p2d1Path = os.path.join(io.DeepLearningRoot(),'Data/Person2Day1_baseline.csv')
p2d2Path = os.path.join(io.DeepLearningRoot(),'Data/Person2Day2_baseline.csv')
p1Dae = load_model(os.path.join(io.DeepLearningRoot(),'savedModels/person1_baseline_DAE.h5'))
p2Dae = load_model(os.path.join(io.DeepLearningRoot(),'savedModels/person2_baseline_DAE.h5'))
p1d1 = genfromtxt(p1d1Path, delimiter=',', skip_header=0)
p1d2 = genfromtxt(p1d2Path, delimiter=',', skip_header=0)
p2d1 = genfromtxt(p2d1Path, delimiter=',', skip_header=0)
p2d2 = genfromtxt(p2d2Path, delimiter=',', skip_header=0)
# pre-process data: log transformation, a standard practice with CyTOF data
p1d1 = dh.preProcessCytofData(p1d1)
p1d2 = dh.preProcessCytofData(p1d2)
p2d1 = dh.preProcessCytofData(p2d1)
p2d2 = dh.preProcessCytofData(p2d2)
if denoise:
p1d1 = p1Dae.predict(p1d1)
p1d2 = p1Dae.predict(p1d2)
p2d1 = p2Dae.predict(p2d1)
p2d2 = p2Dae.predict(p2d2)
# rescale source to have zero mean and unit variance
# apply same transformation to the target
p1d1_pp = prep.StandardScaler().fit(p1d1)
p1d2_pp = prep.StandardScaler().fit(p1d2)
p2d1_pp = prep.StandardScaler().fit(p2d1)
if data == 'person1_3month':
sourcePath = os.path.join(io.DeepLearningRoot(),
'Data/Person1Day1_3month.csv')
targetPath = os.path.join(io.DeepLearningRoot(),
'Data/Person1Day2_3month.csv')
if data == 'person2_3month':
sourcePath = os.path.join(io.DeepLearningRoot(),
'Data/Person2Day1_3month.csv')
targetPath = os.path.join(io.DeepLearningRoot(),
'Data/Person2Day2_3month.csv')
source = genfromtxt(sourcePath, delimiter=',', skip_header=0)
target = genfromtxt(targetPath, delimiter=',', skip_header=0)
# pre-process data: log transformation, a standard practice with CyTOF data
target = dh.preProcessCytofData(target)
source = dh.preProcessCytofData(source)
numZerosOK = 1
toKeepS = np.sum((source == 0), axis=1) <= numZerosOK
print(np.sum(toKeepS))
toKeepT = np.sum((target == 0), axis=1) <= numZerosOK
print(np.sum(toKeepT))
inputDim = target.shape[1]
if denoise:
trainTarget_ae = np.concatenate([source[toKeepS], target[toKeepT]], axis=0)
np.random.shuffle(trainTarget_ae)
trainData_ae = trainTarget_ae * np.random.binomial(
n=1, p=keepProb, size=trainTarget_ae.shape)
'Data/Person2Day2_3month.csv')
sourceLabelPath = os.path.join(io.DeepLearningRoot(),
'Data/Person2Day1_3month_label.csv')
targetLabelPath = os.path.join(io.DeepLearningRoot(),
'Data/Person2Day2_3month_label.csv')
autoencoder2 = load_model(
os.path.join(io.DeepLearningRoot(),
'savedModels/person2_3month_DAE.h5'))
source2 = genfromtxt(sourcePath, delimiter=',', skip_header=0)
target2 = genfromtxt(targetPath, delimiter=',', skip_header=0)
sourceLabels2 = genfromtxt(sourceLabelPath, delimiter=',', skip_header=0)
targetLabels2 = genfromtxt(targetLabelPath, delimiter=',', skip_header=0)
# pre-process data: log transformation, a standard practice with CyTOF data
target1 = dh.preProcessCytofData(target1)
source1 = dh.preProcessCytofData(source1)
target2 = dh.preProcessCytofData(target2)
source2 = dh.preProcessCytofData(source2)
if denoise:
source1 = autoencoder1.predict(source1)
target1 = autoencoder1.predict(target1)
source2 = autoencoder2.predict(source2)
target2 = autoencoder2.predict(target2)
# rescale source to have zero mean and unit variance
# apply same transformation to the target
preprocessor1 = prep.StandardScaler().fit(source1)
source1 = preprocessor1.transform(source1)
target1 = preprocessor1.transform(target1)