def get_data(self,
balanced=1,
batch_size=20,
tra_val_split=0.8,
use_validation=True):
self.use_validation = use_validation
Data = DATA()
Data.Fetch_OASIS(balanced=balanced)
Data.Train_Test(TRAIN_TEST_SPLIT, random=RANDOM_SEED)
selectors = self.ROI_nums
Data.Add_MRI(selectors)
Data.Split_Data()
# GET TRAINING AND TEST SETS
X = Data.features_train
y = Data.labels_train
y[y > 0] = 1
y[y <= 0] = 0
y.shape = (len(y), 1)
X_test = Data.features_test
y_test = Data.labels_test
y_test[y_test > 0] = 1
y_test[y_test <= 0] = 0
y_test.shape = (len(y_test), 1)
self.y_test = y_test
self.X_test = X_test
# SPLIT TRAINING INTO TRAINING/VALIDATION
len_yt = y.shape[0]
if use_validation:
training_size = floor(tra_val_split * len_yt)
else:
training_size = len_yt - 1
y_tra = y[:training_size]
X_tra = X[:training_size, ...]
y_val = y[training_size:]
X_val = X[training_size:, ...]
# CREATE TENSORFLOW DATASETS
self.train_ds = tf.data.Dataset.from_tensor_slices(
(X_tra, y_tra)).shuffle(3000).batch(batch_size)
if use_validation:
self.val_ds = tf.data.Dataset.from_tensor_slices(
(X_val, y_val)).batch(batch_size)
else:
self.val_ds = tf.data.Dataset.from_tensor_slices(
(X_val, y_val)).batch(1)
from DataPrep import DATA
from ATLAS import ATLAS
Atlas = ATLAS()
FileDir = os.getcwd()+'/SVM_Outputs/'
file_object = open(FileDir+'SVM_Log.txt', "a")
for i in range(3,46):
ROI = Atlas.dataset_files.labels[i]
FigName = "SVM_"+str(i)+"_"+ROI+".png"
Data = DATA()
Data.Train_Test(0.8)
selectors = [i] #brain" #Hippo mask [34,35], whole brain "brain"
Data.Add_MRI(selectors)
Data.Split_Data()
C_space = np.logspace(-4,1,50)
train_score = []
test_score = []
print("Feature Size = "+str(Data.features_train.shape[1]))
print("Started Training for "+FigName+"....")
for C in C_space:
SVM = svm.SVC(kernel='linear', C=C)
cvs = cross_validate(SVM,Data.features_train,Data.labels_train, cv=4,return_train_score=True)
train_score.append(np.mean(cvs["train_score"]))
test_score.append(np.mean(cvs["test_score"]))
# print("C="+str(C)+"Train"+str(np.mean(cvs["train_score"]))+"Valid"+str(np.mean(cvs["test_score"])))
maxC = C_space[test_score.index(max(test_score))]
def get_data(self, balanced=1, tra_val_split=0.8, use_validation=True):
self.use_validation = use_validation
Data = DATA()
Data.Fetch_OASIS(balanced=balanced)
Data.Train_Test(TRAIN_TEST_SPLIT, random=RANDOM_SEED)
Data.Split_Data()
# GET TRAINING AND TEST SETS
features_train = Data.features_train
y = Data.labels_train
y[y > 0] = 1
y[y < 0] = 0
y.shape = (len(y), 1)
features_test = Data.features_test
y_test = Data.labels_test
y_test[y_test > 0] = 1
y_test[y_test <= 0] = 0
y_test.shape = (len(y_test), 1)
Data.load_images()
gm_imgs_3D = Data.gm_imgs_3D
wm_imgs_3D = Data.wm_imgs_3D
gm_imgs_3D = gm_imgs_3D[..., np.newaxis]
wm_imgs_3D = wm_imgs_3D[..., np.newaxis]
idx_train = Data.idx_train
idx_test = Data.idx_test
gm_imgs_3D_train = gm_imgs_3D[idx_train, ...]
wm_imgs_3D_train = wm_imgs_3D[idx_train, ...]
gm_imgs_3D_test = gm_imgs_3D[idx_test, ...]
wm_imgs_3D_test = wm_imgs_3D[idx_test, ...]
# SPLIT TRAINING INTO TRAINING/VALIDATION
len_yt = y.shape[0]
if use_validation:
train_size = floor(tra_val_split * len_yt)
else:
train_size = len_yt - 1
y_tra = y[:train_size]
features_tra = features_train[:train_size, ...]
y_val = y[train_size:]
features_val = features_train[train_size:, ...]
gm_imgs_3D_val = gm_imgs_3D_train[train_size:, ...]
wm_imgs_3D_val = wm_imgs_3D_train[train_size:, ...]
gm_imgs_3D_tra = gm_imgs_3D_train[:train_size, ...]
wm_imgs_3D_tra = wm_imgs_3D_train[:train_size, ...]
X_tra = [gm_imgs_3D_tra] + [wm_imgs_3D_tra] + [features_tra]
X_val = [gm_imgs_3D_val] + [wm_imgs_3D_val] + [features_val]
X_test = [gm_imgs_3D_test] + [wm_imgs_3D_test] + [features_test]
self.y_test = y_test
self.X_test = X_test
self.y_tra = y_tra
self.X_tra = X_tra
self.y_val = y_val
self.X_val = X_val
def get_data(self, balanced=1, tra_val_split=0.8, use_validation=True):
self.use_validation = use_validation
Data = DATA()
Data.Fetch_OASIS(balanced=balanced)
Data.Train_Test(TRAIN_TEST_SPLIT, random=RANDOM_SEED)
Data.Split_Data()
# GET TRAINING AND TEST SETS
features_train = Data.features_train
y = Data.labels_train
y[y > 0] = 1
y[y < 0] = 0
y.shape = (len(y), 1)
features_test = Data.features_test
y_test = Data.labels_test
y_test[y_test > 0] = 1
y_test[y_test <= 0] = 0
y_test.shape = (len(y_test), 1)
Data.load_images()
selectors = self.ROI_nums
Data.get_3D_ROI(selectors)
ROIs_3D_gm = Data.ROIs_3D_gm
ROIs_3D_wm = Data.ROIs_3D_wm
idx_train = Data.idx_train
idx_test = Data.idx_test
ROIs_3D_gm_train = []
ROIs_3D_gm_test = []
ROIs_3D_wm_train = []
ROIs_3D_wm_test = []
for k in range(self.len_ROI):
ROIs_3D_gm_train.append(ROIs_3D_gm[k][idx_train, ...])
ROIs_3D_gm_test.append(ROIs_3D_gm[k][idx_test, ...])
ROIs_3D_wm_train.append(ROIs_3D_wm[k][idx_train, ...])
ROIs_3D_wm_test.append(ROIs_3D_wm[k][idx_test, ...])
# SPLIT TRAINING INTO TRAINING/VALIDATION
len_yt = y.shape[0]
if use_validation:
train_size = floor(tra_val_split * len_yt)
else:
train_size = len_yt - 1
y_tra = y[:train_size]
features_tra = features_train[:train_size, ...]
y_val = y[train_size:]
features_val = features_train[train_size:, ...]
train_ROIs_3D_gm = []
val_ROIs_3D_gm = []
train_ROIs_3D_wm = []
val_ROIs_3D_wm = []
for k in range(self.len_ROI):
train_ROIs_3D_gm.append(ROIs_3D_gm_train[k][:train_size, ...])
val_ROIs_3D_gm.append(ROIs_3D_gm_train[k][train_size:, ...])
train_ROIs_3D_wm.append(ROIs_3D_wm_train[k][:train_size, ...])
val_ROIs_3D_wm.append(ROIs_3D_wm_train[k][train_size:, ...])
X_tra = train_ROIs_3D_gm + train_ROIs_3D_wm + [features_tra]
X_val = val_ROIs_3D_gm + val_ROIs_3D_wm + [features_val]
X_test = ROIs_3D_gm_test + ROIs_3D_wm_test + [features_test]
self.y_test = y_test
self.X_test = X_test
self.y_tra = y_tra
self.X_tra = X_tra
self.y_val = y_val
self.X_val = X_val