Function train

a09_DynamicMemoryNet/a8_dynamic_memory_network.py:315–354

()

#step1. run train function to train the model. it will save checkpoint
#step2. run predict function to make a prediction based on the model restore from the checkpoint.
def train():
    # below is a function test; if you use this for text classifiction, you need to tranform sentence to indices of vocabulary first. then feed data to the graph.
    num_classes = 15
    learning_rate = 0.001
    batch_size = 8
    decay_steps = 1000
    decay_rate = 0.9
    sequence_length = 10
    vocab_size = 10000
    embed_size = 100
    hidden_size = 100
    is_training = True
    story_length = 3
    dropout_keep_prob = 1
    model = DynamicMemoryNetwork(num_classes, learning_rate, batch_size, decay_steps, decay_rate, sequence_length,
                          story_length, vocab_size, embed_size, hidden_size, is_training,
                          multi_label_flag=False)
    ckpt_dir = 'checkpoint_dmn/dummy_test/'
    saver = tf.train.Saver()
    with tf.Session() as sess:
        sess.run(tf.global_variables_initializer())
        for i in range(1500):
            # input_x should be:[batch_size, num_sentences,self.sequence_length]
            story = np.random.randn(batch_size, story_length, sequence_length)
            story[story > 0] = 1
            story[story <= 0] = 0
            query = np.random.randn(batch_size, sequence_length)  # [batch_size, sequence_length]
            query[query > 0] = 1
            query[query <= 0] = 0
            answer_single = np.sum(query, axis=1) + np.round(0.1 * np.sum(np.sum(story, axis=1),
                                                                          axis=1))  # [batch_size].e.g. np.array([1, 0, 1, 1, 1, 2, 1, 1])
            loss, acc, predict, _ = sess.run(
                [model.loss_val, model.accuracy, model.predictions, model.train_op],
                feed_dict={model.query: query, model.story: story, model.answer_single: answer_single,
                           model.dropout_keep_prob: dropout_keep_prob})
            print(i, "query:", query, "=====================>")
            print(i, "loss:", loss, "acc:", acc, "label:", answer_single, "prediction:", predict)
            if i % 300 == 0:
                save_path = ckpt_dir + "model.ckpt"
                saver.save(sess, save_path, global_step=i * 300)

def predict():
    num_classes = 15