def test_mixedLogistic_train_tribolium(self):
# --- prepare data ---
data = prepareTriboliumData()
# --- set initial ---
beta0, alpha0 = set_initial(data['c'], data['xm'].shape[1], data['xr'].shape[1])
vBetaAlpha0 = mapOptimizationMatrices2Vector(beta0, alpha0)
# --- Estimation ---
res = mixedLogistic_train(vBetaAlpha0, data['c'], data['y'], data['xm'], data['xr'], data['m'])
mapOptimizationVector2Matrices(res.x, data['c'], data['xm'].shape[1])
def test_mixedLogistic_QNtrain(self):
# prepare data
data = prepareTriboliumData()
# set initial
beta0, alpha0 = set_initial(data['c'], data['xm'].shape[1], data['xr'].shape[1])
vBetaAlpha0 = mapOptimizationMatrices2Vector(beta0, alpha0)
nollk = nollkForOptimization(data['c'], data['y'], data['xm'], data['xr'], data['m'])
res = mixedLogistic_QNtrain(nollk, vBetaAlpha0)
print '\n'
print 'parameter estimates =\n', res.x
print 'minimum fval =', res.fun
mapOptimizationVector2Matrices(res.x, data['c'], data['xm'].shape[1])
def test_minimize_negLogLikelihoodForOptimize_tribolium(self):
# read data
df = import_data()
xm = pd.get_dummies(df.Replicate).ix[:, 2:]
xr = pd.get_dummies(df.Species).ix[:, 1:]
m = np.matrix(df.Total).T
y = np.matrix(df.Remaining).T
# pre-process data
xm = addIntercept(xm) # add a leading 1s column
xr = addIntercept(xr) # add a leading 1s column
c = 3 # three replicates as three components
# set initial
beta0, alpha0 = set_initial(c, xm.shape[1], xr.shape[1])
vBetaAlpha0 = mapOptimizationMatrices2Vector(beta0, alpha0)
res = sp.optimize.minimize(nellkForOptimization, x0=vBetaAlpha0,
args=(vBetaAlpha0,c,y,xm,xr,m), method='BFGS')
def test_mixedLogistic_EMtrain_evaluation_tribolium(self):
# --- prepare data ---
data = prepareTriboliumData()
# --- set initial ---
beta0, alpha0 = set_initial(data['c'], data['xm'].shape[1], data['xr'].shape[1])
vBetaAlpha0 = mapOptimizationMatrices2Vector(beta0, alpha0)
# vBetaAlpha0 = np.repeat(1., 15)
# --- EM ---
# param, fval, nIter = mixedLogistic_EMtrain(vBetaAlpha0, c, y, xm, xr, m)
res = mixedLogistic_EMtrain(vBetaAlpha0, data['c'], data['y'], data['xm'], data['xr'], data['m'])
# param in matrix form
beta, alpha = mapOptimizationVector2Matrices(res['param'], data['c'], data['xm'].shape[1])
print 'beta ='
print beta
print 'alpha ='
print alpha
def test_mixedLogistic_Mstep_evaluation_tribolium(self):
# read data
df = import_data()
xm = pd.get_dummies(df.Replicate).ix[:, 2:]
xr = pd.get_dummies(df.Species).ix[:, 1:]
m = np.matrix(df.Total).T
y = np.matrix(df.Remaining).T
# pre-process data
xm = addIntercept(xm) # add a leading 1s column
xr = addIntercept(xr) # add a leading 1s column
c = 3 # three replicates as three components
# set initial
beta0, alpha0 = set_initial(c, xm.shape[1], xr.shape[1])
vBetaAlpha0 = mapOptimizationMatrices2Vector(beta0, alpha0)
# E-step
qfn = mixedLogistic_Estep(vBetaAlpha0, c, y, xm, xr, m)
# M-step
param, fval = mixedLogistic_Mstep(qfn, vBetaAlpha0)
print '\n'
print 'parameter estimates = \n', param
print 'function value = ', fval
print '\n'
c_candidates = np.arange(10) + 1 # candidates for number of hidden groups
# It is tuned by cross-validation
nfolds = 10 # 'nfolds' cross-validation
batchSize = trainAndValidateXm.shape[0] / nfolds
# params = []
print('--- Start %d-fold cross-validation ---' % (nfolds))
avgValidateScores = []
for c in c_candidates:
print(" \tCurrent number of hidden groups 'c' = %d " % (c))
validateScores = []
for batchIndex in xrange(nfolds):
print('\t\t Batch %d ...' % (batchIndex))
# Initial parameters
random.seed(25)
b0, a0 = set_initial(c, data.dxm + 1, data.dxr + 1)
param0 = mapOptimizationMatrices2Vector(b0, a0)
validateIds = np.arange(batchIndex * batchSize, (batchIndex + 1) * batchSize)
trainIds = list(set(np.arange(trainAndValidateXm.shape[0])) - set(validateIds))
res = mixedLogistic_EMtrain(param0=param0, c=c,
y=np.matrix(trainAndValidateY[trainIds]).T,
xm=trainAndValidateXm[trainIds, :],
xr=trainAndValidateXr[trainIds, :],
m=1)
pred = mixedLogistic_pred(res=res,
xmt=trainAndValidateXm[validateIds, :],
xrt=trainAndValidateXr[validateIds, :],
c=c,
