if __name__ == '__main__':
# Load and compile model
K.clear_session()
model = Xception(include_top=True, weights=None)
model.compile(optimizer='Adadelta',
loss='categorical_crossentropy',
metrics=['accuracy'])
batch_count = 0
try:
for i in range(0, 30000):
gc.collect()
print('----------- On Epoch: ' + str(i) + ' ----------')
for x_train, y_train, x_test, y_test in load_batches():
# Model input requires numpy array
#K.clear_session()
x_train = np.array(x_train)
x_train_0 = x_train[:, 0]
x_train_1 = x_train[:, 1]
x_train_2 = x_train[:, 2]
'''
x_train1 = np.array([x_train])
x_train_0_5S = np.array(x_train_0_5S)
x_train1 = append_data(x_train1,x_train_0_5S)
x_train_2S = np.array(x_train_2S)
x_train1 = append_data(x_train1,x_train_2S)
x_train_5S = np.array(x_train_5S)
x_train1 = append_data(x_train1,x_train_5S)
'''
print('We have loaded a previous model!!!!')
'''
txtfile = 'data_rgb.txt'
print("Starting Training...")
batch_count = 0
#try:
for i in range(1, 10):
count = 0
print('----------- On Epoch: ' + str(i) + ' ----------')
f = open(txtfile, 'w+')
data_dump_no = 1
f.write(str(data_dump_no))
f.close()
for x_train, y_train, x_test, y_test, samples_per_batch, batch_count in load_batches(
epoch=i, txtfile=txtfile, samples_per_batch=1000):
# Model input requires numpy array
x_train = np.array(x_train)
x_test = np.array(x_test)
# Classification to one-hot vector
y_train = np.array(y_train)
y_test = np.array(y_test)
# Fit model to batch
'''
def fit(self, X_inputs, Y_targets, n_epoch=10, validation_set=None,
show_metric=False, batch_size=None, shuffle=None,
snapshot_epoch=True, snapshot_step=None, excl_trainops=None,
validation_batch_size=None, run_id=None, callbacks=[]):
'''
model.fit({'input': x_train}, {'targets': y_train},
n_epoch=1,
#
# for j, layer in enumerate(model.layers):
# print(j, layer.name)
print("Starting Training...")
batch_count = 0
# if True:
try:
for i in range(1, 6): # 6 epochs done!
count = 0
print('----------- On Epoch: ' + str(i) + ' ----------')
f = open(txtfile, 'w+')
data_dump_no = 6
f.write(str(data_dump_no))
f.close()
for x_train, y_train, x_test, y_test, batch_size, batch_count in load_batches(
epoch=i, samples_per_batch=4000, txtfile=txtfile):
# Model input requires numpy array
x_train = np.array(x_train)
x_test = np.array(x_test)
# Classification to one-hot vector
y_train = np.array(y_train)
y_test = np.array(y_test)
tensorboard = TensorBoard(
log_dir="logs/{}-{}-added_data-turn-only-v9.2".format(NAME, i))
# Fit model to batch
print('Training......')
# train the model on the new data for a few epochs
model.fit(x_train,
y_train,
MODEL_NAME = NAME+'.model'
PREV_MODEL = NAME+'.model'
LOAD_MODEL = True
model = googlenet(WIDTH, HEIGHT, 3, LR, output=35, model_name=MODEL_NAME)
if LOAD_MODEL:
model.load(PREV_MODEL)
print('We have loaded a previous model!!!!')
count = 0
for epoch in range(EPOCHS):
try:
for X, Y, test_x, test_y, batch_size, batch_count in load_batches(epoch=epoch,txtfile=txtfile):
model.fit({'input': X}, {'targets': Y}, n_epoch=6,
validation_set=({'input': test_x}, {'targets': test_y}), snapshot_step=2500,
show_metric=True, run_id=MODEL_NAME)
print('SAVING MODEL!')
model.save(MODEL_NAME)
print("Saved model to disk")
if batch_count % 2 == 0:
print('SAVING MODEL!')
model.save(MODEL_NAME)
print("Saved model to disk")
if (batch_count + 1) % 2 == 0:
print('SAVING MODEL!')
model.save(MODEL_NAME)
print("Saved model to disk")
