文本分类(Text classification) ============================= 文本分类任务是将一句话或一段话划分到某个具体的类别。比如垃圾邮件识别，文本情绪分类等。 .. code-block:: text 1, 商务大床房，房间很大，床有2M宽，整体感觉经济实惠不错! 其中开头的1是只这条评论的标签，表示是正面的情绪。我们将使用到的数据可以通过 `此链接 `_ 下载并解压，当然也可以通过fastNLP自动下载该数据。 数据中的内容如下图所示。接下来，我们将用fastNLP在这个数据上训练一个分类网络。 .. figure:: ./cn_cls_example.png :alt: jupyter jupyter 步骤 ---- 一共有以下的几个步骤: 1. `读取数据 <#id4>`_ 2. `预处理数据 <#id5>`_ 3. `选择预训练词向量 <#id6>`_ 4. `创建模型 <#id7>`_ 5. `训练模型 <#id8>`_ (1) 读取数据 ~~~~~~~~~~~~~~~~~~~~ fastNLP提供多种数据的自动下载与自动加载功能，对于这里我们要用到的数据，我们可以用 :class:`~fastNLP.io.Loader` 自动下载并加载该数据。 更多有关Loader的使用可以参考 :mod:`~fastNLP.io.loader` .. code-block:: python from fastNLP.io import ChnSentiCorpLoader loader = ChnSentiCorpLoader() # 初始化一个中文情感分类的loader data_dir = loader.download() # 这一行代码将自动下载数据到默认的缓存地址, 并将该地址返回 data_bundle = loader.load(data_dir) # 这一行代码将从{data_dir}处读取数据至DataBundle DataBundle的相关介绍，可以参考 :class:`~fastNLP.io.DataBundle` 。我们可以打印该data\_bundle的基本信息。 .. code-block:: python print(data_bundle) .. code-block:: text In total 3 datasets: dev has 1200 instances. train has 9600 instances. test has 1200 instances. In total 0 vocabs: 可以看出，该data\_bundle中一个含有三个 :class:`~fastNLP.DataSet` 。通过下面的代码，我们可以查看DataSet的基本情况 .. code-block:: python print(data_bundle.get_dataset('train')[:2]) # 查看Train集前两个sample .. code-block:: text DataSet({'raw_chars': 选择珠江花园的原因就是方便，有电动扶梯直接到达海边，周围餐馆、食廊、商场、超市、摊位一应俱全。酒店装修一般，但还算整洁。 泳池在大堂的屋顶，因此很小，不过女儿倒是喜欢。 包的早餐是西式的，还算丰富。 服务吗，一般 type=str, 'target': 1 type=str}, {'raw_chars': 15.4寸笔记本的键盘确实爽，基本跟台式机差不多了，蛮喜欢数字小键盘，输数字特方便，样子也很美观，做工也相当不错 type=str, 'target': 1 type=str}) (2) 预处理数据 ~~~~~~~~~~~~~~~~~~~~ 在NLP任务中，预处理一般包括: (a) 将一整句话切分成汉字或者词; (b) 将文本转换为index fastNLP中也提供了多种数据集的处理类，这里我们直接使用fastNLP的ChnSentiCorpPipe。更多关于Pipe的说明可以参考 :mod:`~fastNLP.io.pipe` 。 .. code-block:: python from fastNLP.io import ChnSentiCorpPipe pipe = ChnSentiCorpPipe() data_bundle = pipe.process(data_bundle) # 所有的Pipe都实现了process()方法，且输入输出都为DataBundle类型 print(data_bundle) # 打印data_bundle，查看其变化 .. code-block:: text In total 3 datasets: dev has 1200 instances. train has 9600 instances. test has 1200 instances. In total 2 vocabs: chars has 4409 entries. target has 2 entries. 可以看到除了之前已经包含的3个 :class:`~fastNLP.DataSet` ,还新增了两个 :class:`~fastNLP.Vocabulary` 。我们可以打印DataSet中的内容 .. code-block:: python print(data_bundle.get_dataset('train')[:2]) .. code-block:: text DataSet({'raw_chars': 选择珠江花园的原因就是方便，有电动扶梯直接到达海边，周围餐馆、食廊、商场、超市、摊位一应俱全。酒店装修一般，但还算整洁。 泳池在大堂的屋顶，因此很小，不过女儿倒是喜欢。 包的早餐是西式的，还算丰富。 服务吗，一般 type=str, 'target': 1 type=int, 'chars': [338, 464, 1400, 784, 468, 739, 3, 289, 151, 21, 5, 88, 143, 2, 9, 81, 134, 2573, 766, 233, 196, 23, 536, 342, 297, 2, 405, 698, 132, 281, 74, 744, 1048, 74, 420, 387, 74, 412, 433, 74, 2021, 180, 8, 219, 1929, 213, 4, 34, 31, 96, 363, 8, 230, 2, 66, 18, 229, 331, 768, 4, 11, 1094, 479, 17, 35, 593, 3, 1126, 967, 2, 151, 245, 12, 44, 2, 6, 52, 260, 263, 635, 5, 152, 162, 4, 11, 336, 3, 154, 132, 5, 236, 443, 3, 2, 18, 229, 761, 700, 4, 11, 48, 59, 653, 2, 8, 230] type=list, 'seq_len': 106 type=int}, {'raw_chars': 15.4寸笔记本的键盘确实爽，基本跟台式机差不多了，蛮喜欢数字小键盘，输数字特方便，样子也很美观，做工也相当不错 type=str, 'target': 1 type=int, 'chars': [50, 133, 20, 135, 945, 520, 343, 24, 3, 301, 176, 350, 86, 785, 2, 456, 24, 461, 163, 443, 128, 109, 6, 47, 7, 2, 916, 152, 162, 524, 296, 44, 301, 176, 2, 1384, 524, 296, 259, 88, 143, 2, 92, 67, 26, 12, 277, 269, 2, 188, 223, 26, 228, 83, 6, 63] type=list, 'seq_len': 56 type=int}) 新增了一列为数字列表的chars，以及变为数字的target列。可以看出这两列的名称和刚好与data\_bundle中两个Vocabulary的名称是一致的，我们可以打印一下Vocabulary看一下里面的内容。 .. code-block:: python char_vocab = data_bundle.get_vocab('chars') print(char_vocab) .. code-block:: text Vocabulary(['选', '择', '珠', '江', '花']...) Vocabulary是一个记录着词语与index之间映射关系的类，比如 .. code-block:: python index = char_vocab.to_index('选') print("'选'的index是{}".format(index)) # 这个值与上面打印出来的第一个instance的chars的第一个index是一致的 print("index:{}对应的汉字是{}".format(index, char_vocab.to_word(index))) .. code-block:: text '选'的index是338 index:338对应的汉字是选 (3) 选择预训练词向量 ~~~~~~~~~~~~~~~~~~~~ 由于Word2vec, Glove, Elmo, Bert等预训练模型可以增强模型的性能，所以在训练具体任务前，选择合适的预训练词向量非常重要。 在fastNLP中我们提供了多种Embedding使得加载这些预训练模型的过程变得更加便捷。 这里我们先给出一个使用word2vec的中文汉字预训练的示例，之后再给出一个使用Bert的文本分类。 这里使用的预训练词向量为'cn-fastnlp-100d'，fastNLP将自动下载该embedding至本地缓存， fastNLP支持使用名字指定的Embedding以及相关说明可以参见 :mod:`fastNLP.embeddings` .. code-block:: python from fastNLP.embeddings import StaticEmbedding word2vec_embed = StaticEmbedding(char_vocab, model_dir_or_name='cn-char-fastnlp-100d') .. code-block:: text Found 4321 out of 4409 compound in the pre-training embedding. (4) 创建模型 ~~~~~~~~~~~~ 这里我们使用到的模型结构如下所示 .. todo:: 补图 .. code-block:: python from torch import nn from fastNLP.modules import LSTM import torch # 定义模型 class BiLSTMMaxPoolCls(nn.Module): def __init__(self, embed, num_classes, hidden_size=400, num_layers=1, dropout=0.3): super().__init__() self.embed = embed self.lstm = LSTM(self.embed.embedding_dim, hidden_size=hidden_size//2, num_layers=num_layers, batch_first=True, bidirectional=True) self.dropout_layer = nn.Dropout(dropout) self.fc = nn.Linear(hidden_size, num_classes) def forward(self, chars, seq_len): # 这里的名称必须和DataSet中相应的field对应，比如之前我们DataSet中有chars，这里就必须为chars # chars:[batch_size, max_len] # seq_len: [batch_size, ] chars = self.embed(chars) outputs, _ = self.lstm(chars, seq_len) outputs = self.dropout_layer(outputs) outputs, _ = torch.max(outputs, dim=1) outputs = self.fc(outputs) return {'pred':outputs} # [batch_size,], 返回值必须是dict类型，且预测值的key建议设为pred # 初始化模型 model = BiLSTMMaxPoolCls(word2vec_embed, len(data_bundle.get_vocab('target'))) (5) 训练模型 ~~~~~~~~~~~~ fastNLP提供了Trainer对象来组织训练过程，包括完成loss计算(所以在初始化Trainer的时候需要指定loss类型)，梯度更新(所以在初始化Trainer的时候需要提供优化器optimizer)以及在验证集上的性能验证(所以在初始化时需要提供一个Metric) .. code-block:: python from fastNLP import Trainer from fastNLP import CrossEntropyLoss from torch.optim import Adam from fastNLP import AccuracyMetric loss = CrossEntropyLoss() optimizer = Adam(model.parameters(), lr=0.001) metric = AccuracyMetric() device = 0 if torch.cuda.is_available() else 'cpu' # 如果有gpu的话在gpu上运行，训练速度会更快 trainer = Trainer(train_data=data_bundle.get_dataset('train'), model=model, loss=loss, optimizer=optimizer, batch_size=32, dev_data=data_bundle.get_dataset('dev'), metrics=metric, device=device) trainer.train() # 开始训练，训练完成之后默认会加载在dev上表现最好的模型 # 在测试集上测试一下模型的性能 from fastNLP import Tester print("Performance on test is:") tester = Tester(data=data_bundle.get_dataset('test'), model=model, metrics=metric, batch_size=64, device=device) tester.test() .. code-block:: text input fields after batch(if batch size is 2): target: (1)type:torch.Tensor (2)dtype:torch.int64, (3)shape:torch.Size([2]) chars: (1)type:torch.Tensor (2)dtype:torch.int64, (3)shape:torch.Size([2, 106]) seq_len: (1)type:torch.Tensor (2)dtype:torch.int64, (3)shape:torch.Size([2]) target fields after batch(if batch size is 2): target: (1)type:torch.Tensor (2)dtype:torch.int64, (3)shape:torch.Size([2]) seq_len: (1)type:torch.Tensor (2)dtype:torch.int64, (3)shape:torch.Size([2]) Evaluate data in 0.01 seconds! training epochs started 2019-09-03-23-57-10 HBox(children=(IntProgress(value=0, layout=Layout(flex='2'), max=3000), HTML(value='')), layout=Layout(display… HBox(children=(IntProgress(value=0, layout=Layout(flex='2'), max=38), HTML(value='')), layout=Layout(display='… Evaluate data in 0.43 seconds! Evaluation on dev at Epoch 1/10. Step:300/3000: AccuracyMetric: acc=0.81 HBox(children=(IntProgress(value=0, layout=Layout(flex='2'), max=38), HTML(value='')), layout=Layout(display='… Evaluate data in 0.44 seconds! Evaluation on dev at Epoch 2/10. Step:600/3000: AccuracyMetric: acc=0.8675 HBox(children=(IntProgress(value=0, layout=Layout(flex='2'), max=38), HTML(value='')), layout=Layout(display='… Evaluate data in 0.44 seconds! Evaluation on dev at Epoch 3/10. Step:900/3000: AccuracyMetric: acc=0.878333 HBox(children=(IntProgress(value=0, layout=Layout(flex='2'), max=38), HTML(value='')), layout=Layout(display='… Evaluate data in 0.43 seconds! Evaluation on dev at Epoch 4/10. Step:1200/3000: AccuracyMetric: acc=0.873333 HBox(children=(IntProgress(value=0, layout=Layout(flex='2'), max=38), HTML(value='')), layout=Layout(display='… Evaluate data in 0.44 seconds! Evaluation on dev at Epoch 5/10. Step:1500/3000: AccuracyMetric: acc=0.878333 HBox(children=(IntProgress(value=0, layout=Layout(flex='2'), max=38), HTML(value='')), layout=Layout(display='… Evaluate data in 0.42 seconds! Evaluation on dev at Epoch 6/10. Step:1800/3000: AccuracyMetric: acc=0.895833 HBox(children=(IntProgress(value=0, layout=Layout(flex='2'), max=38), HTML(value='')), layout=Layout(display='… Evaluate data in 0.44 seconds! Evaluation on dev at Epoch 7/10. Step:2100/3000: AccuracyMetric: acc=0.8975 HBox(children=(IntProgress(value=0, layout=Layout(flex='2'), max=38), HTML(value='')), layout=Layout(display='… Evaluate data in 0.43 seconds! Evaluation on dev at Epoch 8/10. Step:2400/3000: AccuracyMetric: acc=0.894167 HBox(children=(IntProgress(value=0, layout=Layout(flex='2'), max=38), HTML(value='')), layout=Layout(display='… Evaluate data in 0.48 seconds! Evaluation on dev at Epoch 9/10. Step:2700/3000: AccuracyMetric: acc=0.8875 HBox(children=(IntProgress(value=0, layout=Layout(flex='2'), max=38), HTML(value='')), layout=Layout(display='… Evaluate data in 0.43 seconds! Evaluation on dev at Epoch 10/10. Step:3000/3000: AccuracyMetric: acc=0.895833 In Epoch:7/Step:2100, got best dev performance: AccuracyMetric: acc=0.8975 Reloaded the best model. HBox(children=(IntProgress(value=0, layout=Layout(flex='2'), max=19), HTML(value='')), layout=Layout(display='… Evaluate data in 0.34 seconds! [tester] AccuracyMetric: acc=0.8975 {'AccuracyMetric': {'acc': 0.8975}} 使用Bert进行文本分类 ~~~~~~~~~~~~~~~~~~~~ .. code-block:: python # 只需要切换一下Embedding即可 from fastNLP.embeddings import BertEmbedding # 这里为了演示一下效果，所以默认Bert不更新权重 bert_embed = BertEmbedding(char_vocab, model_dir_or_name='cn', auto_truncate=True, requires_grad=False) model = BiLSTMMaxPoolCls(bert_embed, len(data_bundle.get_vocab('target')), ) import torch from fastNLP import Trainer from fastNLP import CrossEntropyLoss from torch.optim import Adam from fastNLP import AccuracyMetric loss = CrossEntropyLoss() optimizer = Adam(model.parameters(), lr=2e-5) metric = AccuracyMetric() device = 0 if torch.cuda.is_available() else 'cpu' # 如果有gpu的话在gpu上运行，训练速度会更快 trainer = Trainer(train_data=data_bundle.get_dataset('train'), model=model, loss=loss, optimizer=optimizer, batch_size=16, dev_data=data_bundle.get_dataset('test'), metrics=metric, device=device, n_epochs=3) trainer.train() # 开始训练，训练完成之后默认会加载在dev上表现最好的模型 # 在测试集上测试一下模型的性能 from fastNLP import Tester print("Performance on test is:") tester = Tester(data=data_bundle.get_dataset('test'), model=model, metrics=metric, batch_size=64, device=device) tester.test() .. code-block:: text loading vocabulary file /home/yh/.fastNLP/embedding/bert-chinese-wwm/vocab.txt Load pre-trained BERT parameters from file /home/yh/.fastNLP/embedding/bert-chinese-wwm/chinese_wwm_pytorch.bin. Start to generating word pieces for word. Found(Or segment into word pieces) 4286 words out of 4409. input fields after batch(if batch size is 2): target: (1)type:torch.Tensor (2)dtype:torch.int64, (3)shape:torch.Size([2]) chars: (1)type:torch.Tensor (2)dtype:torch.int64, (3)shape:torch.Size([2, 106]) seq_len: (1)type:torch.Tensor (2)dtype:torch.int64, (3)shape:torch.Size([2]) target fields after batch(if batch size is 2): target: (1)type:torch.Tensor (2)dtype:torch.int64, (3)shape:torch.Size([2]) seq_len: (1)type:torch.Tensor (2)dtype:torch.int64, (3)shape:torch.Size([2]) Evaluate data in 0.05 seconds! training epochs started 2019-09-04-00-02-37 HBox(children=(IntProgress(value=0, layout=Layout(flex='2'), max=3600), HTML(value='')), layout=Layout(display… HBox(children=(IntProgress(value=0, layout=Layout(flex='2'), max=150), HTML(value='')), layout=Layout(display=… Evaluate data in 15.89 seconds! Evaluation on dev at Epoch 1/3. Step:1200/3600: AccuracyMetric: acc=0.9 HBox(children=(IntProgress(value=0, layout=Layout(flex='2'), max=150), HTML(value='')), layout=Layout(display=… Evaluate data in 15.92 seconds! Evaluation on dev at Epoch 2/3. Step:2400/3600: AccuracyMetric: acc=0.904167 HBox(children=(IntProgress(value=0, layout=Layout(flex='2'), max=150), HTML(value='')), layout=Layout(display=… Evaluate data in 15.91 seconds! Evaluation on dev at Epoch 3/3. Step:3600/3600: AccuracyMetric: acc=0.918333 In Epoch:3/Step:3600, got best dev performance: AccuracyMetric: acc=0.918333 Reloaded the best model. Performance on test is: HBox(children=(IntProgress(value=0, layout=Layout(flex='2'), max=19), HTML(value='')), layout=Layout(display='… Evaluate data in 29.24 seconds! [tester] AccuracyMetric: acc=0.919167 {'AccuracyMetric': {'acc': 0.919167}}