using shared layers

class Scscore(TensorGraph):

  tg = dc.models.TensorGraph(

      tensorboard=True, model_dir='/tmp/scscore', use_queue=True)

  Input1 = InputFifoQueue()

  reactant_features = Feature()

  product_features = Feature()

  hidden_1 = Dense(out_channels=300, in_layer=Input1, activation_fn=tf.nn.relu)

  hidden_2 = Dense(

      out_channels=300, in_layer=hidden_1, activation_fn=tf.nn.relu)

  hidden_3 = Dense(

      out_channels=300, in_layer=hidden_2, activation_fn=tf.nn.relu)

  hidden_4 = Dense(

      out_channels=300, in_layer=hidden_3, activation_fn=tf.nn.relu)

  hidden_5 = Dense(

      out_channels=300, in_layer=hidden_4, activation_fn=tf.nn.relu)

  dense_reactant_1 = Dense(

      out_channels=300, in_layers=[reactant_features], activation_fn=tf.nn.relu)

  dense_product_1 = dense_reactant_1.shared(in_layers=[product_features])

  dense_reactant_2 = Dense(

      out_channels=300, in_layers=[dense_reactant_1], activation_fn=tf.nn.relu)

  dense_product_2 = dense_reactant_2.shared(in_layers=[dense_product_1])

  dense_reactant_3 = Dense(

      out_channels=300, in_layers=[dense_reactant_2], activation_fn=tf.nn.relu)

  dense_product_3 = dense_reactant_3.shared(in_layers=[dense_product_2])

  dense_reactant_4 = Dense(

      out_channels=300, in_layers=[dense_reactant_3], activation_fn=tf.nn.relu)

  dense_product_4 = dense_reactant_4.shared(in_layers=[dense_product_3])

  dense_reactant_5 = Dense(

      out_channels=300, in_layers=[dense_reactant_4], activation_fn=tf.nn.relu)

  dense_product_5 = dense_reactant_5.shared(in_layers=[dense_product_4])

  label = Label(shape=(300,))

  output_dense = Add(in_layers=[dense_reactant_5, dense_product_5])

  loss = Hingeloss(in_layers=[label, hidden_5])

  tg.set_loss(loss)

  output = Sigmoid(in_layers=[hidden_5])

  output = Sigmoid(in_layers=[output_dense])

  tg.add_output(output)

  label = Label(shape=(None, 1))

  loss = Hingeloss(in_layers=[label, output])

  tg.set_loss(loss)