The class constructor lets us load the data and candidates, and then build a vocabulary and subsequently initialize our TensorFlow session and memory network object:

class ChatBotWrapper(object):
    def __init__(self, train_data, test_data, val_data, 
                 candidates, candidates_to_idx,
                 memory_size, batch_size, learning_rate, 
                 evaluation_interval, hops,
                 epochs, embedding_size):
        self.memory_size = memory_size
        self.batch_size = batch_size
        self.evaluation_interval = evaluation_interval
        self.epochs = epochs

        self.candidates = candidates 
        self.candidates_to_idx = candidates_to_idx
        self.candidates_size = len(candidates)
        self.idx_to_candidates = dict((self.candidates_to_idx[key], key) 
                                      for key in self.candidates_to_idx)
        # Initialize data and build vocabulary
        self.train_data = train_data
        self.test_data = test_data
        self.val_data = val_data
        self.build_vocab(train_data + test_data + val_data, candidates)
        # Vectorize candidates
        self.candidates_vec = vectorize_candidates(
            candidates, self.word_idx, self.candidate_sentence_size)
        # Initialize optimizer
        optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate)
        # Initialize TensorFlow session and Memory Network model
        self.sess = tf.Session()
        self.model = MemoryNetwork(
                        self.sentence_size, self.vocab_size, 
                        self.candidates_size, self.candidates_vec, 
                        embedding_size, hops,
                        optimizer=optimizer, session=self.sess)