def run2( reset = False,
base_net = 'kn',
comp_net = 'fn',
demand_bdtnp = False):
bd = nio.getBDTNP()
ktgs,ktfs = nio.getKNet()
tgs,tfs = nio.getNet()
sush = nio.getSush(on_fail = 'compute')
tfset = set(ktfs.keys())
tgset = set(ktgs.keys())
tg_int = set(tgs.keys()).intersection(ktgs.keys())
tf_int = set(tfs.keys()).intersection(ktfs.keys())
if demand_bdtnp:
tg_int = tg_int.intersection(bd.keys())
tf_int = tf_int.intersection(bd.keys())
if base_net =='kn':
b_edges = [(tf, tg, float(wt))
for tg, elt in ktgs.iteritems() if tg in tg_int
for tf, wt in zip(elt['tfs'], elt['weights']) if tf in tf_int]
if comp_net == 'fn':
c_edges = [(tf, tg, float(wt))
for tg, elt in tgs.iteritems() if tg in tg_int
for tf, wt in zip(elt['tfs'], elt['weights']) if tf in tf_int]
elif comp_net == 'sn':
#Sushmita network with signed edges
c_edges = [(tf, tg, float(wt))
for tg, elt in sush.iteritems() if tg in tg_int
for tf, wt in zip(elt['tfs'], elt['weights']) if tf in tf_int]
elif comp_net == 'sna':
#Sushmita network with unsigned edges
c_edges = [(tf, tg, abs(float(wt)))
for tg, elt in sush.iteritems() if tg in tg_int
for tf, wt in zip(elt['tfs'], elt['weights']) if tf in tf_int]
elif comp_net == 'kn':
c_edges = [(tf, tg, float(wt))
for tg, elt in ktgs.iteritems() if tg in tg_int
for tf, wt in zip(elt['tfs'], elt['weights']) if tf in tf_int]
ng = 4
nodes = array(list(tf_int.union(tg_int)))
bg = nx.DiGraph()
bg.add_nodes_from(nodes)
bg.add_weighted_edges_from(b_edges)
cg = nx.DiGraph()
cg.add_nodes_from(nodes)
cg.add_weighted_edges_from(c_edges)
cgraphs = {comp_net:cg}
v0 = cgraphs.values()
k0 = cgraphs.keys()
for k,g in zip(k0,v0):
for prc in [10]:
thr = percentile([e[2]['weight']
for e in nx.to_edgelist(g)], prc)
cgraphs.update([('{0}_thr{1:2.2}'.format(k,thr),
nfu.thr_graph(g,thr))])
gt = nfu.thr_graph(g,thr)
v0 = cgraphs.values()
k0 = cgraphs.keys()
for k, v in zip(k0,v0):
tot_edges = len(nx.to_edgelist(v))
for n_c in [1,2,4]:
for max_edges in array([.2,.5,1.]) * tot_edges :
if not 'thr' in k:
continue
gfilt = nfu.filter_graph(v, n_c = n_c)
gfilt = nfu.top_edges(gfilt, max_edges = max_edges)
gthr = nfu.thr_graph(gfilt, 1e-8)
cgraphs.update([('{0}_flt{1}_nedge{2}'.format(k,n_c,max_edges),gfilt)])
cgraphs.update([('{0}_flt{1}_nedge{2}_thr0'.format(k,n_c,max_edges),gthr)])
'''
When you don't have hidden variables, networks can be modelled mby information criterion.
In what settings can you incur causality from datasets.
You need a prior to limit the number of arrowsin your graph:
The idea: come up with an idea from computational learning theory
and come up with a model for interventions.
Spatial, genetic, time data to penalize network edges...
Granger causality uses time varying data to
'''
#nfplots.show(kg,pos,node_color = 'none')
#nfplots.show(fg,pos,node_color = 'white', alpha = .2, with_labels = False)
return bg, cgraphs
def run( reset = False,
base_net = 'kn',
comp_net = 'fn',
demand_bdtnp = False):
tgs,tfs = nio.getNet()
ktgs,ktfs = nio.getKNet()
bd = nio.getBDTNP()
#btgs,btfs = nio.getBDTNP()
sush = nio.getSush(on_fail = 'compute')
tfset = set(ktfs.keys())
tgset = set(ktgs.keys())
tg_int = set(tgs.keys()).intersection(ktgs.keys())
tf_int = set(tfs.keys()).intersection(ktfs.keys())
if demand_bdtnp:
tg_int = tg_int.intersection(bd.keys())
tf_int = tf_int.intersection(bd.keys())
sfRN = [(tf, tg, float(wt))
for tg, elt in tgs.iteritems() if tg in tg_int
for tf, wt in zip(elt['tfs'], elt['weights']) if tf in tf_int]
kRN = [(tf, tg, float(wt))
for tg, elt in ktgs.iteritems() if tg in tg_int
for tf, wt in zip(elt['tfs'], elt['weights']) if tf in tf_int]
#Sushmita network with signed edges
suRN = [(tf, tg, float(wt))
for tg, elt in sush.iteritems() if tg in tg_int
for tf, wt in zip(elt['tfs'], elt['weights']) if tf in tf_int]
#Sushmita network with unsigned edges
suaRN = [(tf, tg, abs(float(wt)))
for tg, elt in sush.iteritems() if tg in tg_int
for tf, wt in zip(elt['tfs'], elt['weights']) if tf in tf_int]
edges = [ kRN, sfRN, suRN, suaRN]
ng = 4
fg, kg, sug, suag = [nx.DiGraph() for i in range(4)]
nodes = array(list(tf_int.union(tg_int)))
graphs = {'kg':kg,'fg':fg,'sug':sug,'suag':suag}
for g, edges in zip(graphs.values(), edges):
g.add_nodes_from(nodes)
g.add_weighted_edges_from(edges)
for gname in ['fg','suag']:
for prc in [10,50,75,85,90,95,98,99]:
thr = percentile([e[2]['weight'] for e in nx.to_edgelist(graphs[gname])], prc)
graphs.update([('{0}_thr{1:2.2}'.format(gname,thr),
nfu.thr_graph(graphs[gname],thr))])
v0 = graphs.values()
k0 = graphs.keys()
tot_edges = len(nx.to_edgelist(graphs['fg']))
for k, v in zip(k0,v0):
for n_c in [2,4,8 ,12]:
for max_edges in array([.5,1.,2.,5.]) * tot_edges :
if not 'thr' in k:
continue
gfilt = nfu.filter_graph(v, n_c = n_c)
gfilt = nfu.top_edges(gfilt, max_edges = max_edges)
gthr = nfu.thr_graph(gfilt, 1e-8)
graphs.update([('{0}_flt{1}'.format(k,n_c),gfilt)])
graphs.update([('{0}_flt{1}_thr0'.format(k,n_c),gthr)])
#nfplots.show(kg,pos,node_color = 'none')
#nfplots.show(fg,pos,node_color = 'white', alpha = .2, with_labels = False)
return graphs
