def experiment3(delt=1.5,
numIter=30,
low=100,
high=600,
step=100,
numTrain=30000,
numTest=10000):
print "experiment 3"
dimDict = {}
for numCovs in np.arange(low,high+step,step):
print " numCovs=", numCovs
datgen = dc.DataSim(nCovariates=numCovs,
nHeterogenous=1,
delta=np.array([[delt]]))
datmodel=mdl.simModel(generator=datgen)
data=[]
for i in np.arange(0,numIter):
print " iter=",i
Xtrain, Ztrain, Ytrain = datmodel.generate(numTrain)
datmodel.train(Xtrain, Ztrain, Ytrain)
Xtest, Ztest, Ytest = datmodel.generate(numTest)
accuracy, h1, h2 = datmodel.test(Xtest,Ztest,Ytest)
errs,avgerrs,stderrs = datmodel.getMeansError()
data.append((accuracy,avgerrs,h1,h2,datmodel))
datmodel.regenerate() #re-generate the coefficients
dimDict[numCovs]=data
return dimDict
def experiment(args):
nPatients = args[0]
nDims = args[1]
nSparse = args[2]
nHeter = args[3]
heterMeans = args[4] #delta
nIters = args[5]
nTest = args[6]
Bayes = args[7]
datgen = dc.DataSim(nCovariates=nDims,
nHeterogenous=nHeter,
delta=heterMeans) #needs sparsity
datmodel = mdl.simModel(generator = datgen, BIC = Bayes)
def experiment1(numIter=30,
numTrain=30000,
numTest=10000):
print "experiment 1"
deltaDict = {}
for delt in np.arange(-3.0,3.5,0.5):
print " delta=",delt
datgen = dc.DataSim(nCovariates=100,
nHeterogenous=1,
delta=np.array([[delt]]))
datmodel = mdl.simModel(generator = datgen)
data = []
for i in np.arange(0,numIter):
print " iter=",i
Xtrain, Ztrain, Ytrain = datmodel.generate(numTrain)
datmodel.train(Xtrain, Ztrain, Ytrain)
Xtest, Ztest, Ytest = datmodel.generate(numTest)
accuracy, h1, h2 = datmodel.test(Xtest,Ztest,Ytest)
errs,avgerrs,stderrs = datmodel.getMeansError()
data.append((accuracy,avgerrs,h1,h2,datmodel))
datmodel.regenerate() #re-generate the coefficients
deltaDict[delt]=data
return deltaDict