1220 lines
46 KiB
Python
1220 lines
46 KiB
Python
import pygtk
|
|
import gobject
|
|
import gtk
|
|
|
|
import matplotlib
|
|
from matplotlib.backends.backend_gtkagg import FigureCanvasGTKAgg
|
|
from matplotlib.figure import Figure
|
|
from matplotlib.collections import LineCollection
|
|
from matplotlib.patches import Polygon,Rectangle,Circle
|
|
from matplotlib.lines import Line2D
|
|
from matplotlib.mlab import prctile
|
|
from matplotlib.colors import ColorConverter
|
|
import networkx
|
|
import scipy
|
|
from numpy import matlib
|
|
|
|
import laydi
|
|
import logger
|
|
import view
|
|
|
|
def plotlogger(func, name=None):
|
|
def wrapped(parent, *args, **kw):
|
|
parent.__args = args
|
|
parent.__kw = kw
|
|
return func(parent, *args, **kw)
|
|
return wrapped
|
|
|
|
|
|
class Plot(view.View):
|
|
def __init__(self, title):
|
|
view.View.__init__(self, title)
|
|
logger.log('debug', 'plot %s init' %title)
|
|
self.selection_listener = None
|
|
self.current_dim = None
|
|
self._current_selection = None
|
|
self._frozen = False
|
|
self._init_mpl()
|
|
|
|
def _init_mpl(self):
|
|
# init matplotlib related stuff
|
|
self._background = None
|
|
self._colorbar = None
|
|
self._mappable = None
|
|
self._use_blit = False
|
|
self.fig = Figure()
|
|
self.canvas = FigureCanvasGTKAgg(self.fig)
|
|
self.axes = self.fig.gca()
|
|
self._toolbar = view.PlotToolbar(self)
|
|
self._key_press = self.canvas.mpl_connect(
|
|
'key_press_event', self.on_key_press)
|
|
self.canvas.add_events(gtk.gdk.ENTER_NOTIFY_MASK)
|
|
self.add(self.canvas)
|
|
self.canvas.show()
|
|
|
|
def set_frozen(self, frozen):
|
|
"""A frozen plot will not be updated when the current
|
|
selection is changed."""
|
|
self._frozen = frozen
|
|
if not frozen and self._current_selection != None:
|
|
self.set_current_selection(self._current_selection)
|
|
|
|
def get_title(self):
|
|
return self.title
|
|
|
|
def get_toolbar(self):
|
|
return self._toolbar
|
|
|
|
def selection_changed(self, dim_name, selection):
|
|
""" Selection observer handle.
|
|
|
|
A selection change in a plot is only drawn if:
|
|
1.) plot is sensitive to selections (not freezed)
|
|
2.) plot is visible (has a view)
|
|
3.) the selections dim_name is the plot's dimension.
|
|
"""
|
|
|
|
self._current_selection = selection
|
|
if self._frozen \
|
|
or not self.get_property('visible') \
|
|
or self.current_dim != dim_name:
|
|
return
|
|
|
|
self.set_current_selection(selection)
|
|
|
|
def set_selection_listener(self, listener):
|
|
"""Allow project to listen to selections.
|
|
|
|
The selection will propagate back to all plots through the
|
|
selection_changed() method. The listener will be called as
|
|
listener(dimension_name, ids).
|
|
"""
|
|
self.selection_listener = listener
|
|
|
|
def update_selection(self, ids, key=None):
|
|
"""Returns updated current selection from ids.
|
|
If a key is pressed we use the appropriate mode.
|
|
|
|
key map:
|
|
shift : union
|
|
control : intersection
|
|
|
|
"""
|
|
if key == 'shift':
|
|
ids = set(ids).union(self._current_selection[self.current_dim])
|
|
elif key == 'control':
|
|
ids = set(ids).intersection(self._current_selection[self.current_dim])
|
|
return ids
|
|
|
|
def set_current_selection(self, selection):
|
|
"""Called whenever the plot should change the selection.
|
|
|
|
This method is a dummy method, so that specialized plots that have
|
|
no implemented selection can ignore selections alltogether.
|
|
"""
|
|
pass
|
|
|
|
def rectangle_select_callback(self, *args):
|
|
"""Overrriden in subclass."""
|
|
if hasattr(self, 'canvas'):
|
|
self.canvas.draw()
|
|
|
|
def lasso_select_callback(self, *args):
|
|
"""Overrriden in subclass."""
|
|
if hasattr(self, 'canvas'):
|
|
self.canvas.draw()
|
|
|
|
def get_index_from_selection(self, dataset, selection):
|
|
"""Returns the index vector of current selection in given dim."""
|
|
if not selection: return []
|
|
ids = selection.get(self.current_dim, []) # current identifiers
|
|
if not ids : return []
|
|
return dataset.get_indices(self.current_dim, ids)
|
|
|
|
def on_key_press(self, event):
|
|
if event.key == 'c':
|
|
self._toggle_colorbar()
|
|
|
|
def _toggle_colorbar(self):
|
|
if self._colorbar == None:
|
|
if self._mappable == None:
|
|
logger.log('notice', 'No mappable in this plot')
|
|
return
|
|
if self._mappable._A != None: # we need colormapping
|
|
# get axes original position
|
|
self._ax_last_pos = self.axes.get_position()
|
|
self._colorbar = self.fig.colorbar(self._mappable)
|
|
self._colorbar.draw_all()
|
|
self.canvas.draw()
|
|
else:
|
|
# remove colorbar
|
|
# remove, axes, observers, colorbar instance, and restore viewlims
|
|
cb, ax = self._mappable.colorbar
|
|
self.fig.delaxes(ax)
|
|
self._mappable.observers = [obs for obs in self._mappable.observers if obs !=self._colorbar]
|
|
self._colorbar = None
|
|
self.axes.set_position(self._ax_last_pos)
|
|
self.canvas.draw()
|
|
|
|
|
|
class LineViewPlot(Plot):
|
|
"""Line view plot with percentiles.
|
|
|
|
A line view of vectors across a specified dimension of input dataset.
|
|
No selection interaction is defined.
|
|
Only support for 2d-arrays.
|
|
input:
|
|
-- major_axis : dim_number for line dim (see scipy.ndarray for axis def.)
|
|
-- minor_axis : needs definition only for higher order arrays
|
|
|
|
fixme: slow
|
|
"""
|
|
@plotlogger
|
|
def __init__(self, dataset, major_axis=1, minor_axis=None, center=True,name="Line view"):
|
|
Plot.__init__(self, name)
|
|
self.dataset = dataset
|
|
self._data = dataset.asarray()
|
|
if len(self._data.shape)==2 and not minor_axis:
|
|
minor_axis = major_axis - 1
|
|
self.major_axis = major_axis
|
|
self.minor_axis = minor_axis
|
|
self.current_dim = self.dataset.get_dim_name(major_axis)
|
|
self.data_is_centered = False
|
|
self._mn_data = 0
|
|
if center and len(self._data.shape)==2:
|
|
if minor_axis==0:
|
|
self._mn_data = self._data.mean(minor_axis)
|
|
else:
|
|
self._mn_data = self._data.mean(minor_axis)[:,newaxis]
|
|
self._data = self._data - self._mn_data
|
|
self.data_is_centered = True
|
|
#initial line collection
|
|
self.line_coll = None
|
|
self.make_lines()
|
|
# draw background
|
|
self.set_background()
|
|
|
|
# Disable selection modes
|
|
self._toolbar.freeze_button.set_sensitive(False)
|
|
self._toolbar.set_mode_sensitive('select', False)
|
|
self._toolbar.set_mode_sensitive('lassoselect', False)
|
|
|
|
def make_lines(self):
|
|
"""Creates one line for each item along major axis."""
|
|
if self.line_coll: # remove any previous selection lines, if any
|
|
self.axes.collections.remove(self.line_coll)
|
|
self.line_coll = None
|
|
self.line_segs = []
|
|
x_axis = scipy.arange(self._data.shape[self.minor_axis])
|
|
for xi in range(self._data.shape[self.major_axis]):
|
|
yi = self._data.take([xi], self.major_axis).ravel()
|
|
self.line_segs.append([(xx,yy) for xx,yy in zip(x_axis, yi)])
|
|
|
|
def set_background(self):
|
|
"""Add three patches representing [min max],[5,95] and [25,75] percentiles, and a line at the median.
|
|
"""
|
|
if self._data.shape[self.minor_axis]<6:
|
|
return
|
|
# clean old patches if any
|
|
if len(self.axes.patches)>0:
|
|
self.axes.patches = []
|
|
# clean old lines (median) if any
|
|
if len(self.axes.lines)>0:
|
|
self.axes.lines = []
|
|
# settings
|
|
patch_color = 'b' #blue
|
|
patch_lw = 0 #no edges
|
|
patch_alpha = .15 # transparancy
|
|
median_color = 'b' #blue
|
|
median_width = 1.5 #linewidth
|
|
percentiles = [0, 5, 25, 50, 75, 100]
|
|
|
|
# ordinate
|
|
xax = scipy.arange(self._data.shape[self.minor_axis])
|
|
|
|
#vertices
|
|
verts_0 = [] #100,0
|
|
verts_1 = [] # 90,10
|
|
verts_2 = [] # 75,25
|
|
med = []
|
|
# add top vertices the low vertices (do i need an order?)#background
|
|
for i in xax:
|
|
prct = prctile(self._data.take([i], self.minor_axis), percentiles)
|
|
verts_0.append((i, prct[0]))
|
|
verts_1.append((i, prct[1]))
|
|
verts_2.append((i, prct[2]))
|
|
med.append(prct[3])
|
|
for i in xax[::-1]:
|
|
prct = prctile(self._data.take([i], self.minor_axis), percentiles)
|
|
verts_0.append((i, prct[-1]))
|
|
verts_1.append((i, prct[-2]))
|
|
verts_2.append((i, prct[-3]))
|
|
|
|
# make polygons from vertices
|
|
bck0 = Polygon(verts_0, alpha=patch_alpha, lw=patch_lw,
|
|
facecolor=patch_color)
|
|
bck1 = Polygon(verts_1, alpha=patch_alpha, lw=patch_lw,
|
|
facecolor=patch_color)
|
|
bck2 = Polygon(verts_2, alpha=patch_alpha, lw=patch_lw,
|
|
facecolor=patch_color)
|
|
|
|
# add polygons to axes
|
|
self.axes.add_patch(bck0)
|
|
self.axes.add_patch(bck1)
|
|
self.axes.add_patch(bck2)
|
|
# median line
|
|
self.axes.plot(xax, med, median_color, linewidth=median_width)
|
|
|
|
# set y-limits
|
|
padding = 0.1
|
|
self.axes.set_ylim([self._data.min() - padding, self._data.max() + padding])
|
|
|
|
def set_current_selection(self, selection):
|
|
"""Draws the current selection.
|
|
"""
|
|
index = self.get_index_from_selection(self.dataset, selection)
|
|
|
|
if self.line_coll:
|
|
self.axes.collections.remove(self.line_coll)
|
|
segs = [self.line_segs[i] for i in index]
|
|
self.line_coll = LineCollection(segs, colors=(1,0,0,1))
|
|
self.axes.add_collection(self.line_coll)
|
|
|
|
#draw
|
|
if self._use_blit:
|
|
if self._background is None:
|
|
self._background = self.canvas.copy_from_bbox(self.axes.bbox)
|
|
self.canvas.restore_region(self._background)
|
|
self.axes.draw_artist(self.lines)
|
|
self.canvas.blit()
|
|
else:
|
|
self.canvas.draw()
|
|
|
|
|
|
class ScatterMarkerPlot(Plot):
|
|
"""The ScatterMarkerPlot is faster than regular scatterplot, but
|
|
has no color and size options."""
|
|
@plotlogger
|
|
def __init__(self, dataset_1, dataset_2, id_dim, sel_dim,
|
|
id_1, id_2, s=6, name="Scatter plot"):
|
|
Plot.__init__(self, name)
|
|
self.current_dim = id_dim
|
|
self.dataset_1 = dataset_1
|
|
self.ms = s
|
|
x_index = dataset_1[sel_dim][id_1]
|
|
y_index = dataset_2[sel_dim][id_2]
|
|
self.xaxis_data = dataset_1.asarray()[:, x_index]
|
|
self.yaxis_data = dataset_2.asarray()[:, y_index]
|
|
|
|
# init draw
|
|
self._selection_line = None
|
|
self.line = self.axes.plot(self.xaxis_data,
|
|
self.yaxis_data, 'o',
|
|
markeredgewidth=0,
|
|
markersize=s)
|
|
self.axes.axhline(0, color='k', lw=1., zorder=1)
|
|
self.axes.axvline(0, color='k', lw=1., zorder=1)
|
|
|
|
def rectangle_select_callback(self, x1, y1, x2, y2, key):
|
|
ydata = self.yaxis_data
|
|
xdata = self.xaxis_data
|
|
|
|
# find indices of selected area
|
|
if x1>x2:
|
|
x1, x2 = x2, x1
|
|
if y1>y2:
|
|
y1, y2 = y2, y1
|
|
assert x1<=x2
|
|
assert y1<=y2
|
|
index = scipy.nonzero((xdata>x1) & (xdata<x2) & (ydata>y1) & (ydata<y2))[0]
|
|
ids = self.dataset_1.get_identifiers(self.current_dim, index)
|
|
ids = self.update_selection(ids, key)
|
|
self.selection_listener(self.current_dim, ids)
|
|
|
|
def lasso_select_callback(self, verts, key=None):
|
|
xys = scipy.c_[self.xaxis_data[:,scipy.newaxis], self.yaxis_data[:,scipy.newaxis]]
|
|
index = scipy.nonzero(verts.contains_points(xys))
|
|
ids = self.dataset_1.get_identifiers(self.current_dim, index)
|
|
ids = self.update_selection(ids, key)
|
|
self.selection_listener(self.current_dim, ids)
|
|
|
|
def set_current_selection(self, selection):
|
|
print "set_current_selection"
|
|
#remove old selection
|
|
if self._selection_line:
|
|
self.axes.lines.remove(self._selection_line)
|
|
index = self.get_index_from_selection(self.dataset_1, selection)
|
|
if len(index)==0:
|
|
# no selection
|
|
self.canvas.draw()
|
|
self._selection_line = None
|
|
return
|
|
|
|
xdata_new = self.xaxis_data.take(index) #take data
|
|
ydata_new = self.yaxis_data.take(index)
|
|
self._selection_line = Line2D(xdata_new, ydata_new
|
|
,marker='o', markersize=self.ms,
|
|
linewidth=0, markerfacecolor='r',
|
|
markeredgewidth=1.0)
|
|
self.axes.add_line(self._selection_line)
|
|
|
|
if self._use_blit:
|
|
if self._background is None:
|
|
self._background = self.canvas.copy_from_bbox(self.axes.bbox)
|
|
self.canvas.restore_region(self._background)
|
|
if self.selection_line:
|
|
self.axes.draw_artist(self._selection_line)
|
|
self.canvas.blit()
|
|
else:
|
|
self.canvas.draw()
|
|
|
|
|
|
class ScatterPlot(Plot):
|
|
"""The ScatterPlot is slower than scattermarker, but has size option."""
|
|
@plotlogger
|
|
def __init__(self, dataset_1, dataset_2, id_dim, sel_dim, id_1, id_2, c='b', s=30, sel_dim_2=None, name="Scatter plot", **kw):
|
|
"""Initializes a scatter plot.
|
|
"""
|
|
|
|
Plot.__init__(self, name)
|
|
self.dataset_1 = dataset_1
|
|
self.s = s
|
|
self.c = c
|
|
self.kw = kw
|
|
self.current_dim = id_dim
|
|
self._map_ids = dataset_1.get_identifiers(id_dim, sorted=True)
|
|
|
|
x_index = dataset_1[sel_dim][id_1]
|
|
if sel_dim_2:
|
|
y_index = dataset_2[sel_dim_2][id_2]
|
|
else:
|
|
y_index = dataset_2[sel_dim][id_2]
|
|
self.xaxis_data = dataset_1.asarray()[:, x_index]
|
|
self.yaxis_data = dataset_2.asarray()[:, y_index]
|
|
|
|
# init draw
|
|
self.init_draw()
|
|
|
|
# fixme: taking out blit support for now
|
|
#
|
|
# # signals to enable correct use of blit
|
|
# self.connect('zoom-changed', self.onzoom)
|
|
# self.connect('pan-changed', self.onpan)
|
|
# self.need_redraw = False
|
|
# self.canvas.mpl_connect('resize_event', self.onresize)
|
|
|
|
def add_axes_spin_buttons(self, max=None, absi=0, ordi=1):
|
|
self._absi = absi
|
|
self._ordi = ordi
|
|
if max == None:
|
|
max = 5
|
|
sb_a = gtk.SpinButton(climb_rate=1)
|
|
sb_a.set_range(1, max)
|
|
sb_a.set_value(absi+1)
|
|
sb_a.set_increments(1, 5)
|
|
sb_a.connect('value_changed', self.set_absicca)
|
|
sb_o = gtk.SpinButton(climb_rate=1)
|
|
sb_o.set_range(1, max)
|
|
sb_o.set_value(ordi+1)
|
|
sb_o.set_increments(1, 5)
|
|
sb_o.connect('value_changed', self.set_ordinate)
|
|
hbox = gtk.HBox()
|
|
gtk_label_a = gtk.Label(" A:")
|
|
gtk_label_o = gtk.Label(" O:")
|
|
toolitem = gtk.ToolItem()
|
|
toolitem.set_expand(False)
|
|
toolitem.set_border_width(2)
|
|
toolitem.add(hbox)
|
|
hbox.pack_start(gtk_label_a)
|
|
hbox.pack_start(sb_a)
|
|
hbox.pack_start(gtk_label_o)
|
|
hbox.pack_start(sb_o)
|
|
self._toolbar.insert(toolitem, -1)
|
|
toolitem.set_tooltip(self._toolbar.tooltips, "Set axis")
|
|
self._toolbar.show_all() #do i need this?
|
|
|
|
def set_absicca(self, sb):
|
|
self._absi = sb.get_value_as_int() - 1
|
|
xy = self.dataset_1.asarray()[:,[self._absi, self._ordi]]
|
|
self.xaxis_data = xy[:,0]
|
|
self.yaxis_data = xy[:,1]
|
|
self.sc._offsets = xy
|
|
self.selection_collection._offsets = xy
|
|
self._reset_limits(0)
|
|
self.canvas.draw_idle()
|
|
|
|
def set_ordinate(self, sb):
|
|
self._ordi = sb.get_value_as_int() - 1
|
|
xy = self.dataset_1.asarray()[:,[self._absi, self._ordi]]
|
|
self.xaxis_data = xy[:,0]
|
|
self.yaxis_data = xy[:,1]
|
|
self.sc._offsets = xy
|
|
self.selection_collection._offsets = xy
|
|
self._reset_limits(1)
|
|
self.canvas.draw_idle()
|
|
|
|
def onzoom(self, widget, mode):
|
|
#logger.log('notice', 'Zoom in widget: %s' %widget)
|
|
self.clean_redraw()
|
|
|
|
def onpan(self, widget, mode):
|
|
#logger.log('notice', 'Pan in widget: %s' %widget)
|
|
self.clean_redraw()
|
|
|
|
def onresize(self, widget):
|
|
#logger.log('notice', 'resize event')
|
|
self.clean_redraw()
|
|
|
|
def clean_redraw(self):
|
|
if self._use_blit == True:
|
|
#logger.log('notice', 'blit -> clean redraw ')
|
|
self.set_current_selection(None)
|
|
self._background = self.canvas.copy_from_bbox(self.axes.bbox)
|
|
self.set_current_selection(self._current_selection)
|
|
else:
|
|
self._background = None
|
|
|
|
def init_draw(self):
|
|
lw = scipy.zeros(self.xaxis_data.shape)
|
|
self.sc = self.axes.scatter(self.xaxis_data, self.yaxis_data, edgecolor='r',
|
|
s=self.s, c=self.c, linewidth=lw,
|
|
zorder=3, **self.kw)
|
|
self._mappable = self.sc
|
|
self.selection_collection = self.axes.scatter(self.xaxis_data,
|
|
self.yaxis_data,
|
|
alpha=0,
|
|
c='w',s=self.s,
|
|
edgecolor='r',
|
|
linewidth=4,
|
|
zorder=4)
|
|
self.axes.axhline(0, color='k', lw=1., zorder=1)
|
|
self.axes.axvline(0, color='k', lw=1., zorder=1)
|
|
# axhline and axvline mistakes the max/min using *all* data
|
|
self._reset_limits(0)
|
|
self._reset_limits(1)
|
|
|
|
self._background = self.canvas.copy_from_bbox(self.axes.bbox)
|
|
|
|
def _reset_limits(self, axis):
|
|
""" Resets vievlimits to datarange."""
|
|
if axis == 0:
|
|
cax = self.xaxis_data
|
|
set_lim = self.axes.set_xlim
|
|
elif axis == 1 or axis == -1:
|
|
set_lim = self.axes.set_ylim
|
|
cax = self.yaxis_data
|
|
else:
|
|
raise ValueError("Axis needs to be 0 or 1")
|
|
pad = abs(cax.min()- cax.max())*0.05
|
|
new_lims = (cax.min() - pad, cax.max() + pad)
|
|
set_lim(new_lims, emit=True)
|
|
|
|
def is_mappable_with(self, obj):
|
|
"""Returns True if dataset/selection is mappable with this plot.
|
|
"""
|
|
if isinstance(obj, laydi.dataset.Dataset):
|
|
if self.current_dim in obj.get_dim_name():
|
|
return True
|
|
|
|
elif isinstance(obj, laydi.dataset.Selection):
|
|
if self.current_dim in obj.get_dim_name():
|
|
return True
|
|
|
|
else:
|
|
return False
|
|
|
|
def _update_color_from_dataset(self, data):
|
|
"""Updates the facecolors from a dataset.
|
|
"""
|
|
array = data.asarray()
|
|
#only support for 2d-arrays:
|
|
try:
|
|
m, n = array.shape
|
|
except:
|
|
raise ValueError, "No support for more than 2 dimensions."
|
|
|
|
# is dataset a vector or matrix?
|
|
if not n==1:
|
|
# we have a category dataset
|
|
if isinstance(data, laydi.dataset.CategoryDataset):
|
|
vec = scipy.dot(array, scipy.diag(scipy.arange(n))).sum(1)
|
|
else:
|
|
vec = array.sum(1)
|
|
else:
|
|
vec = array.ravel()
|
|
|
|
# 1.) Set all ids present in plot but not in dataset to gray
|
|
# 2.) Set all ids which are inf/-inf/nan in dataset to gray
|
|
|
|
# ids of scatterplot
|
|
identifiers = self.dataset_1.get_identifiers(self.current_dim, sorted=True)
|
|
# corresponding indices of input data
|
|
indices = data.get_indices(self.current_dim, identifiers)
|
|
# ids of scatterplot that is present in input data
|
|
existing_ids = data.existing_identifiers(self.current_dim, identifiers)
|
|
|
|
# use only values present in scatterplot
|
|
v = vec[indices]
|
|
vec_min = min(vec[vec > -scipy.inf])
|
|
vec_max = max(vec[vec < scipy.inf])
|
|
ptp = abs(vec_max - vec_min)
|
|
# set all infs equal to max value + delta
|
|
delta = 2.
|
|
v[v==scipy.inf] = vec_max + delta
|
|
v[v==-scipy.inf] = vec_max + delta
|
|
v[v==scipy.nan] = vec_max + delta
|
|
|
|
# get the indices of scatterplot ids present in input data
|
|
indices = self.dataset_1.get_indices(self.current_dim, existing_ids)
|
|
map_vec = vec_max*scipy.ones(len(identifiers))
|
|
# setting all present values to v
|
|
map_vec[indices] = v
|
|
|
|
# update facecolors
|
|
self.sc.set_array(map_vec)
|
|
self.sc.set_clim(vec_min, vec_max)
|
|
self.sc.update_scalarmappable() #sets facecolors from array
|
|
# adjust max values so they get one (1) bin in lut
|
|
|
|
if hasattr(self.sc.cmap, "_lut"):
|
|
#! fixme this is just a hack, not even tested
|
|
lut_l = 1.*self.sc.cmap._lut.shape[0]
|
|
map_vec_range = map_vec.ptp()
|
|
delta_lut = map_vec_range/lut_l
|
|
# just add a small value to distinct nans/infs etc.
|
|
map_vec[map_vec==(vec_max + delta)] = vec_max + delta_lut/2.
|
|
self.sc.cmap._lut[-1,:] = [.5, .5, .5, 1]
|
|
else:
|
|
print "No lut present"
|
|
self.canvas.draw()
|
|
|
|
def rectangle_select_callback(self, x1, y1, x2, y2, key):
|
|
ydata = self.yaxis_data
|
|
xdata = self.xaxis_data
|
|
|
|
# find indices of selected area
|
|
if x1>x2:
|
|
x1, x2 = x2, x1
|
|
if y1>y2:
|
|
y1, y2 = y2, y1
|
|
assert x1<=x2
|
|
assert y1<=y2
|
|
|
|
index = scipy.nonzero((xdata>x1) & (xdata<x2) & (ydata>y1) & (ydata<y2))[0]
|
|
ids = self.dataset_1.get_identifiers(self.current_dim, index)
|
|
ids = self.update_selection(ids, key)
|
|
self.selection_listener(self.current_dim, ids)
|
|
|
|
def lasso_select_callback(self, verts, key=None):
|
|
xys = scipy.c_[self.xaxis_data[:,scipy.newaxis], self.yaxis_data[:,scipy.newaxis]]
|
|
poly = matplotlib.path.Path(verts, closed=True)
|
|
index = scipy.nonzero(poly.contains_points(xys))[0]
|
|
ids = self.dataset_1.get_identifiers(self.current_dim, index)
|
|
ids = self.update_selection(ids, key)
|
|
self.selection_listener(self.current_dim, ids)
|
|
|
|
def set_current_selection(self, selection):
|
|
linewidth = scipy.zeros(self.xaxis_data.shape, 'f')
|
|
index = self.get_index_from_selection(self.dataset_1, selection)
|
|
if len(index) > 0:
|
|
linewidth[index] = 1.5
|
|
self.sc.set_linewidth(linewidth)
|
|
|
|
if self._use_blit and len(index)>0 :
|
|
if self._background is None:
|
|
self._background = self.canvas.copy_from_bbox(self.axes.bbox)
|
|
self.canvas.restore_region(self._background)
|
|
self.axes.draw_artist(self.selection_collection)
|
|
self.canvas.blit()
|
|
else:
|
|
self.canvas.draw()
|
|
|
|
|
|
class ImagePlot(Plot):
|
|
@plotlogger
|
|
def __init__(self, dataset, **kw):
|
|
Plot.__init__(self, kw.get('name', 'Image Plot'))
|
|
self.dataset = dataset
|
|
|
|
# Initial draw
|
|
self.axes.grid(False)
|
|
self.axes.imshow(dataset.asarray(), interpolation='nearest')
|
|
self.axes.axis('tight')
|
|
self._mappable = self.axes.images[0]
|
|
|
|
# Disable selection modes
|
|
self._toolbar.freeze_button.set_sensitive(False)
|
|
self._toolbar.set_mode_sensitive('select', False)
|
|
self._toolbar.set_mode_sensitive('lassoselect', False)
|
|
|
|
|
|
class HistogramPlot(Plot):
|
|
""" Histogram plot.
|
|
If dataset is 1-dim the current_dim is set and selections may
|
|
be performed. For dataset> 1.dim the histogram is over all values
|
|
and selections are not defined,"""
|
|
@plotlogger
|
|
def __init__(self, dataset, **kw):
|
|
Plot.__init__(self, kw['name'])
|
|
self.dataset = dataset
|
|
self._data = dataset.asarray()
|
|
|
|
# If dataset is 1-dim we may do selections
|
|
if dataset.shape[0]==1:
|
|
self.current_dim = dataset.get_dim_name(1)
|
|
if dataset.shape[1]==1:
|
|
self.current_dim = dataset.get_dim_name(0)
|
|
|
|
# Set default paramteters
|
|
if not kw.has_key('bins'):
|
|
kw['bins'] = self._get_binsize()
|
|
|
|
# Initial draw
|
|
self.axes.grid(False)
|
|
bins = min(self._data.size, kw['bins'])
|
|
count, lims, self.patches = self.axes.hist(self._data, bins=bins)
|
|
# Add identifiers to the individual patches
|
|
if self.current_dim != None:
|
|
for i, patch in enumerate(self.patches):
|
|
if i==len(self.patches)-1:
|
|
end_lim = self._data.max() + 1
|
|
else:
|
|
end_lim = lims[i+1]
|
|
bool_ind = scipy.bitwise_and(self._data>=lims[i],
|
|
self._data<=end_lim)
|
|
patch.index = scipy.where(bool_ind)[0]
|
|
|
|
if self.current_dim==None:
|
|
# Disable selection modes
|
|
logger.log('notice', 'Disabled selections in Histogram Plot')
|
|
self._toolbar.freeze_button.set_sensitive(False)
|
|
self._toolbar.set_mode_sensitive('select', False)
|
|
self._toolbar.set_mode_sensitive('lassoselect', False)
|
|
|
|
def rectangle_select_callback(self, x1, y1, x2, y2, key):
|
|
if self.current_dim == None: return
|
|
# make (x1, y1) the lower left corner
|
|
if x1>x2:
|
|
x1, x2 = x2, x1
|
|
if y1>y2:
|
|
y1, y2 = y2, y1
|
|
|
|
self.active_patches = []
|
|
for patch in self.patches:
|
|
xmin = patch.xy[0]
|
|
xmax = xmin + patch.get_width()
|
|
ymin, ymax = -0.001, patch.get_height()
|
|
if xmax>x1 and xmin<x2 and (ymax> y2 or ymax>y1):
|
|
self.active_patches.append(patch)
|
|
if not self.active_patches: return
|
|
|
|
ids = set()
|
|
for patch in self.active_patches:
|
|
ids.update(self.dataset.get_identifiers(self.current_dim,
|
|
patch.index))
|
|
ids = self.update_selection(ids, key)
|
|
self.selection_listener(self.current_dim, ids)
|
|
|
|
def lasso_select_callback(self, verts, key):
|
|
if self.current_dim == None: return
|
|
self.active_patches = []
|
|
for patch in self.patches:
|
|
if scipy.any(verts.contains_points(patch.get_verts())):
|
|
self.active_patches.append(patch)
|
|
if not self.active_patches: return
|
|
ids = set()
|
|
for patch in self.active_patches:
|
|
ids.update(self.dataset.get_identifiers(self.current_dim,
|
|
patch.index))
|
|
ids = self.update_selection(ids, key)
|
|
self.selection_listener(self.current_dim, ids)
|
|
|
|
def set_current_selection(self, selection):
|
|
index = self.get_index_from_selection(self.dataset, selection)
|
|
for patch in self.patches:
|
|
patch.set_facecolor('b')
|
|
for patch in self.patches:
|
|
bin_selected = scipy.intersect1d(patch.index, index).size
|
|
if bin_selected>0:
|
|
bin_total = len(patch.index)
|
|
# fixme: engineering color
|
|
prop = -scipy.log(1.0*bin_selected/bin_total)
|
|
b = min(prop, 1)
|
|
r = max(.5, 1-b)
|
|
g = 0
|
|
patch.set_facecolor((r,g,b,1))
|
|
self.canvas.draw()
|
|
|
|
def _get_binsize(self, min_bins=2, max_bins=100):
|
|
""" Automatic bin selection, as described by Shimazaki."""
|
|
bin_vec = scipy.arange(min_bins, max_bins, 1)
|
|
D = self._data.ptp()/bin_vec
|
|
cost = scipy.empty((bin_vec.shape[0],), 'f')
|
|
for i, bins in enumerate(bin_vec):
|
|
count, lims = scipy.histogram(self._data, bins)
|
|
cost[i] = (2*count.mean() - count.var())/(D[i]**2)
|
|
best_bin_size = bin_vec[scipy.argmin(cost)]
|
|
return best_bin_size
|
|
|
|
class BoxPlot(Plot):
|
|
"""Box plot.
|
|
"""
|
|
def __init__(self, ds, name="Box Plot"):
|
|
Plot.__init__(self, name)
|
|
|
|
self.axes = self.fig.add_subplot(111)
|
|
|
|
m = ds.asarray()
|
|
|
|
box_plot_lines = self.axes.boxplot(m)
|
|
self.axes.grid(False)
|
|
|
|
dim = ds.get_dim_name(1)
|
|
labels = ds.get_identifiers(dim, sorted=True)
|
|
self.axes.set_xticklabels(labels)
|
|
|
|
self.add(self.canvas)
|
|
self.canvas.show()
|
|
|
|
def set_current_selection(self, selection):
|
|
pass
|
|
|
|
|
|
class BarPlot(Plot):
|
|
"""Bar plot.
|
|
|
|
Ordinary bar plot for (column) vectors.
|
|
For matrices there is one color for each row.
|
|
"""
|
|
@plotlogger
|
|
def __init__(self, dataset, **kw):
|
|
Plot.__init__(self, kw.get('name', 'Bar Plot'))
|
|
self.dataset = dataset
|
|
|
|
# Initial draw
|
|
self.axes.grid(False)
|
|
n, m = dataset.shape
|
|
if m>1:
|
|
clrs = matplotlib.cm.ScalarMappable().to_rgba(range(n))
|
|
for i, row in enumerate(dataset.asarray()):
|
|
left = scipy.arange(i+1, m*n+1, n)
|
|
height = row
|
|
color = clrs[i]
|
|
c = (color[0], color[1], color[2])
|
|
self.axes.bar(left, height,color=c)
|
|
else:
|
|
height = dataset.asarray().ravel()
|
|
left = scipy.arange(1, n+1, 1)
|
|
self.axes.bar(left, height)
|
|
|
|
# Disable selection modes
|
|
self._toolbar.freeze_button.set_sensitive(False)
|
|
self._toolbar.set_mode_sensitive('select', False)
|
|
self._toolbar.set_mode_sensitive('lassoselect', False)
|
|
|
|
|
|
class NetworkPlot(Plot):
|
|
@plotlogger
|
|
def __init__(self, dataset, pos=None, nodecolor='b', nodesize=40,
|
|
prog='neato', with_labels=False, name='Network Plot'):
|
|
|
|
Plot.__init__(self, name)
|
|
self.dataset = dataset
|
|
self.graph = dataset.asnetworkx()
|
|
self._prog = prog
|
|
self._pos = pos
|
|
self._nodesize = nodesize
|
|
self._nodecolor = nodecolor
|
|
self._with_labels = with_labels
|
|
|
|
self.current_dim = self.dataset.get_dim_name(0)
|
|
|
|
if not self._pos:
|
|
self._pos = networkx.graphviz_layout(self.graph, self._prog)
|
|
self._xy = scipy.asarray([self._pos[node] for node in self.dataset.get_identifiers(self.current_dim, sorted=True)])
|
|
self.xaxis_data = self._xy[:,0]
|
|
self.yaxis_data = self._xy[:,1]
|
|
|
|
# Initial draw
|
|
self.default_props = {'nodesize' : 50,
|
|
'nodecolor' : 'blue',
|
|
'edge_color' : 'gray',
|
|
'edge_color_selected' : 'red'}
|
|
self.node_collection = None
|
|
self.edge_collection = None
|
|
self.node_labels = None
|
|
lw = scipy.zeros(self.xaxis_data.shape)
|
|
self.node_collection = self.axes.scatter(self.xaxis_data, self.yaxis_data,
|
|
s=self._nodesize,
|
|
c=self._nodecolor,
|
|
linewidth=lw,
|
|
zorder=3)
|
|
self._mappable = self.node_collection
|
|
# selected nodes is a transparent graph that adjust node-edge visibility
|
|
# according to the current selection needed to get get the selected
|
|
# nodes 'on top' as zorder may not be defined individually
|
|
self.selected_nodes = self.axes.scatter(self.xaxis_data,
|
|
self.yaxis_data,
|
|
s=self._nodesize,
|
|
c=self._nodecolor,
|
|
linewidth=lw,
|
|
zorder=4,
|
|
alpha=0)
|
|
|
|
edge_color = self.default_props['edge_color']
|
|
self.edge_collection = networkx.draw_networkx_edges(self.graph,
|
|
self._pos,
|
|
ax=self.axes,
|
|
edge_color=edge_color)
|
|
# edge color rgba-arrays
|
|
self._edge_color_rgba = matlib.repmat(ColorConverter().to_rgba(edge_color),
|
|
self.graph.number_of_edges(),1)
|
|
self._edge_color_selected = ColorConverter().to_rgba(self.default_props['edge_color_selected'])
|
|
if self._with_labels:
|
|
self.node_labels = networkx.draw_networkx_labels(self.graph,
|
|
self._pos,
|
|
ax=self.axes)
|
|
|
|
# remove axes, frame and grid
|
|
self.axes.set_xticks([])
|
|
self.axes.set_yticks([])
|
|
self.axes.grid(False)
|
|
self.axes.set_frame_on(False)
|
|
self.fig.subplots_adjust(left=0, right=1, bottom=0, top=1)
|
|
|
|
def rectangle_select_callback(self, x1, y1, x2, y2, key):
|
|
ydata = self.yaxis_data
|
|
xdata = self.xaxis_data
|
|
|
|
# find indices of selected area
|
|
if x1>x2:
|
|
x1, x2 = x2, x1
|
|
if y1>y2:
|
|
y1, y2 = y2, y1
|
|
assert x1<=x2
|
|
assert y1<=y2
|
|
|
|
index = scipy.nonzero((xdata>x1) & (xdata<x2) & (ydata>y1) & (ydata<y2))[0]
|
|
ids = self.dataset.get_identifiers(self.current_dim, index)
|
|
print "ids in rsc: %s" %str(ids)
|
|
ids = self.update_selection(ids, key)
|
|
self.selection_listener(self.current_dim, ids)
|
|
|
|
def lasso_select_callback(self, verts, key=None):
|
|
xys = scipy.c_[self.xaxis_data[:,scipy.newaxis], self.yaxis_data[:,scipy.newaxis]]
|
|
index = scipy.nonzero(verts.contains_ponts(xys))
|
|
ids = self.dataset.get_identifiers(self.current_dim, index)
|
|
ids = self.update_selection(ids, key)
|
|
self.selection_listener(self.current_dim, ids)
|
|
|
|
def set_current_selection(self, selection):
|
|
linewidth = scipy.zeros(self.xaxis_data.shape)
|
|
edge_color_rgba = self._edge_color_rgba.copy()
|
|
index = self.get_index_from_selection(self.dataset, selection)
|
|
if len(index) > 0:
|
|
linewidth[index] = 2
|
|
idents = selection[self.current_dim]
|
|
edge_index = [i for i,edge in enumerate(self.graph.edges()) if (edge[0] in idents and edge[1] in idents)]
|
|
if len(edge_index)>0:
|
|
for i in edge_index:
|
|
edge_color_rgba[i,:] = self._edge_color_selected
|
|
self._A = None
|
|
|
|
self.edge_collection._colors = edge_color_rgba
|
|
self.selected_nodes.set_linewidth(linewidth)
|
|
self.canvas.draw()
|
|
|
|
|
|
class VennPlot(Plot):
|
|
@plotlogger
|
|
def __init__(self, name="Venn diagram"):
|
|
Plot.__init__(self, name)
|
|
|
|
# init draw
|
|
self._init_bck()
|
|
for c in self._venn_patches:
|
|
self.axes.add_patch(c)
|
|
for mrk in self._markers:
|
|
self.axes.add_patch(mrk)
|
|
self.axes.set_xlim([-3, 3])
|
|
self.axes.set_ylim([-2.5, 3.5])
|
|
self._last_active = set()
|
|
self.axes.set_xticks([])
|
|
self.axes.set_yticks([])
|
|
self.axes.axis('equal')
|
|
self.axes.grid(False)
|
|
self.axes.set_frame_on(False)
|
|
self.fig.subplots_adjust(left=0, right=1, bottom=0, top=1)
|
|
|
|
def _init_bck(self):
|
|
res = 50
|
|
a = .5
|
|
r = 1.5
|
|
mr = .2
|
|
self.c1 = c1 = Circle((-1,0), radius=r, alpha=a, facecolor='b')
|
|
self.c2 = c2 = Circle((1,0), radius=r, alpha=a, facecolor='r')
|
|
self.c3 = c3 = Circle((0, scipy.sqrt(3)), radius=r, alpha=a, facecolor='g')
|
|
|
|
self.c1marker = Circle((-1.25, -.25), radius=mr, facecolor='y', alpha=0)
|
|
self.c2marker = Circle((1.25, -.25), radius=mr, facecolor='y', alpha=0)
|
|
self.c3marker = Circle((0, scipy.sqrt(3)+.25), radius=mr, facecolor='y', alpha=0)
|
|
self.c1c2marker = Circle((0, -.15), radius=mr, facecolor='y', alpha=0)
|
|
|
|
self.c1c3marker = Circle((-scipy.sqrt(2)/2, 1), radius=mr, facecolor='y', alpha=0)
|
|
self.c2c3marker = Circle((scipy.sqrt(2)/2, 1), radius=mr, facecolor='y', alpha=0)
|
|
self.c1c2c3marker = Circle((0, .6), radius=mr, facecolor='y', alpha=0)
|
|
|
|
c1.elements = set(['a', 'b', 'c', 'f'])
|
|
c2.elements = set(['a', 'c', 'd', 'e'])
|
|
c3.elements = set(['a', 'e', 'f', 'g'])
|
|
self.active_elements = set()
|
|
self.all_elements = c1.elements.union(c2.elements).union(c3.elements)
|
|
|
|
c1.active = False
|
|
c2.active = False
|
|
c3.active = False
|
|
|
|
c1.name = 'Blue'
|
|
c2.name = 'Red'
|
|
c3.name = 'Green'
|
|
|
|
self._venn_patches = [c1, c2, c3]
|
|
self._markers = [self.c1marker, self.c2marker, self.c3marker,
|
|
self.c1c2marker, self.c1c3marker,
|
|
self.c2c3marker, self.c1c2c3marker]
|
|
|
|
self._tot_label = 'Tot: ' + str(len(self.all_elements))
|
|
self._sel_label = 'Sel: ' + str(len(self.active_elements))
|
|
self._legend = self.axes.legend((self._tot_label, self._sel_label),
|
|
loc='upper right')
|
|
|
|
def set_selection(self, selection, patch=None):
|
|
if patch:
|
|
patch.selection = selection
|
|
else:
|
|
selection_set = False
|
|
for patch in self._venn_patches:
|
|
if len(patch.elements)==0:
|
|
patch.elements = selection
|
|
selection_set = True
|
|
if not selection_set:
|
|
self.venn_patches[0].elements = selection
|
|
|
|
def lasso_select_callback(self, verts, key=None):
|
|
if verts==None:
|
|
verts = (self._event.xdata, self._event.ydata)
|
|
if key!='shift':
|
|
for m in self._markers:
|
|
m.set_alpha(0)
|
|
|
|
self._patches_within_verts(verts, key)
|
|
active = [i.active for i in self._venn_patches]
|
|
if active==[True, False, False]:
|
|
self.c1marker.set_alpha(1)
|
|
self.active_elements = self.c1.elements.difference(self.c2.elements.union(self.c3.elements))
|
|
elif active== [False, True, False]:
|
|
self.c2marker.set_alpha(1)
|
|
self.active_elements = self.c2.elements.difference(self.c1.elements.union(self.c3.elements))
|
|
elif active== [False, False, True]:
|
|
self.c3marker.set_alpha(1)
|
|
self.active_elements = self.c3.elements.difference(self.c2.elements.union(self.c1.elements))
|
|
elif active==[True, True, False]:
|
|
self.c1c2marker.set_alpha(1)
|
|
self.active_elements = self.c1.elements.intersection(self.c2.elements)
|
|
elif active==[True, False, True]:
|
|
self.c1c3marker.set_alpha(1)
|
|
self.active_elements = self.c1.elements.intersection(self.c3.elements)
|
|
elif active==[False, True, True]:
|
|
self.c2c3marker.set_alpha(1)
|
|
self.active_elements = self.c2.elements.intersection(self.c3.elements)
|
|
elif active==[True, True, True]:
|
|
self.c1c2c3marker.set_alpha(1)
|
|
self.active_elements = self.c1.elements.intersection(self.c3.elements).intersection(self.c2.elements)
|
|
|
|
if key=='shift':
|
|
self.active_elements = self.active_elements.union(self._last_active)
|
|
self._last_active = self.active_elements.copy()
|
|
self._sel_label = 'Sel: ' + str(len(self.active_elements))
|
|
self._legend.texts[1].set_text(self._sel_label)
|
|
self.axes.figure.canvas.draw()
|
|
|
|
def rectangle_select_callback(self, x1, y1, x2, y2, key):
|
|
verts = [(x1, y1), (x2, y2)]
|
|
if key!='shift':
|
|
for m in self._markers:
|
|
m.set_alpha(0)
|
|
|
|
self._patches_within_verts(verts, key)
|
|
active = [i.active for i in self._venn_patches]
|
|
if active==[True, False, False]:
|
|
self.c1marker.set_alpha(1)
|
|
self.active_elements = self.c1.elements.difference(self.c2.elements.union(self.c3.elements))
|
|
elif active== [False, True, False]:
|
|
self.c2marker.set_alpha(1)
|
|
self.active_elements = self.c2.elements.difference(self.c1.elements.union(self.c3.elements))
|
|
elif active== [False, False, True]:
|
|
self.c3marker.set_alpha(1)
|
|
self.active_elements = self.c3.elements.difference(self.c2.elements.union(self.c1.elements))
|
|
elif active==[True, True, False]:
|
|
self.c1c2marker.set_alpha(1)
|
|
self.active_elements = self.c1.elements.intersection(self.c2.elements)
|
|
elif active==[True, False, True]:
|
|
self.c1c3marker.set_alpha(1)
|
|
self.active_elements = self.c1.elements.intersection(self.c3.elements)
|
|
elif active==[False, True, True]:
|
|
self.c2c3marker.set_alpha(1)
|
|
self.active_elements = self.c2.elements.intersection(self.c3.elements)
|
|
elif active==[True, True, True]:
|
|
self.c1c2c3marker.set_alpha(1)
|
|
self.active_elements = self.c1.elements.intersection(self.c3.elements).intersection(self.c2.elements)
|
|
|
|
if key=='shift':
|
|
self.active_elements = self.active_elements.union(self._last_active)
|
|
self._last_active = self.active_elements.copy()
|
|
self._sel_label = 'Sel: ' + str(len(self.active_elements))
|
|
self._legend.texts[1].set_text(self._sel_label)
|
|
self.axes.figure.canvas.draw()
|
|
|
|
def _patches_within_verts(self, verts, key):
|
|
xy = scipy.array(verts).mean(0)
|
|
for venn_patch in self._venn_patches:
|
|
venn_patch.active = False
|
|
if self._distance(venn_patch.center,xy)<venn_patch.radius:
|
|
venn_patch.active = True
|
|
|
|
def _distance(self, (x1,y1),(x2,y2)):
|
|
return scipy.sqrt( (x2-x1)**2 + (y2-y1)**2 )
|
|
|
|
|
|
class PlotThresholder:
|
|
"""Mixin class for plots that needs to filter nodes within a threshold
|
|
range.
|
|
"""
|
|
def __init__(self, text="x"):
|
|
"""Constructor.
|
|
|
|
@param text: Name of the variable the threshold is on.
|
|
"""
|
|
self._threshold_ds = None
|
|
self._add_spin_buttons(text)
|
|
self._sb_min.set_sensitive(False)
|
|
self._sb_max.set_sensitive(False)
|
|
|
|
def set_threshold_dataset(self, ds):
|
|
"""Sets the dataset to threshold on.
|
|
|
|
@param ds: A dataset where one dimension corresponds to the select dimension
|
|
in the plot, and any other dimensions have length 1
|
|
"""
|
|
self._threshold_ds = ds
|
|
self._sb_min.set_sensitive(True)
|
|
self._sb_max.set_sensitive(True)
|
|
|
|
def _add_spin_buttons(self, text):
|
|
"""Adds spin buttons to the toolbar for selecting minimum and maximum
|
|
threshold values on information content."""
|
|
sb_min = gtk.SpinButton(digits=2)
|
|
sb_min.set_range(0, 100)
|
|
sb_min.set_value(0)
|
|
sb_min.set_increments(.1, 1.)
|
|
sb_min.connect('value-changed', self._on_value_changed)
|
|
self._sb_min = sb_min
|
|
|
|
sb_max = gtk.SpinButton(digits=2)
|
|
sb_max.set_range(0, 100)
|
|
sb_max.set_value(1)
|
|
sb_max.set_increments(.1, 1.)
|
|
sb_max.connect('value-changed', self._on_value_changed)
|
|
self._sb_max = sb_max
|
|
|
|
label = gtk.Label(" < %s < " % text)
|
|
hbox = gtk.HBox()
|
|
hbox.pack_start(sb_min)
|
|
hbox.pack_start(label)
|
|
hbox.pack_start(sb_max)
|
|
ti = gtk.ToolItem()
|
|
ti.set_expand(False)
|
|
ti.add(hbox)
|
|
sb_min.show()
|
|
sb_max.show()
|
|
label.show()
|
|
hbox.show()
|
|
ti.show()
|
|
self._toolbar.insert(ti, -1)
|
|
ti.set_tooltip(self._toolbar.tooltips, "Set threshold")
|
|
|
|
def set_threshold(self, min, max):
|
|
"""Sets min and max to the given values.
|
|
Updates the plot accordingly to show only values that have a
|
|
value within the boundaries. Other values are
|
|
also excluded from being selected from the plot.
|
|
@param ic_min Do not show nodes with IC below this value.
|
|
@param ic_max Do not show nodes with IC above this value.
|
|
"""
|
|
ds = self._threshold_ds
|
|
if ds == None:
|
|
return
|
|
|
|
icnodes = ds.existing_identifiers('go-terms', self._map_ids)
|
|
icindices = ds.get_indices('go-terms', icnodes)
|
|
a = ds.asarray()[icindices].sum(1)
|
|
|
|
good = set(scipy.array(icnodes)[(a>=min) & (a<=max)])
|
|
|
|
sizes = scipy.zeros(len(self._map_ids))
|
|
visible = set()
|
|
for i, n in enumerate(self._map_ids):
|
|
if n in good:
|
|
sizes[i] = 50
|
|
visible.add(n)
|
|
else:
|
|
sizes[i] = 0
|
|
self.visible = visible
|
|
|
|
self._mappable._sizes = sizes
|
|
self.canvas.draw()
|
|
|
|
def get_nodes_within_bounds(self):
|
|
"""Get a list of all nodes within the bounds of the selection in the
|
|
seleted dataset.
|
|
"""
|
|
pass
|
|
|
|
def filter_nodes(self, nodes):
|
|
"""Filter a list of nodes and return only those that are within the
|
|
threshold boundaries."""
|
|
pass
|
|
|
|
def _on_value_changed(self, sb):
|
|
"""Callback on spin button value changes."""
|
|
min = self._sb_min.get_value()
|
|
max = self._sb_max.get_value()
|
|
self.set_threshold(min, max)
|
|
|
|
|
|
#class PlotContextMenu(gtk.Menu):
|
|
# def __init__(self, plot):
|
|
# gtk.Menu(self)
|
|
# self._plot = plot
|
|
#
|
|
# # Populate main menu
|
|
## self.color_item = gtk.MenuItem('Color')
|
|
# self.append(self.color_item)
|
|
# self.color_item.set_submenu(self.build_dataset_menu()):
|
|
#
|
|
# def build_dataset_menu(self):
|
|
# return gtk.Menu()
|
|
|
|
|
|
# Create zoom-changed signal
|
|
gobject.signal_new('zoom-changed', Plot, gobject.SIGNAL_RUN_LAST, None,
|
|
(gobject.TYPE_PYOBJECT,))
|
|
|
|
# Create pan/zoom-changed signal
|
|
gobject.signal_new('pan-changed', Plot, gobject.SIGNAL_RUN_LAST, None,
|
|
(gobject.TYPE_PYOBJECT,))
|
|
|
|
# Create plot-resize-changed signal
|
|
gobject.signal_new('plot-resize-changed', Plot, gobject.SIGNAL_RUN_LAST, None,
|
|
(gobject.TYPE_PYOBJECT,))
|