def _get_par(self, datao):
gaussiane = [ estimate_gaussian_per_class(datao, at, common_if_extreme=True) for at in range(len(datao.domain.attributes)) ]
normalizec = []
for i,g in zip(range(len(datao.domain.attributes)), gaussiane):
r = [ _llrlogratio(ex[i].value, *g) for ex in datao ]
normalizec.append((mean(r), std(r)))
return gaussiane, normalizec
def _get_par(self, datao):
gaussiane = [
estimate_gaussian_per_class(datao, at, common_if_extreme=True)
for at in range(len(datao.domain.attributes))
]
normalizec = []
for i, g in zip(range(len(datao.domain.attributes)), gaussiane):
r = [_llrlogratio(ex[i].value, *g) for ex in datao]
normalizec.append((mean(r), std(r)))
return gaussiane, normalizec
def estimate_gaussian_per_class(data,
i,
a=None,
b=None,
common_if_extreme=False):
cv = data.domain.class_var
if a == None: a = cv.values[0]
if b == None: b = cv.values[1]
def avWCVal(value):
return [
ex[i].value for ex in data
if ex[-1].value == value and not ex[i].isSpecial()
]
list1 = avWCVal(a)
list2 = avWCVal(b)
mi1 = mi2 = st1 = st2 = None
try:
mi1 = statc.mean(list1)
st1 = statc.std(list1)
except:
pass
try:
mi2 = statc.mean(list2)
st2 = statc.std(list2)
except:
pass
def extreme():
return st1 == 0 or st2 == 0
if common_if_extreme and extreme():
st1 = st2 = statc.std(list1 + list2)
return mi1, st1, mi2, st2
def build_feature(self, data, gs):
at = Orange.feature.Continuous(name=str(gs))
geneset = list(gs.genes)
nm, name_ind, genes, takegenes, to_geneset = self._match_data(
data, geneset, odic=True)
gsi = [name_ind[g] for g in genes]
gausse = compute_llr(data, gsi, self._gauss_cache)
genes_gs = [to_geneset[g] for g in genes]
if self.normalize: # per (3) in the paper
#compute log ratios for all samples and genes from this gene set
for i, gene_gs, g in zip(gsi, genes_gs, gausse):
if gene_gs not in self._normalizec: #skip if computed already
r = [_llrlogratio(ex[i].value, *g) for ex in data]
self._normalizec[gene_gs] = (mean(r), std(r))
def t(ex,
w,
genes_gs=genes_gs,
gausse=gausse,
normalizec=self._normalizec):
nm2, name_ind2, genes2 = self._match_instance(ex, genes_gs, None)
gsvalues = [vou(ex, gn, name_ind2) for gn in genes2]
vals = [
_llrlogratio(v, *g) if v != "?" else 0.0
for v, g in zip(gsvalues, gausse)
]
if len(normalizec): #normalize according to (3)
vals2 = []
for v, g in zip(vals, genes_gs):
m, s = normalizec[g]
if s == 0: #disregard attributes without differences
vals2.append(0.)
else:
vals2.append((v - m) / s)
vals = vals2
return sum(vals)
at.get_value_from = t
return at
def build_feature(self, data, gs):
at = Orange.feature.Continuous(name=str(gs))
geneset = list(gs.genes)
nm, name_ind, genes, takegenes, to_geneset = self._match_data(data, geneset, odic=True)
gsi = [ name_ind[g] for g in genes ]
gausse = compute_llr(data, gsi, self._gauss_cache)
genes_gs = [ to_geneset[g] for g in genes ]
if self.normalize: # per (3) in the paper
#compute log ratios for all samples and genes from this gene set
for i, gene_gs, g in zip(gsi, genes_gs, gausse):
if gene_gs not in self._normalizec: #skip if computed already
r = [ _llrlogratio(ex[i].value, *g) for ex in data ]
self._normalizec[gene_gs] = (mean(r), std(r))
def t(ex, w, genes_gs=genes_gs, gausse=gausse, normalizec=self._normalizec):
nm2, name_ind2, genes2 = self._match_instance(ex, genes_gs, None)
gsvalues = [ vou(ex, gn, name_ind2) for gn in genes2 ]
vals = [ _llrlogratio(v, *g) if v != "?" else 0.0 for v,g in zip(gsvalues, gausse) ]
if len(normalizec): #normalize according to (3)
vals2 = []
for v,g in zip(vals, genes_gs):
m,s = normalizec[g]
if s == 0: #disregard attributes without differences
vals2.append(0.)
else:
vals2.append((v-m)/s)
vals = vals2
return sum(vals)
at.get_value_from = t
return at
def stdev(l):
return statc.std(l)
