def train(self, session, data):
# data separate
self.left = data_split(X=data[0], train_rate=0.8)
self.right = data_split(X=data[1], train_rate=0.8)
self.label = data_split(X=data[2], train_rate=0.8)
print('[Step2 Data] Train samples:{0}, valid samples:{1}'.format(
len(self.left[0]), len(self.left[1])))
# train
for iters in range(self.max_iterations):
batch_x_left = get_next_batch(self.left[0], None, iters,
self.batch_size)
batch_x_right = get_next_batch(self.right[0], None, iters,
self.batch_size)
batch_y = get_next_batch(self.label[0], None, iters,
self.batch_size)
loss, _ = session.run([self.loss_func, self.optimizer],
feed_dict={
self.XL: batch_x_left,
self.XR: batch_x_right,
self.Y: batch_y
})
if iters % self.display_iters == 0:
print('[Step2 Train] RegNet Iters:{0}, loss:{1}'.format(
iters, loss))
if iters % self.test_iters == 0:
self.val_internal(session)
def val_internal(self, session):
max_batches = len(self.data[2]) // self.test_batch
rights = 0.0
for iters in range(max_batches):
batch_x, batch_y = get_next_batch(self.data[2], self.data[3], iters, self.test_batch)
sums = session.run(self.correct,feed_dict={self.X:batch_x,self.Y:batch_y})
rights += sums
print '[Step1 Valid] EyeNet val accuracy:{0}'.format(rights/len(self.data[2]))
def val_internal(self, session):
max_batches = len(self.left[1]) // self.test_batch
rights = 0.0
for iters in range(max_batches):
batch_x_left = get_next_batch(self.left[1], None, iters,
self.test_batch)
batch_x_right = get_next_batch(self.right[1], None, iters,
self.test_batch)
batch_y = get_next_batch(self.label[1], None, iters,
self.test_batch)
sums = session.run(self.correct,
feed_dict={
self.XL: batch_x_left,
self.XR: batch_x_right,
self.Y: batch_y
})
rights += sums
print('[Step2 Valid] RegNet val accuracy:{0}'.format(
rights / len(self.left[1])))
def train(self, session, data):
# prepare data
self.data = data_split(X=data[0],Y=data[1],train_rate=0.8)
print '[Step1 Data] Train samples:{0}, valid samples:{1}'.format(len(self.data[0]), len(self.data[3]))
# train step by step
for iters in range(self.max_iterations):
batch_x, batch_y = get_next_batch(self.data[0], self.data[1], iters, self.batch_size)
loss, _ = session.run([self.loss_func, self.optimizer],feed_dict={self.X:batch_x, self.Y:batch_y})
if iters % self.display_iters == 0:
print '[Step1 Train] EyeNet Iters:{0}, loss:{1}'.format(iters,loss)
if iters % self.test_iters == 0:
self.val_internal(session)