def depExportGraph(self, f, n_int=1, print_bits = 1, black_white = 0, undirected = 1, significant_digits = 2, pretty_names = 1, pcutoff=-1, postscript=1, spanning_tree = 1, TAN=1, source=-1, labelled=1,jaccard=1,filter=[],diagonal=0,pvlabel=0):
NA = self.NA
### SELECTION OF INTERACTIONS AND ATTRIBUTES ###
links = []
maxlink = -1e6
if n_int == 1 and spanning_tree:
# prepare table
lmm = []
for i in range(1,NA):
ei = self.ents[(i,)]
for j in range(i):
ej = self.ents[(j,)]
if TAN:
# I(A;B|C)
v = self.way3[(j,i,-1)].InteractionInformation()
v += self.way2[(j,i)].InteractionInformation()
else:
if jaccard:
v = self.way2[(j,i)].JaccardInteraction() # I(A;B) chow-liu, mutual information
else:
v = self.way2[(j,i)].InteractionInformation() # I(A;B) chow-liu, mutual information
if ei > ej:
lmm.append((abs(v),v,ej,(j,i)))
else:
lmm.append((abs(v),v,ei,(i,j)))
lmm.sort()
maxlink = lmm[-1][0]
# use Prim's algorithm here
mapped = []
for i in range(NA):
mapped.append(i)
n = NA
idx = -1 # running index in the sorted array of possible links
while n > 1:
# find the cheapest link
while 1:
(av,v,e,(i,j)) = lmm[idx]
idx -= 1
if mapped[i] != mapped[j]:
break
links.append((v,(i,j),e))
toremove = mapped[j]
for k in range(NA):
if mapped[k] == toremove:
mapped[k] = mapped[i]
n -= 1
else:
# select the top
lmm = []
for i in range(NA):
if filter==[] or self.names[i] in filter:
for j in range(i):
if filter==[] or self.names[j] in filter:
ii = max(i,j)
jj = min(i,j)
if jaccard and pcutoff < 0.0:
if self.ents[(jj,ii)] == 0.0:
v = 1.0
else:
v = self.way2[(jj,ii)].JaccardInteraction()
lmm.append((v,(i,j)))
else:
v = self.way2[(jj,ii)].InteractionInformation()
if pcutoff >= 0.0:
xt = self.way2[(jj,ii)]
dof = 1.0
dof *= len(xt.values[0])-1
dof *= len(xt.values[1])-1
pv = orngContingency.getPvalueDOF(v,xt,dof)
if pv <= pcutoff:
v = 1-pv
lmm.append((v,(i,j)))
else:
lmm.append((v,(i,j)))
lmm.sort()
maxlink = max(lmm[-1][0],maxlink)
links += [(v,p,1.0) for (v,p) in lmm[-n_int:]]
# mark vertices
mv = [0 for x in range(NA)]
for (v,(i,j),e) in links:
mv[i] = 1
mv[j] = 1
# output the attributes
f.write("digraph G {\n")
if print_bits:
shap = 'record'
else:
shap = 'box'
for n in range(NA):
if mv[n]:
if source != -1 and not type(source)==type(1):
# find the name
if string.upper(self.names[n])==string.upper(source):
source = n
t = '%s'%self.names[n]
if pretty_names:
t = string.replace(t,"ED_","")
t = string.replace(t,"D_","")
t = string.replace(t,"M_","")
t = string.replace(t," ","\\n")
t = string.replace(t,"-","\\n")
t = string.replace(t,"_","\\n")
if print_bits:
#t = "{%s|%s}"%(t,_nicefloat(self.ents[(n,)],significant_digits))
t = "{%s|%s}"%(t,_nicefloat(self.way2[(n,-1)].total,significant_digits))
f.write("\tnode [ shape=%s, label = \"%s\"] %d;\n"%(shap,t,n))
if source != -1:
# redirect all links
age = [-1]*NA
age[source] = 0
phase = 1
remn = NA-1
premn = -1
while remn > 0 and premn != remn:
premn = remn
for (v,(i,j),e) in links:
if age[i] >= 0 and age[i] < phase and age[j] < 0:
age[j] = phase
remn -= 1
if age[j] >= 0 and age[j] < phase and age[i] < 0:
age[i] = phase
remn -= 1
phase += 1
### EDGE DRAWING ###
for (v,(i,j),e) in links:
if v > 0:
c = v/e
perc = int(100*v/maxlink + 0.5)
style = ''
if postscript:
style += "style=\"setlinewidth(%d)\","%(abs(perc)/30+1)
if not black_white:
l = 0.3+0.7*perc/100.0
style += 'color="0.5 %f %f",'%(l,1-l) # adjust saturation
if labelled:
if diagonal:
ct = self.way2[(min(i,j),max(i,j))]
(ni,nj) = numpy.shape(ct.m)
cc = 0.0
if ni==nj:
for x in range(ni):
cc += ct.m[x,x]
style += 'label=\"%s%%\",'%_nicefloat(100.0*cc/ct.total,significant_digits)
elif pvlabel and pcutoff >= 0.0:
style += 'label=\"%e\",'%(1-v)
else:
style += 'label=\"%s%%\",'%_nicefloat(100.0*c,significant_digits)
if source == -1 or undirected:
f.write("\t%d -> %d [%sweight=%d,dir=none];\n"%(j,i,style,(perc/30+1)))
else:
if age[i] > age[j]:
f.write("\t%d -> %d [%sweight=%d];\n"%(j,i,style,(perc/30+1)))
else:
f.write("\t%d -> %d [%sweight=%d];\n"%(i,j,style,(perc/30+1)))
f.write("}\n")
def depExportGraph(self,
f,
n_int=1,
print_bits=1,
black_white=0,
undirected=1,
significant_digits=2,
pretty_names=1,
pcutoff=-1,
postscript=1,
spanning_tree=1,
TAN=1,
source=-1,
labelled=1,
jaccard=1,
filter=[],
diagonal=0,
pvlabel=0):
NA = self.NA
### SELECTION OF INTERACTIONS AND ATTRIBUTES ###
links = []
maxlink = -1e6
if n_int == 1 and spanning_tree:
# prepare table
lmm = []
for i in range(1, NA):
ei = self.ents[(i, )]
for j in range(i):
ej = self.ents[(j, )]
if TAN:
# I(A;B|C)
v = self.way3[(j, i, -1)].InteractionInformation()
v += self.way2[(j, i)].InteractionInformation()
else:
if jaccard:
v = self.way2[(j, i)].JaccardInteraction(
) # I(A;B) chow-liu, mutual information
else:
v = self.way2[(j, i)].InteractionInformation(
) # I(A;B) chow-liu, mutual information
if ei > ej:
lmm.append((abs(v), v, ej, (j, i)))
else:
lmm.append((abs(v), v, ei, (i, j)))
lmm.sort()
maxlink = lmm[-1][0]
# use Prim's algorithm here
mapped = []
for i in range(NA):
mapped.append(i)
n = NA
idx = -1 # running index in the sorted array of possible links
while n > 1:
# find the cheapest link
while 1:
(av, v, e, (i, j)) = lmm[idx]
idx -= 1
if mapped[i] != mapped[j]:
break
links.append((v, (i, j), e))
toremove = mapped[j]
for k in range(NA):
if mapped[k] == toremove:
mapped[k] = mapped[i]
n -= 1
else:
# select the top
lmm = []
for i in range(NA):
if filter == [] or self.names[i] in filter:
for j in range(i):
if filter == [] or self.names[j] in filter:
ii = max(i, j)
jj = min(i, j)
if jaccard and pcutoff < 0.0:
if self.ents[(jj, ii)] == 0.0:
v = 1.0
else:
v = self.way2[(jj,
ii)].JaccardInteraction()
lmm.append((v, (i, j)))
else:
v = self.way2[(jj,
ii)].InteractionInformation()
if pcutoff >= 0.0:
xt = self.way2[(jj, ii)]
dof = 1.0
dof *= len(xt.values[0]) - 1
dof *= len(xt.values[1]) - 1
pv = orngContingency.getPvalueDOF(
v, xt, dof)
if pv <= pcutoff:
v = 1 - pv
lmm.append((v, (i, j)))
else:
lmm.append((v, (i, j)))
lmm.sort()
maxlink = max(lmm[-1][0], maxlink)
links += [(v, p, 1.0) for (v, p) in lmm[-n_int:]]
# mark vertices
mv = [0 for x in range(NA)]
for (v, (i, j), e) in links:
mv[i] = 1
mv[j] = 1
# output the attributes
f.write("digraph G {\n")
if print_bits:
shap = 'record'
else:
shap = 'box'
for n in range(NA):
if mv[n]:
if source != -1 and not type(source) == type(1):
# find the name
if string.upper(self.names[n]) == string.upper(source):
source = n
t = '%s' % self.names[n]
if pretty_names:
t = string.replace(t, "ED_", "")
t = string.replace(t, "D_", "")
t = string.replace(t, "M_", "")
t = string.replace(t, " ", "\\n")
t = string.replace(t, "-", "\\n")
t = string.replace(t, "_", "\\n")
if print_bits:
#t = "{%s|%s}"%(t,_nicefloat(self.ents[(n,)],significant_digits))
t = "{%s|%s}" % (t,
_nicefloat(self.way2[(n, -1)].total,
significant_digits))
f.write("\tnode [ shape=%s, label = \"%s\"] %d;\n" %
(shap, t, n))
if source != -1:
# redirect all links
age = [-1] * NA
age[source] = 0
phase = 1
remn = NA - 1
premn = -1
while remn > 0 and premn != remn:
premn = remn
for (v, (i, j), e) in links:
if age[i] >= 0 and age[i] < phase and age[j] < 0:
age[j] = phase
remn -= 1
if age[j] >= 0 and age[j] < phase and age[i] < 0:
age[i] = phase
remn -= 1
phase += 1
### EDGE DRAWING ###
for (v, (i, j), e) in links:
if v > 0:
c = v / e
perc = int(100 * v / maxlink + 0.5)
style = ''
if postscript:
style += "style=\"setlinewidth(%d)\"," % (abs(perc) / 30 +
1)
if not black_white:
l = 0.3 + 0.7 * perc / 100.0
style += 'color="0.5 %f %f",' % (l, 1 - l
) # adjust saturation
if labelled:
if diagonal:
ct = self.way2[(min(i, j), max(i, j))]
(ni, nj) = numpy.shape(ct.m)
cc = 0.0
if ni == nj:
for x in range(ni):
cc += ct.m[x, x]
style += 'label=\"%s%%\",' % _nicefloat(
100.0 * cc / ct.total, significant_digits)
elif pvlabel and pcutoff >= 0.0:
style += 'label=\"%e\",' % (1 - v)
else:
style += 'label=\"%s%%\",' % _nicefloat(
100.0 * c, significant_digits)
if source == -1 or undirected:
f.write("\t%d -> %d [%sweight=%d,dir=none];\n" %
(j, i, style, (perc / 30 + 1)))
else:
if age[i] > age[j]:
f.write("\t%d -> %d [%sweight=%d];\n" %
(j, i, style, (perc / 30 + 1)))
else:
f.write("\t%d -> %d [%sweight=%d];\n" %
(i, j, style, (perc / 30 + 1)))
f.write("}\n")