Welcome, guest | Sign In | My Account | Store | Cart
# Author: Anand Patil
# License: MIT License

import matplotlib.pyplot as pl
import numpy as np

def symmetric(sorted_streams, stream_bounds):
   
"""Symmetric baseline"""
    lb
, ub = np.min(stream_bounds[:,0,:],axis=0), np.max(stream_bounds[:,1,:],axis=0)
   
return .5*(lb+ub)
def pos_only(sorted_streams, stream_bounds):
   
"""Lumps will only be positive"""
    lb
, ub = np.min(stream_bounds[:,0,:],axis=0), np.max(stream_bounds[:,1,:],axis=0)
   
return lb
def zero(sorted_streams, stream_bounds):
   
"""Zero baseline"""
   
return np.zeros(stream_bounds.shape[2])
def min_weighted_wiggles(sorted_streams, stream_bounds):
   
"""Baseline recommended by Byron and Wattenberg"""
    lb
, ub = np.min(stream_bounds[:,0,:],axis=0), np.max(stream_bounds[:,1,:],axis=0)
    weight
= ub-lb
   
    sorted_streams
= np.abs(sorted_streams)
   
for i in xrange(len(sorted_streams)):
        sorted_streams
[i,:] *= (-1)**i
    cusum_f
= np.vstack((np.zeros(sorted_streams.shape[1]),
                        np
.cumsum(sorted_streams[:-1,:], axis=0)))
    f_prime
= np.diff(sorted_streams, axis=1)
    cusum_f_prime
= np.diff(cusum_f, axis=1)
    g_prime
= np.hstack(([0],-np.sum((f_prime*.5  + cusum_f_prime)*sorted_streams[:,1:],axis=0) / weight[1:]))
    g_prime
[np.where(weight==0)] = 0
    g
= np.cumsum(g_prime)
   
   
return g
   

def stacked_graph(streams, cmap=pl.cm.bone, color_seq='linear', baseline_fn=min_weighted_wiggles):
   
"""
    Produces stacked graphs using matplotlib.
   
    Reference: 'Stacked graphs- geometry & aesthetics' by Byron and Wattenberg
    http://www.leebyron.com/else/streamgraph/download.php?file=stackedgraphs_byron_wattenberg.pdf
   
    Parameters:
      - streams: A list of time-series of positive values. Each element must be of the same length.
      - cmap: A matplotlib color map. Defaults to 'bone'.
      - colo_seq: 'linear' or 'random'.
      - baseline_fn: Current options are symmetric, pos_only, zero and min_weighted_wiggles.
    """

   
# Sort by onset times
    onset_times
= [np.where(np.abs(stream)>0)[0][0] for stream in streams]
    order
= np.argsort(onset_times)
    streams
= np.asarray(streams)
    sorted_streams
= streams[order]
   
    t
= np.arange(streams.shape[1])
   
   
# Establish bounds
    stream_bounds
= [np.vstack((np.zeros(streams.shape[1]), sorted_streams[0])),
                    np
.vstack((-sorted_streams[1], (np.zeros(streams.shape[1]))))]
    side
= -1
   
for stream in sorted_streams[2:]:
        side
*= -1
       
if side==1:
            stream_bounds
.append(np.vstack((stream_bounds[-2][1], stream_bounds[-2][1]+stream)))
       
else:
            stream_bounds
.append(np.vstack((stream_bounds[-2][0]-stream, stream_bounds[-2][0])))
           
    stream_bounds
= np.array(stream_bounds)
   
   
# Compute baseline
    baseline
= baseline_fn(sorted_streams, stream_bounds)
   
   
# Choose colors
    t_poly
= np.hstack((t,t[::-1]))
   
if color_seq=='linear':
        colors
= np.linspace(0,1,streams.shape[1])
   
elif color_seq=='random':
        colors
= np.random.random(size=streams.shape[1])
   
else:
       
raise ValueError, 'Color sequence %s unrecognized'%color_seq
   
   
# Plot    
    pl
.axis('off')        
   
for i in xrange(len(stream_bounds)):
        bound
= stream_bounds[i]
        color
= cmap(colors[i])
        pl
.fill(t_poly, np.hstack((bound[0]-baseline,(bound[1]-baseline)[::-1])), facecolor=color, linewidth=0.,edgecolor='none')
       
       
# Demo
if __name__ == '__main__':
    pl
.clf()
    N_dsets
= 50
    T
= 100
    amp
= 1
    fade
= .15
    dsets
= []
   
for i in xrange(N_dsets):
        this_dset
= np.zeros(T)
        t_onset
= np.random.randint(.9*T)-T/3
       
if t_onset >= 0:  
            remaining_t
= np.arange(T-t_onset)    
       
else:
            remaining_t
= np.arange(T)-t_onset
        this_dset
[max(t_onset,0):]=np.exp(-.15*np.random.gamma(10,.1)*remaining_t)\
                           
* remaining_t * np.random.gamma(6,.2)# * np.cos(-fade*remaining_t*np.random.gamma(10,.1))**2
        dsets
.append(this_dset)
    stacked_graph
(dsets, baseline_fn = min_weighted_wiggles, color_seq='random')

History