabel == "ln(x)": item.transformX(transform.toLogX, transform.errToLogX) self.graph._xaxis_transformed("\ln\\ %s" % xname, "%s" % xunits) if self.xLabel == "log10(x)": item.transformX(transform.toX_pos, transform.errToX_pos) _xscale = 'log' self.graph._xaxis_transformed("%s" % xname, "%s" % xunits) if self.xLabel == "log10(x^(4))": item.transformX(transform.toX4, transform.errToX4) xunits = convert_unit(4, xunits) self.graph._xaxis_transformed("%s^{4}" % xname, "%s" % xunits) _xscale = 'log' if self.yLabel == "ln(y)": item.transformY(transform.toLogX, transform.errToLogX) self.graph._yaxis_transformed("\ln\\ %s" % yname, "%s" % yunits) if self.yLabel == "y": item.transformY(transform.toX, transform.errToX) self.graph._yaxis_transformed("%s" % yname, "%s" % yunits) if self.yLabel == "log10(y)": item.transformY(transform.toX_pos, transform.errToX_pos) _yscale = 'log' self.graph._yaxis_transformed("%s" % yname, "%s" % yunits) if self.yLabel == "y^(2)": item.transformY(transform.toX2, transform.errToX2) yunits = convert_unit(2, yunits) self.graph._yaxis_transformed("%s^{2}" % yname, "%s" % yunits) if self.yLabel == "1/y": item.transformY(transform.toOneOverX, transform.errOneOverX) yunits = convert_unit(-1, yunits) self.graph._yaxis_transformed("1/%s" % yname, "%s" % yunits) if self.yLabel == "y*x^(4)": item.transformY(transform.toYX4, transform.errToYX4) xunits = convert_unit(4, self.xaxis_unit) self.graph._yaxis_transformed("%s \ \ %s^{4}" % (yname, xname), "%s%s" % (yunits, xunits)) if self.yLabel == "1/sqrt(y)": item.transformY(transform.toOneOverSqrtX, transform.errOneOverSqrtX) yunits = convert_unit(-0.5, yunits) self.graph._yaxis_transformed("1/\sqrt{%s}" % yname, "%s" % yunits) if self.yLabel == "ln(y*x)": item.transformY(transform.toLogXY, transform.errToLogXY) self.graph._yaxis_transformed("\ln (%s \ \ %s)" % (yname, xname), "%s%s" % (yunits, self.xaxis_unit)) if self.yLabel == "ln(y*x^(2))": item.transformY(transform.toLogYX2, transform.errToLogYX2) xunits = convert_unit(2, self.xaxis_unit) self.graph._yaxis_transformed("\ln (%s \ \ %s^{2})" % (yname, xname), "%s%s" % (yunits, xunits)) if self.yLabel == "ln(y*x^(4))": item.transformY(transform.toLogYX4, transform.errToLogYX4) xunits = convert_unit(4, self.xaxis_unit) self.graph._yaxis_transformed("\ln (%s \ \ %s^{4})" % (yname, xname), "%s%s" % (yunits, xunits)) if self.yLabel == "log10(y*x^(4))": item.transformY(transform.toYX4, transform.errToYX4) xunits = convert_unit(4, self.xaxis_unit) _yscale = 'log' self.graph._yaxis_transformed("%s \ \ %s^{4}" % (yname, xname), "%s%s" % (yunits, xunits)) if self.viewModel == "Guinier lny vs x^(2)": item.transformX(transform.toX2, transform.errToX2) xunits = convert_unit(2, xunits) self.graph._xaxis_transformed("%s^{2}" % xname, "%s" % xunits) item.transformY(transform.toLogX, transform.errToLogX) self.graph._yaxis_transformed("\ln\ \ %s" % yname, "%s" % yunits) if self.viewModel == "Porod y*x^(4) vs x^(4)": item.transformX(transform.toX4, transform.errToX4) xunits = convert_unit(4, self.xaxis_unit) self.graph._xaxis_transformed("%s^{4}" % xname, "%s" % xunits) item.transformY(transform.toYX4, transform.errToYX4) self.graph._yaxis_transformed("%s \ \ %s^{4}" % (yname, xname), "%s%s" % (yunits, xunits)) item.transformView() # set new label and units yname = self.graph.prop["ylabel"] yunits = '' xname = self.graph.prop["xlabel"] xunits = '' self.resetFitView() self.prevXtrans = self.xLabel self.prevYtrans = self.yLabel self.graph.render(self) self.set_xscale(_xscale) self.set_yscale(_yscale) self.xaxis(xname, xunits, self.xaxis_font, self.xaxis_color, self.xaxis_tick) self.yaxis(yname, yunits, self.yaxis_font, self.yaxis_color, self.yaxis_tick) self.subplot.texts = self.textList if show: self.subplot.figure.canvas.draw_idle() def onFitDisplay(self, tempx, tempy, xminView, xmaxView, xmin, xmax, func): """ Add a new plottable into the graph .In this case this plottable will be used to fit some data :param tempx: The x data of fit line :param tempy: The y data of fit line :param xminView: the lower bound of fitting range :param xminView: the upper bound of fitting range :param xmin: the lowest value of data to fit to the line :param xmax: the highest value of data to fit to the line """ # Saving value to redisplay in Fit Dialog when it is opened again self.Avalue, self.Bvalue, self.ErrAvalue, \ self.ErrBvalue, self.Chivalue = func self.xminView = xminView self.xmaxView = xmaxView self.xmin = xmin self.xmax = xmax #In case need to change the range of data plotted list = [] list = self.graph.returnPlottable() for item in list: #item.onFitRange(xminView,xmaxView) item.onFitRange(None, None) # Create new data plottable with result self.fit_result.x = [] self.fit_result.y = [] self.fit_result.x = tempx self.fit_result.y = tempy self.fit_result.dx = None self.fit_result.dy = None #Load the view with the new values self.fit_result.reset_view() # Add the new plottable to the graph self.graph.add(self.fit_result) self.graph.render(self) self._offset_graph() self.subplot.figure.canvas.draw_idle() def onChangeCaption(self, event): """ """ if self.parent == None: return # get current caption old_caption = self.window_caption # Get new caption dialog dial = LabelDialog(None, -1, 'Modify Window Title', old_caption) if dial.ShowModal() == wx.ID_OK: new_caption = dial.getText() # send to guiframe to change the panel caption caption = self.parent.on_change_caption(self.window_name, old_caption, new_caption) # also set new caption in plot_panels list self.parent.plot_panels[self.uid].window_caption = caption # set new caption self.window_caption = caption dial.Destroy() def onResetGraph(self, event): """ Reset the graph by plotting the full range of data """ list = [] list = self.graph.returnPlottable() for item in list: item.onReset() self.graph.render(self) self._onEVT_FUNC_PROPERTY(False) self.is_zoomed = False self.toolbar.update() def onPrinterSetup(self, event=None): """ """ self.canvas.Printer_Setup(event=event) self.Update() def onPrinterPreview(self, event=None): """ Matplotlib camvas can no longer print itself. Thus need to do everything ourselves: need to create a printpreview frame to to see the preview but needs a parent frame object. Also needs a printout object (just as any printing task). """ try: #check if parent is a frame. If not keep getting the higher #parent till we find a frame _plot = self while not isinstance(_plot, wx.Frame): _plot = _plot.GetParent() assert _plot is not None #now create the printpeview object _preview = wx.PrintPreview(PlotPrintout(self,'test'), PlotPrintout(self,'test')) #and tie it to a printpreview frame then show it _frame = wx.PreviewFrame(_preview, _plot, "Print Preview", wx.Point(100, 100), wx.Size(600, 650)) _frame.Centre(wx.BOTH) _frame.Initialize() _frame.Show(True) except: traceback.print_exc() pass def onPrint(self, event=None): """ Matplotlib canvas can no longer print itself so using wx.Printer. Need therefore to also define a Printout object. """ try: _printer = wx.Printer() _printer.Print(self.canvas, PlotPrintout(self,'test'), True) except: traceback.print_exc() pass def OnCopyFigureMenu(self, evt): """ Copy the current figure to clipboard """ try: CopyImage(self.canvas) except: print "Error in copy Image" #--------------------------------------------------------------- class NoRepaintCanvas(FigureCanvasWxAgg): """ We subclass FigureCanvasWxAgg, overriding the _onPaint method, so that the draw method is only called for the first two paint events. After that, the canvas will only be redrawn when it is resized. """ def __init__(self, *args, **kwargs): """ """ FigureCanvasWxAgg.__init__(self, *args, **kwargs) self._drawn = 0 def _onPaint(self, evt): """ Called when wxPaintEvt is generated """ if not self._isRealized: self.realize() if self._drawn < 2: self.draw(repaint=False) self._drawn += 1 self.gui_repaint(drawDC=wx.PaintDC(self)) class PlotPrintout(wx.Printout): """ Create the wx.Printout object for matplotlib figure from the PlotPanel. Provides the required OnPrintPage and HasPage overrides. Other methods may be added/overriden in the future. :TODO: this needs LOTS of TLC .. but fixes immediate problem """ def __init__(self, figure, title, size=None, dpi=None, **kwargs): """ Initialize wx.Printout and get passed figure object """ wx.Printout.__init__(self) self.figure = figure def OnPrintPage(self, page): """ Most rudimentry OnPrintPage overide. instatiates a dc object, gets its size, gets the size of the figure object, scales it to the dc canvas size keeping the aspect ratio intact, then prints as bitmap """ _dc = self.GetDC() (_dcX, _dcY) = _dc.GetSizeTuple() (_bmpX,_bmpY) = self.figure.canvas.GetSize() _scale = min(_dcX/_bmpX, _dcY/_bmpY) _dc.SetUserScale(_scale, _scale) _dc.DrawBitmap(self.figure.canvas.bitmap, 0, 0, False,) return True def GetPageInfo(self): """ just sets the page to 1 - no flexibility for now """ return (1, 1, 1, 1)