def testBroad():
data = Data()
geneExp = data.get_gene_exp_matrix()
labels = data.get_labels()
geneNames = data.get_gene_names()
print(labels)
print(geneExp[0])
print(geneExp[5])
print(geneExp[17])
print(len(geneExp))
print(len(geneExp[0]))
def ecocfn(code_size=52./38, C=50, linSVC_L1=False, kernel ='linear', selection='chi2',numFeatures=330):#try different code sizes
"""l1Reg is used only if linSVC=True"""
data = Data()
gene_exp = data.get_gene_exp_matrix()
labels = data.get_labels()
if linSVC_L1:
clf = svm.LinearSVC(C=C,penalty='l1',dual=False)
else:
clf = svm.SVC(C=C,kernel=kernel)
ecocAlgo = OutputCodeClassifier(clf, code_size=code_size, random_state=0)
print(leaveOneOutCrossValid(gene_exp,labels,ecocAlgo,selection=selection,numFeatures=numFeatures))
def svmfn(featureSelectionMethod = 'none',numFeaturesA = '330'):
data = Data()
gene_exp = data.get_gene_exp_matrix()
labels = data.get_labels()
names = data.get_gene_names()
#USES a 1 vs 1 scheme - how does this work?
clf = svm.SVC(C=125.,kernel='linear') #kernel can be poly, rbf, linear, sigmoid
clfwrapper = OneVsRestClassifier(clf);
#accuracy = leaveOneOutCrossValid(gene_exp,labels,clfwrapper,names=names,selection=featureSelectionMethod,numFeatures=numFeaturesA)
#print 'accuracy is'
#print accuracy
trainingError =trainingSetPerformance(gene_exp,labels,clfwrapper,names=names,selection=featureSelectionMethod,numFeatures=numFeaturesA)
#print 'accuracy is'
#print accuracy
print 'trainingError is'
print trainingError
def svmfn(featureSelectionMethod = 'none',numFeatures = '330'):
data = Data()
#data = RaviNormal()
gene_exp = data.get_gene_exp_matrix()
labels = data.get_labels()
names = data.get_gene_names()
#clf = svm.LinearSVC(C=125.,penalty="l1",dual=False,class_weight='auto')
clf = svm.LinearSVC(C=125,penalty="l1",dual=False)
clfwrapper = OneVsRestClassifier(clf);
#accuracy = leaveOneOutCrossValid(gene_exp,labels,clfwrapper,names=names,selection=featureSelectionMethod,numFeatures=numFeatures)
trainingError =trainingSetPerformance(gene_exp,labels,clfwrapper,names=names,selection=featureSelectionMethod,numFeatures=numFeatures)
print 'accuracy is'
#print accuracy
print 'trainingError is'
print trainingError
estimators = clfwrapper.estimators_
j = 0
totalGeneListLength = 0
for estimator in estimators:
print 'estimator for class'
print data.getCellName(j)
i = range(0,len(estimator.coef_[0]))
b = sorted(zip(estimator.coef_[0], i), reverse=True)[:80] #TODO CHANGE
indices = data.indices_of_celltype(j)
#print 'indices of this class:'
#print indices
arraysum = [0.0]*11927
arraysum = numpy.array(arraysum)
for i in indices:
arraysum = numpy.add(arraysum,gene_exp[i])
arrayavg = numpy.divide(arraysum,len(indices))
k = 0
geneList = []
while k<80 and b[k][0] > 0: #TODO CHANGE
avg_expr = arrayavg[b[k][1]]
geneStr = str(b[k][0])+',' +names[b[k][1]] + ':' + str(avg_expr)
geneList = geneList + [geneStr]
k = k+1
j = j+1
print geneList
print len(geneList)
totalGeneListLength += len(geneList)
print 'avg gene signature size:'
print totalGeneListLength/35
