How can I reuse this tensorflow layer ?

I'm writing a model in Tensorflow that I was used to use with PyTorch. Some mechanisms being very much different, I'm stuck at some points.
In particular:

dense = tf.layers.dense
adam = tf.train.AdamOptimizer
nb_joints= 3
code_size = 8

joints_info = tf.placeholder(tf.float32, shape = [None,nb_joints], name = 'joints_state')
target_info = tf.placeholder(tf.float32, shape = [None,2], name = 'target_pos')

next_joint_info = tf.placeholder(tf.float32, shape = [None,nb_joints], name = 'next_joints_state')

with tf.variable_scope('Encoder'):

e1 = dense(joints_info, 32, activation = tf.nn.relu, name ='encoding_1')
code = dense(e1, code_size, activation = tf.nn.relu, name ='code')

d1 = dense(code, code_size, activation = tf.nn.relu, name ='decoding_1')
recon = dense(d1, code_size, activation = tf.nn.relu, name ='reconstructed')

with tf.variable_scope('EncoderLoss'):

encoder_loss = tf.squared_difference(joints_info, recon)
train_encoder = adam(3e-4).minimize(encoder_loss)

with tf.variable_scope('Task'):

t1 = dense(code, 32, activation = tf.nn.relu, name ='task_code')

t1_targ = dense(target_info, 32, activation = tf.nn.relu, name ='task_target')

task_joint = tf.concat([t1,t1_targ],1, name ='States_concatenation')

t2 = dense(task_joint, 128, activation = tf.nn.relu, name = 'task_joint_transformation')

task_prediction = dense(t2, code_size, activation = None, name = 'task_prediction')

with tf.variable_scope('TaskLoss'):

task_real = here, I want to call the CODE operation from the encoder but using next_joint_info placeholder
task_loss = tf.squared_difference(task_prediction, task_real)


Could someone point me in the right direction ? I have no idea on how to proceed here.

Thanks a lot !

TaskLoss

Encoder

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.