def split_data(classes,k,i,the_names,path,spex,batch_size=1,class_on_char=0):
np.random.shuffle(the_names)
class_names = []
class_names_val = []
for c in classes:
names = [idx for idx in the_names if idx[class_on_char].lower() == c.lower()]
list_names = np.array_split(names,k)
class_names.extend(np.hstack(np.delete(list_names, i, 0)).transpose())
class_names_val.extend(list_names[i])
np.random.shuffle(class_names)
np.random.shuffle(class_names_val)
genVal = signalLoader(class_names_val,path,spex,batch_size=batch_size,class_on_char=class_on_char)
gen = signalLoader(class_names,path,spex,batch_size=batch_size,class_on_char=class_on_char)
return (gen,genVal,len(class_names),len(class_names_val),class_names_val)
if name[0] == 'B':
labels.append([0, 1, 0])
if name[0] == 'C':
labels.append([0, 0, 1])
k_folds = 8
i = 0
length = len(labels)
indices = range(0, length)
list_names = np.array_split(names, k_folds)
val_list_names = list_names[i]
list_names = np.hstack(np.delete(list_names, i, 0)).transpose()
list_labels = np.array_split(labels, k_folds)
val_list_labels = list_labels[i]
list_labels = np.vstack(np.delete(list_labels, i, 0))
inputVal = signalLoader(nchan, val_list_names, val_list_labels, path())
input = signalLoader(nchan, list_names, list_labels, path())
img_size = (L, nchan)
last_conv_layer_name = "pooling"
classifier_layer_names = ["flatten", "dense"]
# Make model
model = define_model(nchan, L, Fs)
model.load_weights(
"C:/Users/Oskar/Documents/GitHub/Exjobb/logs/model_check_points/20210204-103301/fold1\cp-0005.ckpt"
)
date = datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
checkpoint_path_fold = checkpoint_path + "/gradCAM" + date + "/cp-{epoch:04d}.ckpt"
cp_callback = tf.keras.callbacks.ModelCheckpoint(filepath=checkpoint_path_fold,
save_weights_only=True,
VAL_ACC = []
VAL_LOSS = []
k_folds = 5
for i in range(0, k_folds):
print("Fold number " + str(i + 1) + "!")
length = len(labels)
indices = range(0, length)
list_names = np.array_split(names, k_folds)
val_list_names = list_names[i]
list_names = np.hstack(np.delete(list_names, i, 0)).transpose()
list_labels = np.array_split(labels, k_folds)
val_list_labels = list_labels[i]
list_labels = np.hstack(np.delete(list_labels, i, 0)).transpose()
data_generatorVal = signalLoader(nchan, val_list_names, val_list_labels)
data_generator = signalLoader(nchan, list_names, list_labels)
data_generatorPred = signalLoader(nchan, prednames, predlabels)
#testplot = np.loadtxt(path() + "Asimulated_test_1",delimiter=',')
#Modell enligt struktur i Zhao19
model = tensorflow.keras.Sequential()
model.add(layers.InputLayer((L, nchan), batch_size=1))
#TF-lager
model.add(MorletConv([L, nchan], T))
#Spatial faltning?
model.add(
layers.Conv2D(filters=25, kernel_size=[1, nchan], activation='elu')
) #kernel_size specificerar spatial faltning enligt Zhao19
#Resten av nätverket
model.add(layers.Permute((3, 1, 2)))
labels.append([0,1])
k_folds= 5
date = datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
for i in range(0,k_folds-4):
print("Fold number " + str(i+1) + "!")
length = len(labels)
indices = range(0,length)
list_names = np.array_split(names,k_folds)
val_list_names = list_names[i]
list_names = np.hstack(np.delete(list_names, i, 0)).transpose()
list_labels = np.array_split(labels,k_folds)
val_list_labels = list_labels[i]
list_labels = np.vstack(np.delete(list_labels,i,0))
data_generatorVal = signalLoader(nchan,val_list_names,val_list_labels,path(),data_aug=False,doDownsampling=False)
data_generator = signalLoader(nchan,list_names,list_labels,path(),data_aug=True,doDownsampling=False)
print(list_names)
print(val_list_names)
tensorboard_callback = load_tensorboard(who,date,i)
model = define_model(nchan,L,Fs)
#es = EarlyStopping(monitor='val_loss', mode='min', verbose=1, patience=2)
history = model.fit(
data_generator,
validation_data=(data_generatorVal),
steps_per_epoch=len(list_labels),
validation_steps=len(val_list_labels),
epochs=50,
callbacks=[tensorboard_callback]),
model.summary()
if name[0] == 'C':
labels.append([0, 0, 1])
date = datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
k_folds = 7
for i in range(0, k_folds):
print("Fold number " + str(i + 1) + "!")
length = len(labels)
indices = range(0, length)
list_names = np.array_split(names, k_folds)
val_list_names = list_names[i]
list_names = np.hstack(np.delete(list_names, i, 0)).transpose()
list_labels = np.array_split(labels, k_folds)
val_list_labels = list_labels[i]
list_labels = np.vstack(np.delete(list_labels, i, 0))
data_generatorVal = signalLoader(nchan, val_list_names, val_list_labels,
path())
data_generator = signalLoader(nchan, list_names, list_labels, path())
tensorboard_callback = load_tensorboard(who, date, i)
model = define_model_R(nchan, L, Fs)
#es = EarlyStopping(monitor='val_loss', mode='min', verbose=1, patience=2)
history = model.fit(data_generator,
validation_data=(data_generatorVal),
steps_per_epoch=len(list_labels),
validation_steps=len(val_list_labels),
epochs=epochs,
callbacks=[tensorboard_callback]),
model.summary()
print("Fold number " + str(i + 1) + "!")
length = len(labels)
indices = range(0, length)
list_names = np.array_split(names, k_folds)
val_list_names = list_names[i]
list_names = np.hstack(np.delete(list_names, i, 0)).transpose()
list_labels = np.array_split(labels, k_folds)
val_list_labels = list_labels[i]
list_labels = np.vstack(np.delete(list_labels, i, 0))
# if data_aug:
# val_list_labels = np.repeat(val_list_labels,2,axis=0)
# val_list_names = np.repeat(val_list_names,2)
# list_labels = np.repeat(list_labels,2,axis=0)
# list_names = np.repeat(list_names,2)
data_generatorTest = signalLoader(nchan, val_list_names, val_list_labels,
pathPred())
data_generator = signalLoader(nchan,
list_names,
list_labels,
path(),
data_aug=data_aug)
tensorboard_callback = load_tensorboard(who, date, i)
#es = EarlyStopping(monitor='val_loss', mode='min', verbose=1, patience=2)
checkpoint_path_fold = checkpoint_path + date + "/fold" + str(
i + 1) + "/cp-{epoch:04d}.ckpt"
check_point_dir = os.path.dirname(checkpoint_path_fold)
cp_callback = tensorflow.keras.callbacks.ModelCheckpoint(
filepath=checkpoint_path_fold, save_weights_only=True, verbose=1)
model = define_model(nchan, L, Fs)