def build_branch(self, name, W_emb, bias):
    indices = tf.placeholder(tf.int64, shape=[None, 2], name=name + '__indices')
    values = tf.placeholder(tf.float32, shape=[None], name=name + '__values')
    dense_shape = tf.placeholder(tf.int64, shape=[2], name=name + '__dense_shape')

    sparse_input = tf.SparseTensor(indices, values, dense_shape)

    embedding = tf.nn.relu(tf.sparse_tensor_dense_matmul(sparse_input, W_emb) + bias)

    encoder = embedding
    for layer in self.layers:
        encoder = layer(encoder)

    return [indices, values, dense_shape], encoder