[技术分享]使用N-Gram词向量模型句子的下一个单词

本文参考《深度学习工程师认证初级教程》中5.3.3节词向量案例，利用N-gram模型实现在PTB数据集上的单词预测。

词向量是自然语言处理中常见的一个操作，是搜索引擎、广告系统、推荐系统等互联网服务背后常见的基础技术。One-hot vector是一种最简单的词向量，但是用处有限。比如，在互联网广告系统里，如果用户输入的query是“母亲节”，而有一个广告的关键词是“康乃馨”。虽然按照常理，我们知道这两个词之间是有联系的——母亲节通常应该送给母亲一束康乃馨，但是这两个词对应的one-hot vectors之间的距离度量，无论是欧氏距离还是余弦相似度(cosine similarity)，由于其向量正交，都认为这两个词毫无相关性。 在神经网络中，通过学习语义信息可以得到词向量，常用的模型包括：N-gram模型，CBOW模型和Skip-gram模型。

一、数据集

本文中使用的语料数据集是Penn Treebank(PTB)经Tomas Mikolov预处理过的版本数据集。PTB数据集较小，训练速度快，广泛应用于自然语言处理中。在PTB数据集中训练数据2068句，验证数据3370句，测试数据3761句。Paddle中提供了对应PTB数据集的python包paddle.dataset.imikolov，自动做数据的下载与预处理，方便大家使用。

二、配置说明

1、超参数配置

# 定义参数

EMBED_SIZE = 32      # embedding维度
HIDDEN_SIZE = 256    # 隐层大小
N = 5                # ngram大小，这里固定取5
BATCH_SIZE = 100     # batch大小
PASS_NUM = 100       # 训练轮数

use_cuda = True  # 如果用GPU训练，则设置为True

2、加载预览数据

# 读取数据

word_dict = paddle.dataset.imikolovolov.build_dict() #从paddel自带的数据集构建字典
dict_size = len(word_dict)

train_reader = paddle.batch(
    paddle.dataset.imikolov.train(word_dict, N), BATCH_SIZE)
test_reader = paddle.batch(
    paddle.dataset.imikolov.test(word_dict, N), BATCH_SIZE)

for data in train_reader():
    print(len(data))
    print(data[91])
    break

输出为：

100 #一批数据大小为100
(6, 15, 47, 56, 4) #第一条数据（5个单词组成，每个数字代表字典中的某个单词对应的编号，如6是‘to’,15是‘said’,47是‘were’，56是‘than’，4是‘N’）

3、网络配置

N-Gram模型的网络结构为：

#定义N-Gram网络结构

def inference_program(words, is_sparse):

    embed_first = fluid.embedding(
        input=words[0],
        size=[dict_size, EMBED_SIZE],
        dtype='float32',
        is_sparse=is_sparse,
        param_attr='shared_w')
    embed_second = fluid.embedding(
        input=words[1],
        size=[dict_size, EMBED_SIZE],
        dtype='float32',
        is_sparse=is_sparse,
        param_attr='shared_w')
    embed_third = fluid.embedding(
        input=words[2],
        size=[dict_size, EMBED_SIZE],
        dtype='float32',
        is_sparse=is_sparse,
        param_attr='shared_w')
    embed_fourth = fluid.embedding(
        input=words[3],
        size=[dict_size, EMBED_SIZE],
        dtype='float32',
        is_sparse=is_sparse,
        param_attr='shared_w')

    concat_embed = fluid.layers.concat(
        input=[embed_first, embed_second, embed_third, embed_fourth], axis=1)
    hidden1 = fluid.layers.fc(input=concat_embed,
                              size=HIDDEN_SIZE,
                              act='sigmoid')
    predict_word = fluid.layers.fc(input=hidden1, size=dict_size, act='softmax')
    return predict_word

三、训练模型

可以使用paddle自带的语料训练N-Gram模型，也可以使用其他语料，如使用莎士比亚文集参考这里

# 训练方法

def train_program(predict_word):
    # 'next_word'的定义必须要在inference_program的声明之后，
    # 否则train program输入数据的顺序就变成了[next_word, firstw, secondw,
    # thirdw, fourthw], 这是不正确的.
    next_word = fluid.data(name='nextw', shape=[None, 1], dtype='int64')
    cost = fluid.layers.cross_entropy(input=predict_word, label=next_word)
    avg_cost = fluid.layers.mean(cost)
    return avg_cost

def optimizer_func():
    return fluid.optimizer.AdagradOptimizer(
        learning_rate=3e-3,
        regularization=fluid.regularizer.L2DecayRegularizer(8e-4))

# 训练模型

def train(if_use_cuda, params_dirname, is_sparse=True):
    place = fluid.CUDAPlace(0) if if_use_cuda else fluid.CPUPlace()

    train_reader = paddle.batch(
        paddle.dataset.imikolov.train(word_dict, N), BATCH_SIZE)
    test_reader = paddle.batch(
        paddle.dataset.imikolov.test(word_dict, N), BATCH_SIZE)

    first_word = fluid.data(name='firstw', shape=[None, 1], dtype='int64')
    second_word = fluid.data(name='secondw', shape=[None, 1], dtype='int64')
    third_word = fluid.data(name='thirdw', shape=[None, 1], dtype='int64')
    forth_word = fluid.data(name='fourthw', shape=[None, 1], dtype='int64')
    next_word = fluid.data(name='nextw', shape=[None, 1], dtype='int64')

    word_list = [first_word, second_word, third_word, forth_word, next_word]
    feed_order = ['firstw', 'secondw', 'thirdw', 'fourthw', 'nextw']

    main_program = fluid.default_main_program()
    star_program = fluid.default_startup_program()

    predict_word = inference_program(word_list, is_sparse)
    avg_cost = train_program(predict_word)
    test_program = main_program.clone(for_test=True)

    optimizer = optimizer_func()
    optimizer.minimize(avg_cost)

    exe = fluid.Executor(place)

    def train_test(program, reader):
        count = 0
        feed_var_list = [
            program.global_block().var(var_name) for var_name in feed_order
        ]
        feeder_test = fluid.DataFeeder(feed_list=feed_var_list, place=place)
        test_exe = fluid.Executor(place)
        accumulated = len([avg_cost]) * [0]
        for test_data in reader():
            avg_cost_np = test_exe.run(
                program=program,
                feed=feeder_test.feed(test_data),
                fetch_list=[avg_cost])
            accumulated = [
                x[0] + x[1][0] for x in zip(accumulated, avg_cost_np)
            ]
                   count += 1
        return [x / count for x in accumulated]

    def train_loop():
        step = 0
        feed_var_list_loop = [
            main_program.global_block().var(var_name) for var_name in feed_order
        ]
        feeder = fluid.DataFeeder(feed_list=feed_var_list_loop, place=place)
        exe.run(star_program)
        for pass_id in range(PASS_NUM):
            for data in train_reader():
                avg_cost_np = exe.run(
                    main_program, feed=feeder.feed(data), fetch_list=[avg_cost])

                if step % 10 == 0:
                    outs = train_test(test_program, test_reader)

                    print("Step %d: Average Cost %f" % (step, outs[0]))

                    # 整个训练过程要花费几个小时，如果平均损失低于5.8，
                    # 我们就认为模型已经达到很好的效果可以停止训练了。
                    # 注意5.8是一个相对较高的值，为了获取更好的模型，可以将
                    # 这里的阈值设为3.5，但训练时间也会更长。
                    if outs[0] < 5.8:
                        if params_dirname is not None:
                            fluid.io.save_inference_model(params_dirname, [
                                'firstw', 'secondw', 'thirdw', 'fourthw'
                            ], [predict_word], exe)
                        return
                step += 1
                if math.isnan(float(avg_cost_np[0])):
                    sys.exit("got NaN loss, training failed.")

        raise AssertionError("Cost is too large {0:2.2}".format(avg_cost_np[0]))

    train_loop()

四、模型预测

# 模型推理

def infer(use_cuda, params_dirname=None):
    place = fluid.CUDAPlace(0) if use_cuda else fluid.CPUPlace()

    exe = fluid.Executor(place)

    inference_scope = fluid.core.Scope()
    with fluid.scope_guard(inference_scope):
        # 使用fluid.io.load_inference_model获取inference program，
        # feed变量的名称feed_target_names和从scope中fetch的对象fetch_targets
        [inferencer, feed_target_names,
         fetch_targets] = fluid.io.load_inference_model(params_dirname, exe)

        # 设置输入，用四个LoDTensor来表示4个词语。这里每个词都是一个id，
        # 用来查询embedding表获取对应的词向量，因此其形状大小是[1]。
        # recursive_sequence_lengths设置的是基于长度的LoD，因此都应该设为[[1]]
        # 注意recursive_sequence_lengths是列表的列表
        data1 = numpy.asarray([[211]], dtype=numpy.int64)  # 'among'
        data2 = numpy.asarray([[6]], dtype=numpy.int64)  # 'a'
        data3 = numpy.asarray([[96]], dtype=numpy.int64)  # 'group'
        data4 = numpy.asarray([[4]], dtype=numpy.int64)  # 'of'
        lod = numpy.asarray([[1]], dtype=numpy.int64)

        first_word = fluid.create_lod_tensor(data1, lod, place)
        second_word = fluid.create_lod_tensor(data2, lod, place)
        third_word = fluid.create_lod_tensor(data3, lod, place)
        fourth_word = fluid.create_lod_tensor(data4, lod, place)

        assert feed_target_names[0] == 'firstw'
        assert feed_target_names[1] == 'secondw'
        assert feed_target_names[2] == 'thirdw'
        assert feed_target_names[3] == 'fourthw'

        # 构造feed词典 {feed_target_name: feed_target_data}
        # 预测结果包含在results之中
        results = exe.run(
            inferencer,
            feed={
                feed_target_names[0]: first_word,
                feed_target_names[1]: second_word,
                feed_target_names[2]: third_word,
                feed_target_names[3]: fourth_word
            },
            fetch_list=fetch_targets,
            return_numpy=False)

        print(numpy.array(results[0]))
        most_possible_word_index = numpy.argmax(results[0])
        print(most_possible_word_index)
        print([
            key for key, value in six.iteritems(word_dict)
            if value == most_possible_word_index
        ][0])

主程序为：

# 主程序

def main(use_cuda, is_sparse):
    if use_cuda and not fluid.core.is_compiled_with_cuda():
        return

    params_dirname = "word2vec.inference.model"
    paddle.enable_static()
    train(
        if_use_cuda=use_cuda,
        params_dirname=params_dirname,
        is_sparse=is_sparse)

    infer(use_cuda=use_cuda, params_dirname=params_dirname)


main(use_cuda=use_cuda, is_sparse=True)

得到among a group of下一个单词的预测结果为the：

Step 0: Average Cost 7.399707
Step 10: Average Cost 6.203541
Step 20: Average Cost 5.858641
Step 30: Average Cost 5.721037
[[0.04144997 0.03653922 0.04100619 ... 0.00019988 0.00010753 0.0284689 ]]
0
b'the'

该案例可以在百度AI Studio中查看notebook版本。