1. Attention

  H是一个矩阵，它是由LSTM层产生的多个向量$[h_1,h_2,\dots,h_T]$[h1​,h2​,…,hT​]组成的。其中T是句子的长度。句子的表示$r$r是输出向量的权重之和。
$M = tan(H)$M=tan(H) $\alpha =softmax(W^TM)$α=softmax(WTM) $r=H\alpha^T$r=HαT

output_h = tf.add(output_forward, output_backward) # total_num, time_steps,gru_size
output_h的维度为[total_num, time_steps,gru_size]，其中total_num指的是batch_size，time_steps指的是length of sentence。

  attention_w = tf.get_variable(‘attention_omega’, [gru_size, 1]) #attention_w本质上是每个GRU单元不同的权重，也就是对每个单词赋予不同的权重。

• t1 = tf.tanh(output_h) #[total_num, time_steps,gru_size]
• t2 = t1.reshape(t1, [total_numtime_steps, gru_size]) #total_numtime_steps, gru_size
• t3 = tf.matmul(t2, attention_w)#total_num*time_steps,1
• t4 = tf.reshape(t3, [total_num, time_steps]) #total_num * time_steps
• t5 = tf.nn.softmax(t4) # softmax是针对每个词来说的, total_num * time_steps
• t6 = tf.reshape(t5, [total_num, 1, time_steps])
• t7 = tf.matmul(t6, output_h) # total_num * 1 * gru_size，其中三维张量的乘法，第一维不变（第一维比较相同），后两维进行二维矩阵的乘法。
• r = tf.reshape(t7, [total_num, gru_size])
• result = tf.tanh( r )

另外的实现代码如下所示：

def attention(inputs):
    # Trainable parameters
    # B*T*H B*T*1
    hidden_size = inputs.shape[2].value
    u_omega = tf.get_variable("u_omega", [hidden_size], initializer=tf.keras.initializers.glorot_normal())

    with tf.name_scope('v'):
        v = tf.tanh(inputs)

    # For each of the timestamps its vector of size A from `v` is reduced with `u` vector
    vu = tf.tensordot(v, u_omega, axes=1, name='vu')  # (B,T) shape
    alphas = tf.nn.softmax(vu, name='alphas')  # (B,T) shape

    # Output of (Bi-)RNN is reduced with attention vector; the result has (B,D) shape
    output = tf.reduce_sum(inputs * tf.expand_dims(alphas, -1), 1)

    # Final output with tanh
    output = tf.tanh(output)

    return output, alphas

2. BLSTM

class AttLSTM:
    def __init__(self, sequence_length, num_classes, vocab_size, embedding_size,
                 hidden_size, l2_reg_lambda=0.0):
        # Placeholders for input, output and dropout
        self.input_text = tf.placeholder(tf.int32, shape=[None, sequence_length], name='input_text')
        self.input_y = tf.placeholder(tf.float32, shape=[None, num_classes], name='input_y')
        self.emb_dropout_keep_prob = tf.placeholder(tf.float32, name='emb_dropout_keep_prob')
        self.rnn_dropout_keep_prob = tf.placeholder(tf.float32, name='rnn_dropout_keep_prob')
        self.dropout_keep_prob = tf.placeholder(tf.float32, name='dropout_keep_prob')

        initializer = tf.keras.initializers.glorot_normal

        # Word Embedding Layer
        with tf.device('/cpu:0'), tf.variable_scope("word-embeddings"):
            self.W_text = tf.Variable(tf.random_uniform([vocab_size, embedding_size], -0.25, 0.25), name="W_text")
            self.embedded_chars = tf.nn.embedding_lookup(self.W_text, self.input_text)

        # Dropout for Word Embedding
        with tf.variable_scope('dropout-embeddings'):
            self.embedded_chars = tf.nn.dropout(self.embedded_chars, self.emb_dropout_keep_prob)

        # Bidirectional LSTM
        with tf.variable_scope("bi-lstm"):
            _fw_cell = tf.nn.rnn_cell.LSTMCell(hidden_size, initializer=initializer())
            fw_cell = tf.nn.rnn_cell.DropoutWrapper(_fw_cell, self.rnn_dropout_keep_prob)
            _bw_cell = tf.nn.rnn_cell.LSTMCell(hidden_size, initializer=initializer())
            bw_cell = tf.nn.rnn_cell.DropoutWrapper(_bw_cell, self.rnn_dropout_keep_prob)
            self.rnn_outputs, _ = tf.nn.bidirectional_dynamic_rnn(cell_fw=fw_cell,
                                                                  cell_bw=bw_cell,
                                                                  inputs=self.embedded_chars,
                                                                  sequence_length=self._length(self.input_text),
                                                                  dtype=tf.float32)
            self.rnn_outputs = tf.add(self.rnn_outputs[0], self.rnn_outputs[1])

        # Attention
        with tf.variable_scope('attention'):
            self.attn, self.alphas = attention(self.rnn_outputs)

        # Dropout
        with tf.variable_scope('dropout'):
            self.h_drop = tf.nn.dropout(self.attn, self.dropout_keep_prob)

        # Fully connected layer
        with tf.variable_scope('output'):
            self.logits = tf.layers.dense(self.h_drop, num_classes, kernel_initializer=initializer())
            self.predictions = tf.argmax(self.logits, 1, name="predictions")

        # Calculate mean cross-entropy loss
        with tf.variable_scope("loss"):
            losses = tf.nn.softmax_cross_entropy_with_logits_v2(logits=self.logits, labels=self.input_y)
            self.l2 = tf.add_n([tf.nn.l2_loss(v) for v in tf.trainable_variables()])
            self.loss = tf.reduce_mean(losses) + l2_reg_lambda * self.l2

        # Accuracy
        with tf.variable_scope("accuracy"):
            correct_predictions = tf.equal(self.predictions, tf.argmax(self.input_y, 1))
            self.accuracy = tf.reduce_mean(tf.cast(correct_predictions, tf.float32), name="accuracy")

    # Length of the sequence data
    @staticmethod
    def _length(seq):
        relevant = tf.sign(tf.abs(seq))
        length = tf.reduce_sum(relevant, reduction_indices=1)
        length = tf.cast(length, tf.int32)
        return length

3. 训练模型

def train():
    with tf.device('/cpu:0'):
        x_text, y = data_helpers.load_data_and_labels(FLAGS.train_path)

    # Build vocabulary
    # Example: x_text[3] = "A misty <e1>ridge</e1> uprises from the <e2>surge</e2>."
    # ['a misty ridge uprises from the surge <UNK> <UNK> ... <UNK>']
    # =>
    # [27 39 40 41 42  1 43  0  0 ... 0]
    # dimension = FLAGS.max_sentence_length
    vocab_processor = tf.contrib.learn.preprocessing.VocabularyProcessor(FLAGS.max_sentence_length)
    x = np.array(list(vocab_processor.fit_transform(x_text)))
    print("Text Vocabulary Size: {:d}".format(len(vocab_processor.vocabulary_)))
    print("x = {0}".format(x.shape))
    print("y = {0}".format(y.shape))
    print("")

    # Randomly shuffle data to split into train and test(dev)
    np.random.seed(10)
    shuffle_indices = np.random.permutation(np.arange(len(y)))
    x_shuffled = x[shuffle_indices]
    y_shuffled = y[shuffle_indices]

    # Split train/test set
    # TODO: This is very crude, should use cross-validation
    dev_sample_index = -1 * int(FLAGS.dev_sample_percentage * float(len(y)))
    x_train, x_dev = x_shuffled[:dev_sample_index], x_shuffled[dev_sample_index:]
    y_train, y_dev = y_shuffled[:dev_sample_index], y_shuffled[dev_sample_index:]
    print("Train/Dev split: {:d}/{:d}\n".format(len(y_train), len(y_dev)))

    with tf.Graph().as_default():
        session_conf = tf.ConfigProto(
            allow_soft_placement=FLAGS.allow_soft_placement,
            log_device_placement=FLAGS.log_device_placement)
        session_conf.gpu_options.allow_growth = FLAGS.gpu_allow_growth
        sess = tf.Session(config=session_conf)
        with sess.as_default():
            model = AttLSTM(
                sequence_length=x_train.shape[1],
                num_classes=y_train.shape[1],
                vocab_size=len(vocab_processor.vocabulary_),
                embedding_size=FLAGS.embedding_dim,
                hidden_size=FLAGS.hidden_size,
                l2_reg_lambda=FLAGS.l2_reg_lambda)

            # Define Training procedure
            global_step = tf.Variable(0, name="global_step", trainable=False)
            optimizer = tf.train.AdadeltaOptimizer(FLAGS.learning_rate, FLAGS.decay_rate, 1e-6)
            gvs = optimizer.compute_gradients(model.loss)
            capped_gvs = [(tf.clip_by_value(grad, -1.0, 1.0), var) for grad, var in gvs]
            train_op = optimizer.apply_gradients(capped_gvs, global_step=global_step)

            # Output directory for models and summaries
            timestamp = str(int(time.time()))
            out_dir = os.path.abspath(os.path.join(os.path.curdir, "runs", timestamp))
            print("Writing to {}\n".format(out_dir))

            # Summaries for loss and accuracy
            loss_summary = tf.summary.scalar("loss", model.loss)
            acc_summary = tf.summary.scalar("accuracy", model.accuracy)

            # Train Summaries
            train_summary_op = tf.summary.merge([loss_summary, acc_summary])
            train_summary_dir = os.path.join(out_dir, "summaries", "train")
            train_summary_writer = tf.summary.FileWriter(train_summary_dir, sess.graph)

            # Dev summaries
            dev_summary_op = tf.summary.merge([loss_summary, acc_summary])
            dev_summary_dir = os.path.join(out_dir, "summaries", "dev")
            dev_summary_writer = tf.summary.FileWriter(dev_summary_dir, sess.graph)

            # Checkpoint directory. Tensorflow assumes this directory already exists so we need to create it
            checkpoint_dir = os.path.abspath(os.path.join(out_dir, "checkpoints"))
            checkpoint_prefix = os.path.join(checkpoint_dir, "model")
            if not os.path.exists(checkpoint_dir):
                os.makedirs(checkpoint_dir)
            saver = tf.train.Saver(tf.global_variables(), max_to_keep=FLAGS.num_checkpoints)

            # Write vocabulary
            vocab_processor.save(os.path.join(out_dir, "vocab"))

            # Initialize all variables
            sess.run(tf.global_variables_initializer())

            # Pre-trained word2vec
            if FLAGS.embedding_path:
                pretrain_W = utils.load_glove(FLAGS.embedding_path, FLAGS.embedding_dim, vocab_processor)
                sess.run(model.W_text.assign(pretrain_W))
                print("Success to load pre-trained word2vec model!\n")

            # Generate batches
            batches = data_helpers.batch_iter(list(zip(x_train, y_train)), FLAGS.batch_size, FLAGS.num_epochs)
            # Training loop. For each batch...
            best_f1 = 0.0  # For save checkpoint(model)
            for batch in batches:
                x_batch, y_batch = zip(*batch)
                # Train
                feed_dict = {
                    model.input_text: x_batch,
                    model.input_y: y_batch,
                    model.emb_dropout_keep_prob: FLAGS.emb_dropout_keep_prob,
                    model.rnn_dropout_keep_prob: FLAGS.rnn_dropout_keep_prob,
                    model.dropout_keep_prob: FLAGS.dropout_keep_prob
                }
                _, step, summaries, loss, accuracy = sess.run(
                    [train_op, global_step, train_summary_op, model.loss, model.accuracy], feed_dict)
                train_summary_writer.add_summary(summaries, step)

                # Training log display
                if step % FLAGS.display_every == 0:
                    time_str = datetime.datetime.now().isoformat()
                    print("{}: step {}, loss {:g}, acc {:g}".format(time_str, step, loss, accuracy))

                # Evaluation
                if step % FLAGS.evaluate_every == 0:
                    print("\nEvaluation:")
                    feed_dict = {
                        model.input_text: x_dev,
                        model.input_y: y_dev,
                        model.emb_dropout_keep_prob: 1.0,
                        model.rnn_dropout_keep_prob: 1.0,
                        model.dropout_keep_prob: 1.0
                    }
                    summaries, loss, accuracy, predictions = sess.run(
                        [dev_summary_op, model.loss, model.accuracy, model.predictions], feed_dict)
                    dev_summary_writer.add_summary(summaries, step)

                    time_str = datetime.datetime.now().isoformat()
                    f1 = f1_score(np.argmax(y_dev, axis=1), predictions, labels=np.array(range(1, 19)), average="macro")
                    print("{}: step {}, loss {:g}, acc {:g}".format(time_str, step, loss, accuracy))
                    print("[UNOFFICIAL] (2*9+1)-Way Macro-Average F1 Score (excluding Other): {:g}\n".format(f1))

                    # Model checkpoint
                    if best_f1 < f1:
                        best_f1 = f1
                        path = saver.save(sess, checkpoint_prefix + "-{:.3g}".format(best_f1), global_step=step)
                        print("Saved model checkpoint to {}\n".format(path))

4.验证模型

def test():
    with tf.device('/cpu:0'):
        x_text, y = data_helpers.load_data_and_labels(FLAGS.test_path)

    # Map data into vocabulary
    text_path = os.path.join(FLAGS.checkpoint_dir, "..", "vocab")
    text_vocab_processor = tf.contrib.learn.preprocessing.VocabularyProcessor.restore(text_path)
    x = np.array(list(text_vocab_processor.transform(x_text)))

    checkpoint_file = tf.train.latest_checkpoint(FLAGS.checkpoint_dir)

    graph = tf.Graph()
    with graph.as_default():
        session_conf = tf.ConfigProto(
            allow_soft_placement=FLAGS.allow_soft_placement,
            log_device_placement=FLAGS.log_device_placement)
        session_conf.gpu_options.allow_growth = FLAGS.gpu_allow_growth
        sess = tf.Session(config=session_conf)
        with sess.as_default():
            # Load the saved meta graph and restore variables
            saver = tf.train.import_meta_graph("{}.meta".format(checkpoint_file))
            saver.restore(sess, checkpoint_file)

            # Get the placeholders from the graph by name
            input_text = graph.get_operation_by_name("input_text").outputs[0]
            # input_y = graph.get_operation_by_name("input_y").outputs[0]
            emb_dropout_keep_prob = graph.get_operation_by_name("emb_dropout_keep_prob").outputs[0]
            rnn_dropout_keep_prob = graph.get_operation_by_name("rnn_dropout_keep_prob").outputs[0]
            dropout_keep_prob = graph.get_operation_by_name("dropout_keep_prob").outputs[0]

            # Tensors we want to evaluate
            predictions = graph.get_operation_by_name("output/predictions").outputs[0]

            # Generate batches for one epoch
            batches = data_helpers.batch_iter(list(x), FLAGS.batch_size, 1, shuffle=False)

            # Collect the predictions here
            preds = []
            for x_batch in batches:
                pred = sess.run(predictions, {input_text: x_batch,
                                              emb_dropout_keep_prob: 1.0,
                                              rnn_dropout_keep_prob: 1.0,
                                              dropout_keep_prob: 1.0})
                preds.append(pred)
            preds = np.concatenate(preds)
            truths = np.argmax(y, axis=1)

            prediction_path = os.path.join(FLAGS.checkpoint_dir, "..", "predictions.txt")
            truth_path = os.path.join(FLAGS.checkpoint_dir, "..", "ground_truths.txt")
            prediction_file = open(prediction_path, 'w')
            truth_file = open(truth_path, 'w')
            for i in range(len(preds)):
                prediction_file.write("{}\t{}\n".format(i, utils.label2class[preds[i]]))
                truth_file.write("{}\t{}\n".format(i, utils.label2class[truths[i]]))
            prediction_file.close()
            truth_file.close()

            perl_path = os.path.join(os.path.curdir,
                                     "SemEval2010_task8_all_data",
                                     "SemEval2010_task8_scorer-v1.2",
                                     "semeval2010_task8_scorer-v1.2.pl")
            process = subprocess.Popen(["perl", perl_path, prediction_path, truth_path], stdout=subprocess.PIPE)
            for line in str(process.communicate()[0].decode("utf-8")).split("\\n"):
                print(line)