Python to generate nice looking SVG graph http://pygal.org/
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

257 lines
6.4 KiB

#!python
from SVG import Graph
from itertools import chain
__all__ = ( 'VerticalBar', 'HorizontalBar' )
class Bar( Graph ):
"A superclass for bar-style graphs. Do not instantiate directly."
# gap between bars
bar_gap = True
# how to stack adjacent dataset series
# overlap - overlap bars with transparent colors
# top - stack bars on top of one another
# side - stack bars side-by-side
stack = 'overlap'
scale_divisions = None
def __init__( self, fields, *args, **kargs ):
self.fields = fields
super( Bar, self ).__init__( *args, **kargs )
def data_max( self ):
return max( chain( *map( lambda set: set['data'], self.data ) ) )
# above is same as
# return max( map( lambda set: max( set['data'] ), self.data ) )
def data_min( self ):
if not getattr(self, 'min_scale_value') is None: return self.min_scale_value
min_value = min( chain( *map( lambda set: set['data'], self.data ) ) )
min_value = min( min_value, 0 )
return min_value
def get_css( self ):
return """\
/* default fill styles for multiple datasets (probably only use a single dataset on this graph though) */
.key1,.fill1{
fill: #ff0000;
fill-opacity: 0.5;
stroke: none;
stroke-width: 0.5px;
}
.key2,.fill2{
fill: #0000ff;
fill-opacity: 0.5;
stroke: none;
stroke-width: 1px;
}
.key3,.fill3{
fill: #00ff00;
fill-opacity: 0.5;
stroke: none;
stroke-width: 1px;
}
.key4,.fill4{
fill: #ffcc00;
fill-opacity: 0.5;
stroke: none;
stroke-width: 1px;
}
.key5,.fill5{
fill: #00ccff;
fill-opacity: 0.5;
stroke: none;
stroke-width: 1px;
}
.key6,.fill6{
fill: #ff00ff;
fill-opacity: 0.5;
stroke: none;
stroke-width: 1px;
}
.key7,.fill7{
fill: #00ffff;
fill-opacity: 0.5;
stroke: none;
stroke-width: 1px;
}
.key8,.fill8{
fill: #ffff00;
fill-opacity: 0.5;
stroke: none;
stroke-width: 1px;
}
.key9,.fill9{
fill: #cc6666;
fill-opacity: 0.5;
stroke: none;
stroke-width: 1px;
}
.key10,.fill10{
fill: #663399;
fill-opacity: 0.5;
stroke: none;
stroke-width: 1px;
}
.key11,.fill11{
fill: #339900;
fill-opacity: 0.5;
stroke: none;
stroke-width: 1px;
}
.key12,.fill12{
fill: #9966FF;
fill-opacity: 0.5;
stroke: none;
stroke-width: 1px;
}
"""
def float_range( start = 0, stop = None, step = 1 ):
"Much like the built-in function range, but accepts floats"
while start < stop:
yield float( start )
start += step
class VerticalBar( Bar ):
""" # === Create presentation quality SVG bar graphs easily
#
# = Synopsis
#
# require 'SVG/Graph/Bar'
#
# fields = %w(Jan Feb Mar);
# data_sales_02 = [12, 45, 21]
#
# graph = SVG::Graph::Bar.new(
# :height => 500,
# :width => 300,
# :fields => fields
# )
#
# graph.add_data(
# :data => data_sales_02,
# :title => 'Sales 2002'
# )
#
# print "Content-type: image/svg+xml\r\n\r\n"
# print graph.burn
#
# = Description
#
# This object aims to allow you to easily create high quality
# SVG[http://www.w3c.org/tr/svg bar graphs. You can either use the default
# style sheet or supply your own. Either way there are many options which
# can be configured to give you control over how the graph is generated -
# with or without a key, data elements at each point, title, subtitle etc.
#
# = Notes
#
# The default stylesheet handles upto 12 data sets, if you
# use more you must create your own stylesheet and add the
# additional settings for the extra data sets. You will know
# if you go over 12 data sets as they will have no style and
# be in black.
#
# = Examples
#
# * http://germane-software.com/repositories/public/SVG/test/test.rb
#
# = See also
#
# * SVG::Graph::Graph
# * SVG::Graph::BarHorizontal
# * SVG::Graph::Line
# * SVG::Graph::Pie
# * SVG::Graph::Plot
# * SVG::Graph::TimeSeries
"""
top_align = top_font = 1
def get_x_labels( self ):
return self.fields
# adapted from plot (very much like calling data_range('y'))
def data_range( self ):
min_value = self.data_min( )
max_value = self.data_max( )
range = max_value - min_value
top_pad = range / 20.0 or 10
scale_range = ( max_value + top_pad ) - min_value
scale_division = self.scale_divisions or ( scale_range / 10.0 )
if self.scale_integers:
scale_division = round(scale_division) or 1
return min_value, max_value, scale_division
# adapted from Plot
def get_data_values( self ):
min_value, max_value, scale_division = self.data_range()
result = tuple( float_range( min_value, max_value + scale_division, scale_division ) )
if self.scale_integers:
result = map(int, result)
return result
# adapted from Plot
def get_y_labels( self ):
return map( str, self.get_data_values() )
def x_label_offset( self, width ):
return width / 2.0
def draw_data( self ):
min_value = self.data_min()
unit_size = (float(self.graph_height) - self.font_size*2*self.top_font)
unit_size /= (max(self.get_data_values()) - min(self.get_data_values()) )
bar_gap = 0
if self.bar_gap:
bar_gap = 10
if self.get_field_width() < 10:
bar_gap = self.get_field_width() / 2
bar_width = self.get_field_width() - bar_gap
if self.stack == 'side':
bar_width /= len( self.data )
x_mod = ( self.graph_width - bar_gap )/2
if self.stack == 'side':
x_mod -= bar_width/2
bottom = self.graph_height
for field_count, field in enumerate( self.fields ):
for dataset_count, dataset in enumerate( self.data ):
# cases (assume 0 = +ve):
# value min length
# +ve +ve value - min
# +ve -ve value - 0
# -ve -ve value.abs - 0
value = dataset['data'][field_count]
left = self.get_field_width() * field_count
length = ( abs(value) - max( min_value, 0 ) ) * unit_size
# top is 0 if value is negative
top = bottom - (( max(value,0) - min_value ) * unit_size )
if self.stack == 'side':
left += bar_width * dataset_count
rect = self._create_element( 'rect', {
'x': str(left),
'y': str(top),
'width': str(bar_width),
'height': str(length),
'class': 'fill%s' % (dataset_count+1),
} )
self.graph.appendChild( rect )
self.make_datapoint_text( left + bar_width/2.0, top-6, value )