def motifOrder(data,
key,
epsilon,
orderSize=3,
motifSize=3,
degree=10,
swap=False):
"""Sorts graphs into bins based on motif frequency ordering"""
if key == "rand":
graphs = []
for i in xrange(100):
x = np.random.rand(88, 88)
x -= np.diag(np.diag(x))
graphs.append(x)
else:
graphs = data[key]
pattern = defaultdict(float)
for G in graphs:
#calculate threshold
sortedWeights = np.sort(G, axis=None)
threshold = sortedWeights[-len(G) * degree - 1]
#Output graph to txt file
graph = nx.DiGraph(G > threshold)
graph = nx.convert_node_labels_to_integers(graph, 1)
if swap:
graph = gh.randomize_graph(graph, 2000)
with open('result/OUTPUT.txt', 'wb') as f:
f.write(str(len(graph)) + '\n')
nx.write_edgelist(graph, f, data=False)
#Jenky way to use c++ motif finder in python
subprocess.call(['./Kavosh', str(motifSize)])
data = np.loadtxt("result/MotifCount.txt", ndmin=2)
order = []
for iD, total, percent in data:
order.append((int(iD), total))
keys = sorted(order, key=lambda x: -x[1])
index = 0
groupedKeys = []
size = min(orderSize, len(keys))
while index < size:
root = keys[index][1]
element = [keys[index][0]]
index += 1
while index < size and keys[index][1] * epsilon > root:
element.append(keys[index][0])
index += 1
element.sort()
groupedKeys.append(tuple(element))
pattern[tuple(groupedKeys)] += 100 / float(len(graphs))
return pattern
def motifOrder(data,key,epsilon,orderSize=3,motifSize=3,degree=10,swap=False):
"""Sorts graphs into bins based on motif frequency ordering"""
if key == "rand":
graphs = []
for i in xrange(100):
x = np.random.rand(88,88)
x -= np.diag(np.diag(x))
graphs.append(x)
else:
graphs = data[key]
pattern = defaultdict(float)
for G in graphs:
#calculate threshold
sortedWeights = np.sort(G,axis=None)
threshold = sortedWeights[-len(G)*degree-1]
#Output graph to txt file
graph = nx.DiGraph(G>threshold)
graph = nx.convert_node_labels_to_integers(graph,1)
if swap:
graph = gh.randomize_graph(graph, 2000)
with open('result/OUTPUT.txt','wb') as f:
f.write(str(len(graph)) + '\n')
nx.write_edgelist(graph,f,data=False)
#Jenky way to use c++ motif finder in python
subprocess.call(['./Kavosh', str(motifSize)])
data = np.loadtxt("result/MotifCount.txt",ndmin=2)
order = []
for iD,total,percent in data:
order.append((int(iD),total))
keys = sorted(order,key=lambda x:-x[1])
index = 0
groupedKeys = []
size = min(orderSize,len(keys))
while index < size:
root = keys[index][1]
element = [keys[index][0]]
index += 1
while index < size and keys[index][1]*epsilon > root:
element.append(keys[index][0])
index += 1
element.sort()
groupedKeys.append(tuple(element))
pattern[tuple(groupedKeys)] += 100/float(len(graphs))
return pattern
def makeSwapData(degree=10):
with open("aznorbert_corrsd_new.pkl","rb") as f:
data = pickle.load(f)
swapData = {}
for key, graphs in data.iteritems():
print key
keyData = []
for i,G in enumerate(graphs):
print i
sortedWeights = np.sort(G,axis=None)
threshold = sortedWeights[-len(G)*degree-1]
graph = nx.DiGraph(G>threshold)
diff = gh.randomize_graph(graph, 2500)
keyData.append(graph)
swapData[key] = keyData
with open("SwapData"+str(degree)+".pkl",'wb') as f:
pickle.dump(swapData,f)
def findMotifs(data,key,motifSize=3,degree=10,randomize=False):
"""Main finding motifs routine"""
usetotal = False
#Check cache
filename = str(key)+'s'+str(int(motifSize))+'d'+str(int(degree))+str(usetotal)+".pkl"
if os.path.exists('cache/'+filename) and USECACHE:
print "in cache"
with open('cache/'+filename,"rb") as f:
return pickle.load(f)
if key == "rand":
graphs = []
for i in xrange(100):
x = np.random.rand(88,88)
x -= np.diag(np.diag(x))
graphs.append(x)
else:
graphs = data[key]
motifs = defaultdict(list)
numstring ="/"+str(len(graphs))
rejected = 0
for index,G in enumerate(graphs):
#Cull bad graphs
if np.count_nonzero(G)<len(G)*degree:
rejected += 1
continue
#calculate threshold
sortedWeights = np.sort(G,axis=None)
threshold = sortedWeights[-len(G)*degree-1]
#Print progress
sys.stdout.write("\rMotif Finding Progress: "+str(index)+numstring)
sys.stdout.write(" Threshold: "+str(threshold))
sys.stdout.flush()
#Output graph to txt file
graph = nx.DiGraph(G>threshold)
graph = nx.convert_node_labels_to_integers(graph,1)
if randomize:
graph = gh.randomize_graph(graph, 2000)
with open('result/OUTPUT.txt','wb') as f:
f.write(str(len(graph)) + '\n')
nx.write_edgelist(graph,f,data=False)
#Jenky way to use c++ motif finder in python
os.system("./Kavosh "+str(motifSize))
with open("result/MotifCount.txt","rb") as f:
subgraphs = float(f.next())
data = np.loadtxt(f, ndmin=2)
for iD,total in data:
percent = total/subgraphs
motifs[int(iD)].append(percent)
print '\nMotifs Done! Graphs Rejected: '+str(rejected)
#add zeros to graphs that didn't contain motifs
for key,value in motifs.iteritems():
numZero = len(graphs)-len(value)-rejected
value.extend([0 for derp in xrange(numZero)])
motifs[int(key)] = np.array(value)
motifs = dict(motifs)
#add motifs to cache
if USECACHE:
with open('cache/'+filename,'wb') as f:
pickle.dump(motifs,f)
return motifs
f.write(
"\\end{tabular}\n"
"\\end{table}\n"
)
f.write("\\end{document}\n")
os.system("pdflatex -output-directory result " + filename)
os.system("rm result/*.log result/*.aux")
if __name__ == '__main__':
with open("aznorbert_corrsd.pkl","rb") as f:
data = pickle.load(f)
#PDFOrder(data,"AllOrder",1.05)
for i,G in enumerate(data[('NL','corr')]):
print i
sortedWeights = np.sort(G,axis=None)
threshold = sortedWeights[-len(G)*10-1]
graph = nx.DiGraph(G>threshold)
graph = gh.randomize_graph(graph, 1000)
#PDFstats(data,"Motif_Statistics_Mats",False)
#motifStats(data)
#print 'Normal'
#motifOrder(data,('NL','corr'),1.05)
def findMotifs(data, key, motifSize=3, degree=10, randomize=False):
"""Main finding motifs routine"""
usetotal = False
#Check cache
filename = str(key) + 's' + str(int(motifSize)) + 'd' + str(
int(degree)) + str(usetotal) + ".pkl"
if os.path.exists('cache/' + filename) and USECACHE:
print "in cache"
with open('cache/' + filename, "rb") as f:
return pickle.load(f)
if key == "rand":
graphs = []
for i in xrange(100):
x = np.random.rand(88, 88)
x -= np.diag(np.diag(x))
graphs.append(x)
else:
graphs = data[key]
motifs = defaultdict(list)
numstring = "/" + str(len(graphs))
rejected = 0
for index, G in enumerate(graphs):
#Cull bad graphs
if np.count_nonzero(G) < len(G) * degree:
rejected += 1
continue
#calculate threshold
sortedWeights = np.sort(G, axis=None)
threshold = sortedWeights[-len(G) * degree - 1]
#Print progress
sys.stdout.write("\rMotif Finding Progress: " + str(index) + numstring)
sys.stdout.write(" Threshold: " + str(threshold))
sys.stdout.flush()
#Output graph to txt file
graph = nx.DiGraph(G > threshold)
graph = nx.convert_node_labels_to_integers(graph, 1)
if randomize:
graph = gh.randomize_graph(graph, 2000)
with open('result/OUTPUT.txt', 'wb') as f:
f.write(str(len(graph)) + '\n')
nx.write_edgelist(graph, f, data=False)
#Jenky way to use c++ motif finder in python
os.system("./Kavosh " + str(motifSize))
with open("result/MotifCount.txt", "rb") as f:
subgraphs = float(f.next())
data = np.loadtxt(f, ndmin=2)
for iD, total in data:
percent = total / subgraphs
motifs[int(iD)].append(percent)
print '\nMotifs Done! Graphs Rejected: ' + str(rejected)
#add zeros to graphs that didn't contain motifs
for key, value in motifs.iteritems():
numZero = len(graphs) - len(value) - rejected
value.extend([0 for derp in xrange(numZero)])
motifs[int(key)] = np.array(value)
motifs = dict(motifs)
#add motifs to cache
if USECACHE:
with open('cache/' + filename, 'wb') as f:
pickle.dump(motifs, f)
return motifs
f.write(line + " \\\\ \\hline\n")
f.write("\\end{tabular}\n" "\\end{table}\n")
f.write("\\end{document}\n")
os.system("pdflatex -output-directory result " + filename)
os.system("rm result/*.log result/*.aux")
if __name__ == '__main__':
with open("aznorbert_corrsd.pkl", "rb") as f:
data = pickle.load(f)
#PDFOrder(data,"AllOrder",1.05)
for i, G in enumerate(data[('NL', 'corr')]):
print i
sortedWeights = np.sort(G, axis=None)
threshold = sortedWeights[-len(G) * 10 - 1]
graph = nx.DiGraph(G > threshold)
graph = gh.randomize_graph(graph, 1000)
#PDFstats(data,"Motif_Statistics_Mats",False)
#motifStats(data)
#print 'Normal'
#motifOrder(data,('NL','corr'),1.05)