RNN来解决命名实体识别问题（week2-NER，HSE-AML，Natural language processing, 第一周作业，配图，注释)
简介
（本次笔记是基于HSE的课程natural language processing，第二周的作业。完整代码在最后）
使用RNN来解决命名实体识别（NER）是NLP的常见问题。它能够将文字中的实体提取出来，比如人名，组织，地名等。这次作业的训练集来自Twitter，使用的结构是Bi-Directional Long Short-Term Memory Networks (Bi-LSTMs)。作业可以分成三个部分。
1. 文本预处理。
对于twitter中的文本，需要以’@’，’http’开始的字符串替换成’USER’,’URL’.并且将文本转化为一个dict和一个list。dict用来存储tok2idx，list用来存储idx2tok。
2. 构建Bi-LSTM模型.
有以下几个方法，分别用来申明placeholder，定义层，计算输出，计算损失，优化方法。
- declare_placeholders()
- build_layers(vocabulary_size, embedding_dim, n_hidden_rnn, - n_tags)
- compute_predictions()
- compute_loss(n_tags, PAD_index)
- perform_optimization()
3. 训练和评估
细节
1. declare_placeholders(self):
首先定义placeholders
- input_batch — 输入批序列，维度[batch_size, sequence_len];
- ground_truth_tags — 输出批序列，维度[batch_size, sequence_len];
- lengths — 未padding之前的序列长度，小于sequence_len，维度[batch_size];
- dropout_ph — dropout 概率，默认1;
- learning_rate_ph — 学习率，默认1.
注意：RNN的batch，是将一个batch的序列补齐，变成同样的长度，每次输入的是一个batch中多个sequence的同一个位置上的词语。所以这里有padding。
def declare_placeholders(self):
"""Specifies placeholders for the model."""
# Placeholders for input and ground truth output.
self.input_batch = tf.placeholder(dtype=tf.int32, shape=[None, None], name='input_batch')
self.ground_truth_tags = tf.placeholder(dtype = tf.int32, shape = [None,None], name = 'ground_truth')
# Placeholder for lengths of the sequences.
self.lengths = tf.placeholder(dtype=tf.int32, shape=[None], name='lengths')
# Placeholder for a dropout keep probability. If we don't feed
# a value for this placeholder, it will be equal to 1.0.
self.dropout_ph = tf.placeholder_with_default(tf.cast(1.0, tf.float32), shape=[])
# Placeholder for a learning rate (tf.float32).
self.learning_rate_ph = tf.placeholder_with_default(tf.cast(0.0001,tf.float32),shape = []) 2. build_layers(self, vocabulary_size, embedding_dim, n_hidden_rnn, n_tags):
这是最重要的方法，定义了层的具体实现逻辑。
这里有个疑问，embedding在这里是初始化过得的，那这里应该包括了对embedding的trainning， 即word2vec。但是我没有看到是如何做embedding的training。
- 定义前向和后向LSTM. 建议使用BasicLSTMCell。
- 定义 DropoutWrapper，dropout用来防止过拟合。
- 从预定义的embedding_matrix查找到对应输入input_batch的embeddings。
- 将embeddings传递给Bidirectional Dynamic RNN，并设置好前向后向LSTM。
- 使用length来避免对PAD的计算。
- 最后使用一个全连接层，它的输出将会直接输入损失函数。
def build_layers(self, vocabulary_size, embedding_dim, n_hidden_rnn, n_tags):
"""Specifies bi-LSTM architecture and computes logits for inputs."""
# Create embedding variable (tf.Variable) with dtype tf.float32
initial_embedding_matrix = np.random.randn(vocabulary_size, embedding_dim) / np.sqrt(embedding_dim)
embedding_matrix_variable = tf.Variable(initial_value = initial_embedding_matrix, dtype = 'float32', name = 'embedding_matrix')
# Create RNN cells (for example, tf.nn.rnn_cell.BasicLSTMCell) with n_hidden_rnn number of units
# and dropout (tf.nn.rnn_cell.DropoutWrapper), initializing all *_keep_prob with dropout placeholder.
forward_cell = tf.nn.rnn_cell.DropoutWrapper(
tf.nn.rnn_cell.BasicLSTMCell(num_units=n_hidden_rnn, forget_bias=1.0),
input_keep_prob=self.dropout_ph,
output_keep_prob=self.dropout_ph,
state_keep_prob=self.dropout_ph
)
backward_cell = tf.nn.rnn_cell.DropoutWrapper(
tf.nn.rnn_cell.BasicLSTMCell(num_units=n_hidden_rnn, forget_bias=1.0),
input_keep_prob=self.dropout_ph,
output_keep_prob=self.dropout_ph,
state_keep_prob=self.dropout_ph
)
# Look up embeddings for self.input_batch (tf.nn.embedding_lookup).
# Shape: [batch_size, sequence_len, embedding_dim].
embeddings = tf.nn.embedding_lookup(embedding_matrix_variable,self.input_batch)
# Pass them through Bidirectional Dynamic RNN (tf.nn.bidirectional_dynamic_rnn).
# Shape: [batch_size, sequence_len, 2 * n_hidden_rnn].
# Also don't forget to initialize sequence_length as self.lengths and dtype as tf.float32.
(rnn_output_fw, rnn_output_bw), _ = tf.nn.bidirectional_dynamic_rnn(cell_fw = forward_cell,
cell_bw = backward_cell,
dtype = tf.float32,
inputs = embeddings,
sequence_length = self.lengths)
rnn_output = tf.concat([rnn_output_fw, rnn_output_bw], axis=2)
# Dense layer on top.
# Shape: [batch_size, sequence_len, n_tags].
self.logits = tf.layers.dense(rnn_output, n_tags, activation=None)3. compute_predictions(self):
将layer的输出作为softmax的输入，然后使用argmax函数来计算预测结果。
4. compute_loss(self, n_tags, PAD_index):
在训练过程中，并不需要网络预测的结果，但是需要损失函数。这里我们使用cross-entropy loss。
注意：这个损失函数需要使用到logits上，而不是softmax的输出概率。并且我们不需要计算’PAD’造成的损失，提前将他屏蔽掉。
def compute_predictions(self):
"""Transforms logits to probabilities and finds the most probable tags."""
# Create softmax (tf.nn.softmax) function
softmax_output = tf.nn.softmax(self.logits)
# Use argmax (tf.argmax) to get the most probable tags
# Don't forget to set axis=-1
# otherwise argmax will be calculated in a wrong way
self.predictions = tf.argmax(softmax_output, axis = -1)
def compute_loss(self, n_tags, PAD_index):
"""Computes masked cross-entopy loss with logits."""
# Create cross entropy function function (tf.nn.softmax_cross_entropy_with_logits)
ground_truth_tags_one_hot = tf.one_hot(self.ground_truth_tags, n_tags)
loss_tensor = tf.nn.softmax_cross_entropy_with_logits(labels = ground_truth_tags_one_hot, logits = self.logits)
mask = tf.cast(tf.not_equal(self.input_batch, PAD_index), tf.float32)
# Create loss function which doesn't operate with tokens (tf.reduce_mean)
# Be careful that the argument of tf.reduce_mean should be
# multiplication of mask and loss_tensor.
self.loss = tf.reduce_mean(tf.reduce_sum(mask * loss_tensor, axis = -1)/tf.reduce_sum(mask,axis = -1))5. perform_optimization(self):
最后定义优化方法。建议使用Adam方法，并且对梯度使用clipping以消除梯度爆炸的影响。
def perform_optimization(self):
"""Specifies the optimizer and train_op for the model."""
# Create an optimizer (tf.train.AdamOptimizer)
self.optimizer = tf.train.AdamOptimizer(self.learning_rate_ph)######### YOUR CODE HERE #############
self.grads_and_vars = self.optimizer.compute_gradients(self.loss)
# Gradient clipping (tf.clip_by_norm) for self.grads_and_vars
# Pay attention that you need to apply this operation only for gradients
# because self.grads_and_vars contains also variables.
# list comprehension might be useful in this case.
clip_norm = tf.cast(1.0, tf.float32)
#self.grads_and_vars = tf.clip_by_norm(self.grads_and_vars) ######### YOUR CODE HERE #############
self.grads_and_vars = [(tf.clip_by_norm(grad, clip_norm), var) for grad, var in self.grads_and_vars]
self.train_op = self.optimizer.apply_gradients(self.grads_and_vars)（下面是具体的实现。是HSE的课程natural language processing，第二周的作业。）
Recognize named entities on Twitter with LSTMs
In this assignment, you will use a recurrent neural network to solve Named Entity Recognition (NER) problem. NER is a common task in natural language processing systems. It serves for extraction such entities from the text as persons, organizations, locations, etc. In this task you will experiment to recognize named entities from Twitter.
For example, we want to extract persons’ and organizations’ names from the text. Than for the input text:
Ian Goodfellow works for Google Brain
a NER model needs to provide the following sequence of tags:
B-PER I-PER O O B-ORG I-ORG
Where B- and I- prefixes stand for the beginning and inside of the entity, while O stands for out of tag or no tag. Markup with the prefix scheme is called BIO markup. This markup is introduced for distinguishing of consequent entities with similar types.
A solution of the task will be based on neural networks, particularly, on Bi-Directional Long Short-Term Memory Networks (Bi-LSTMs).
Libraries
For this task you will need the following libraries:
- Tensorflow — an open-source software library for Machine Intelligence.
- Numpy — a package for scientific computing.
If you have never worked with Tensorflow, you would probably need to read some tutorials during your work on this assignment, e.g. this one could be a good starting point.
Data
The following cell will download all data required for this assignment into the folder week2/data.
import sys
sys.path.append("..")
from common.download_utils import download_week2_resources
#download_week2_resources()Load the Twitter Named Entity Recognition corpus
We will work with a corpus, which contains twits with NE tags. Every line of a file contains a pair of a token (word/punctuation symbol) and a tag, separated by a whitespace. Different tweets are separated by an empty line.
The function read_data reads a corpus from the file_path and returns two lists: one with tokens and one with the corresponding tags. You need to complete this function by adding a code, which will replace a user’s nickname to
def read_data(file_path):
tokens = []
tags = []
tweet_tokens = []
tweet_tags = []
for line in open(file_path, encoding='utf-8'):
line = line.strip()
if not line:
if tweet_tokens:
tokens.append(tweet_tokens)
tags.append(tweet_tags)
tweet_tokens = []
tweet_tags = []
else:
token, tag = line.split()
# Replace all urls with token
# Replace all users with token
######################################
######### YOUR CODE HERE #############
######################################
if((token.startswith('http://'))or(token.startswith('https://'))):
token = ''
elif(token.startswith('@')):
token = ''
tweet_tokens.append(token)
tweet_tags.append(tag)
return tokens, tagsAnd now we can load three separate parts of the dataset:
- train data for training the model;
- validation data for evaluation and hyperparameters tuning;
- test data for final evaluation of the model.
train_tokens, train_tags = read_data('data/train.txt')
validation_tokens, validation_tags = read_data('data/validation.txt')
test_tokens, test_tags = read_data('data/test.txt')You should always understand what kind of data you deal with. For this purpose, you can print the data running the following cell:
train_tokens[:3]for i in range(6):
for token, tag in zip(train_tokens[i], train_tags[i]):
print('%s\t%s' % (token, tag))
print()Prepare dictionaries
To train a neural network, we will use two mappings:
- {token} → → {token id}: address the row in embeddings matrix for the current token;
- {tag} → → {tag id}: one-hot ground truth probability distribution vectors for computing the loss at the output of the network.
Now you need to implement the function build_dict which will return {token or tag} → → {index} and vice versa.
from collections import defaultdictdef build_dict(tokens_or_tags, special_tokens):
"""
tokens_or_tags: a list of lists of tokens or tags
special_tokens: some special tokens
"""
# Create a dictionary with default value 0
tok2idx = defaultdict(lambda: 0)
idx2tok = []
# Create mappings from tokens (or tags) to indices and vice versa.
# At first, add special tokens (or tags) to the dictionaries.
# The first special token must have index 0.
# Mapping tok2idx should contain each token or tag only once.
# To do so, you should:
# 1. extract unique tokens/tags from the tokens_or_tags variable, which is not
# occure in special_tokens (because they could have non-empty intersection)
# 2. index them (for example, you can add them into the list idx2tok
# 3. for each token/tag save the index into tok2idx).
######################################
######### YOUR CODE HERE #############
######################################
idx = 0
for token in special_tokens:
idx2tok.append(token)
tok2idx[token] = idx
idx += 1
for token_list in tokens_or_tags:
for token in token_list:
if token not in tok2idx:
idx2tok.append(token)
tok2idx[token] = idx
idx += 1
return tok2idx, idx2tokAfter implementing the function build_dict you can make dictionaries for tokens and tags. Special tokens in our case will be:
-
-
special_tokens = ['' , '' ]
special_tags = ['O']
# Create dictionaries #can not understand here, sicne the token sets and train sets can not be matched.
token2idx, idx2token = build_dict(train_tokens + validation_tokens, special_tokens)
tag2idx, idx2tag = build_dict(train_tags, special_tags)type(idx2token)list
The next additional functions will help you to create the mapping between tokens and ids for a sentence.
def words2idxs(tokens_list):
return [token2idx[word] for word in tokens_list]
def tags2idxs(tags_list):
return [tag2idx[tag] for tag in tags_list]
def idxs2words(idxs):
return [idx2token[idx] for idx in idxs]
def idxs2tags(idxs):
return [idx2tag[idx] for idx in idxs]Generate batches
Neural Networks are usually trained with batches. It means that weight updates of the network are based on several sequences at every single time. The tricky part is that all sequences within a batch need to have the same length. So we will pad them with a special
def batches_generator(batch_size, tokens, tags,
shuffle=True, allow_smaller_last_batch=True):
"""Generates padded batches of tokens and tags."""
n_samples = len(tokens)
if shuffle:
order = np.random.permutation(n_samples)
else:
order = np.arange(n_samples)
n_batches = n_samples // batch_size
if allow_smaller_last_batch and n_samples % batch_size:
n_batches += 1
for k in range(n_batches):
batch_start = k * batch_size
batch_end = min((k + 1) * batch_size, n_samples)
current_batch_size = batch_end - batch_start
x_list = []
y_list = []
max_len_token = 0
for idx in order[batch_start: batch_end]:
x_list.append(words2idxs(tokens[idx]))
y_list.append(tags2idxs(tags[idx]))
max_len_token = max(max_len_token, len(tags[idx]))
# the inputs, both tokens and tags, are the index
# Fill in the data into numpy nd-arrays filled with padding indices.
x = np.ones([current_batch_size, max_len_token], dtype=np.int32) * token2idx['' ]
y = np.ones([current_batch_size, max_len_token], dtype=np.int32) * tag2idx['O']
lengths = np.zeros(current_batch_size, dtype=np.int32)
for n in range(current_batch_size):
utt_len = len(x_list[n])
x[n, :utt_len] = x_list[n]
lengths[n] = utt_len
y[n, :utt_len] = y_list[n]
yield x, y, lengthsBuild a recurrent neural network
This is the most important part of the assignment. Here we will specify the network architecture based on TensorFlow building blocks. It’s fun and easy as a lego constructor! We will create an LSTM network which will produce probability distribution over tags for each token in a sentence. To take into account both right and left contexts of the token, we will use Bi-Directional LSTM (Bi-LSTM). Dense layer will be used on top to perform tag classification.
import tensorflow as tf
import numpy as np/home/lika/anaconda3/lib/python3.6/site-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.
from ._conv import register_converters as _register_converters
class BiLSTMModel():
passFirst, we need to create placeholders to specify what data we are going to feed into the network during the execution time. For this task we will need the following placeholders:
- input_batch — sequences of words (the shape equals to [batch_size, sequence_len]);
- ground_truth_tags — sequences of tags (the shape equals to [batch_size, sequence_len]);
- lengths — lengths of not padded sequences (the shape equals to [batch_size]);
- dropout_ph — dropout keep probability; this placeholder has a predefined value 1;
- learning_rate_ph — learning rate; we need this placeholder because we want to change the value during training.
It could be noticed that we use None in the shapes in the declaration, which means that data of any size can be feeded.
You need to complete the function declare_placeholders.
def declare_placeholders(self):
"""Specifies placeholders for the model."""
# Placeholders for input and ground truth output.
self.input_batch = tf.placeholder(dtype=tf.int32, shape=[None, None], name='input_batch')
self.ground_truth_tags = tf.placeholder(dtype = tf.int32, shape = [None,None], name = 'ground_truth')######### YOUR CODE HERE #############
# Placeholder for lengths of the sequences.
self.lengths = tf.placeholder(dtype=tf.int32, shape=[None], name='lengths')
# Placeholder for a dropout keep probability. If we don't feed
# a value for this placeholder, it will be equal to 1.0.
self.dropout_ph = tf.placeholder_with_default(tf.cast(1.0, tf.float32), shape=[])
# Placeholder for a learning rate (tf.float32).
self.learning_rate_ph = tf.placeholder_with_default(tf.cast(0.0001,tf.float32),shape = []) ######### YOUR CODE HERE #############BiLSTMModel.__declare_placeholders = classmethod(declare_placeholders)Now, let us specify the layers of the neural network. First, we need to perform some preparatory steps:
- Create embeddings matrix with tf.Variable. Specify its name (embeddings_matrix), type (tf.float32), and initialize with random values.
- Create forward and backward LSTM cells. TensorFlow provides a number of RNN cells ready for you. We suggest that you use BasicLSTMCell, but you can also experiment with other types, e.g. GRU cells. This blogpost could be interesting if you want to learn more about the differences.
- Wrap your cells with DropoutWrapper. Dropout is an important regularization technique for neural networks. Specify all keep probabilities using the dropout placeholder that we created before.
After that, you can build the computation graph that transforms an input_batch:
- Look up embeddings for an input_batch in the prepared embedding_matrix.
- Pass the embeddings through Bidirectional Dynamic RNN with the specified forward and backward cells. Use the lengths placeholder here to avoid computations for padding tokens inside the RNN.
- Create a dense layer on top. Its output will be used directly in loss function.
Fill in the code below. In case you need to debug something, the easiest way is to check that tensor shapes of each step match the expected ones.
def build_layers(self, vocabulary_size, embedding_dim, n_hidden_rnn, n_tags):
"""Specifies bi-LSTM architecture and computes logits for inputs."""
# Create embedding variable (tf.Variable) with dtype tf.float32
initial_embedding_matrix = np.random.randn(vocabulary_size, embedding_dim) / np.sqrt(embedding_dim)
embedding_matrix_variable = tf.Variable(initial_value = initial_embedding_matrix, dtype = 'float32', name = 'embedding_matrix')######### YOUR CODE HERE #############
# Create RNN cells (for example, tf.nn.rnn_cell.BasicLSTMCell) with n_hidden_rnn number of units
# and dropout (tf.nn.rnn_cell.DropoutWrapper), initializing all *_keep_prob with dropout placeholder.
#forward_cell = tf.nn.rnn_cell.BasicLSTMCell(n_hidden_rnn, forget_bias=1.0, state_is_tuple=True)######### YOUR CODE HERE #############
#backward_cell = tf.nn.rnn_cell.BasicLSTMCell(n_hidden_rnn, forget_bias=1.0, state_is_tuple=True)######### YOUR CODE HERE #############
forward_cell = tf.nn.rnn_cell.DropoutWrapper(
tf.nn.rnn_cell.BasicLSTMCell(num_units=n_hidden_rnn, forget_bias=1.0),
input_keep_prob=self.dropout_ph,
output_keep_prob=self.dropout_ph,
state_keep_prob=self.dropout_ph
)
backward_cell = tf.nn.rnn_cell.DropoutWrapper(
tf.nn.rnn_cell.BasicLSTMCell(num_units=n_hidden_rnn, forget_bias=1.0),
input_keep_prob=self.dropout_ph,
output_keep_prob=self.dropout_ph,
state_keep_prob=self.dropout_ph
)
# Look up embeddings for self.input_batch (tf.nn.embedding_lookup).
# Shape: [batch_size, sequence_len, embedding_dim].
embeddings = tf.nn.embedding_lookup(embedding_matrix_variable,self.input_batch)######### YOUR CODE HERE #############
# Pass them through Bidirectional Dynamic RNN (tf.nn.bidirectional_dynamic_rnn).
# Shape: [batch_size, sequence_len, 2 * n_hidden_rnn].
# Also don't forget to initialize sequence_length as self.lengths and dtype as tf.float32.
(rnn_output_fw, rnn_output_bw), _ = tf.nn.bidirectional_dynamic_rnn(cell_fw = forward_cell,
cell_bw = backward_cell,
dtype = tf.float32,
inputs = embeddings,
sequence_length = self.lengths)######### YOUR CODE HERE #############
rnn_output = tf.concat([rnn_output_fw, rnn_output_bw], axis=2)
# Dense layer on top.
# Shape: [batch_size, sequence_len, n_tags].
self.logits = tf.layers.dense(rnn_output, n_tags, activation=None)BiLSTMModel.__build_layers = classmethod(build_layers)To compute the actual predictions of the neural network, you need to apply softmax to the last layer and find the most probable tags with argmax.
def compute_predictions(self):
"""Transforms logits to probabilities and finds the most probable tags."""
# Create softmax (tf.nn.softmax) function
softmax_output = tf.nn.softmax(self.logits) ######### YOUR CODE HERE #############
# Use argmax (tf.argmax) to get the most probable tags
# Don't forget to set axis=-1
# otherwise argmax will be calculated in a wrong way
self.predictions = tf.argmax(softmax_output, axis = -1)######### YOUR CODE HERE #############BiLSTMModel.__compute_predictions = classmethod(compute_predictions)During training we do not need predictions of the network, but we need a loss function. We will use cross-entropy loss, efficiently implemented in TF as
cross entropy with logits. Note that it should be applied to logits of the model (not to softmax probabilities!). Also note, that we do not want to take into account loss terms coming from
def compute_loss(self, n_tags, PAD_index):
"""Computes masked cross-entopy loss with logits."""
# Create cross entropy function function (tf.nn.softmax_cross_entropy_with_logits)
ground_truth_tags_one_hot = tf.one_hot(self.ground_truth_tags, n_tags)
loss_tensor = tf.nn.softmax_cross_entropy_with_logits(labels = ground_truth_tags_one_hot, logits = self.logits)######### YOUR CODE HERE #############
mask = tf.cast(tf.not_equal(self.input_batch, PAD_index), tf.float32)
# Create loss function which doesn't operate with tokens (tf.reduce_mean)
# Be careful that the argument of tf.reduce_mean should be
# multiplication of mask and loss_tensor.
self.loss = tf.reduce_mean(tf.reduce_sum(mask * loss_tensor, axis = -1)/tf.reduce_sum(mask,axis = -1))######### YOUR CODE HERE #############BiLSTMModel.__compute_loss = classmethod(compute_loss)The last thing to specify is how we want to optimize the loss.
We suggest that you use Adam optimizer with a learning rate from the corresponding placeholder.
You will also need to apply clipping to eliminate exploding gradients. It can be easily done with clip_by_norm function.
def perform_optimization(self):
"""Specifies the optimizer and train_op for the model."""
# Create an optimizer (tf.train.AdamOptimizer)
self.optimizer = tf.train.AdamOptimizer(self.learning_rate_ph)######### YOUR CODE HERE #############
self.grads_and_vars = self.optimizer.compute_gradients(self.loss)
# Gradient clipping (tf.clip_by_norm) for self.grads_and_vars
# Pay attention that you need to apply this operation only for gradients
# because self.grads_and_vars contains also variables.
# list comprehension might be useful in this case.
clip_norm = tf.cast(1.0, tf.float32)
#self.grads_and_vars = tf.clip_by_norm(self.grads_and_vars) ######### YOUR CODE HERE #############
self.grads_and_vars = [(tf.clip_by_norm(grad, clip_norm), var) for grad, var in self.grads_and_vars]
self.train_op = self.optimizer.apply_gradients(self.grads_and_vars)BiLSTMModel.__perform_optimization = classmethod(perform_optimization)Congratulations! You have specified all the parts of your network. You may have noticed, that we didn’t deal with any real data yet, so what you have written is just recipes on how the network should function.
Now we will put them to the constructor of our Bi-LSTM class to use it in the next section.
def init_model(self, vocabulary_size, n_tags, embedding_dim, n_hidden_rnn, PAD_index):
self.__declare_placeholders()
self.__build_layers(vocabulary_size, embedding_dim, n_hidden_rnn, n_tags)
self.__compute_predictions()
self.__compute_loss(n_tags, PAD_index)
self.__perform_optimization()BiLSTMModel.__init__ = classmethod(init_model)Train the network and predict tags
Session.run is a point which initiates computations in the graph that we have defined. To train the network, we need to compute self.train_op, which was declared in perform_optimization. To predict tags, we just need to compute self.predictions. Anyway, we need to feed actual data through the placeholders that we defined before.
def train_on_batch(self, session, x_batch, y_batch, lengths, learning_rate, dropout_keep_probability):
feed_dict = {self.input_batch: x_batch,
self.ground_truth_tags: y_batch,
self.learning_rate_ph: learning_rate,
self.dropout_ph: dropout_keep_probability,
self.lengths: lengths}
session.run(self.train_op, feed_dict=feed_dict)BiLSTMModel.train_on_batch = classmethod(train_on_batch)Implement the function predict_for_batch by initializing feed_dict with input x_batch and lengths and running the session for self.predictions.
def predict_for_batch(self, session, x_batch, lengths):
######################################
######### YOUR CODE HERE #############
######################################
feed_dict = { self.input_batch:x_batch,
self.lengths : lengths}
predictions = session.run(self.predictions, feed_dict = feed_dict)
return predictionsBiLSTMModel.predict_for_batch = classmethod(predict_for_batch)We finished with necessary methods of our BiLSTMModel model and almost ready to start experimenting.
Evaluation
To simplify the evaluation process we provide two functions for you:
- predict_tags: uses a model to get predictions and transforms indices to tokens and tags;
- eval_conll: calculates precision, recall and F1 for the results.
from evaluation import precision_recall_f1def predict_tags(model, session, token_idxs_batch, lengths):
"""Performs predictions and transforms indices to tokens and tags."""
tag_idxs_batch = model.predict_for_batch(session, token_idxs_batch, lengths)
tags_batch, tokens_batch = [], []
for tag_idxs, token_idxs in zip(tag_idxs_batch, token_idxs_batch):
tags, tokens = [], []
for tag_idx, token_idx in zip(tag_idxs, token_idxs):
tags.append(idx2tag[tag_idx])
tokens.append(idx2token[token_idx])
tags_batch.append(tags)
tokens_batch.append(tokens)
return tags_batch, tokens_batch
def eval_conll(model, session, tokens, tags, short_report=True):
"""Computes NER quality measures using CONLL shared task script."""
y_true, y_pred = [], []
for x_batch, y_batch, lengths in batches_generator(1, tokens, tags):
tags_batch, tokens_batch = predict_tags(model, session, x_batch, lengths)
if len(x_batch[0]) != len(tags_batch[0]):
raise Exception("Incorrect length of prediction for the input, "
"expected length: %i, got: %i" % (len(x_batch[0]), len(tags_batch[0])))
predicted_tags = []
ground_truth_tags = []
for gt_tag_idx, pred_tag, token in zip(y_batch[0], tags_batch[0], tokens_batch[0]):
if token != '' :
ground_truth_tags.append(idx2tag[gt_tag_idx])
predicted_tags.append(pred_tag)
# We extend every prediction and ground truth sequence with 'O' tag
# to indicate a possible end of entity.
y_true.extend(ground_truth_tags + ['O'])
y_pred.extend(predicted_tags + ['O'])
results = precision_recall_f1(y_true, y_pred, print_results=True, short_report=short_report)
return resultsRun your experiment
Create BiLSTMModel model with the following parameters:
- vocabulary_size — number of tokens;
- n_tags — number of tags;
- embedding_dim — dimension of embeddings, recommended value: 200;
- n_hidden_rnn — size of hidden layers for RNN, recommended value: 200;
- PAD_index — an index of the padding token (
Set hyperparameters. You might want to start with the following recommended values:
- batch_size: 32;
- 4 epochs;
- starting value of learning_rate: 0.005
- learning_rate_decay: a square root of 2;
- dropout_keep_probability: try several values: 0.1, 0.5, 0.9.
However, feel free to conduct more experiments to tune hyperparameters and earn extra points for the assignment.
tf.reset_default_graph()
model = BiLSTMModel(20505, 21, 200, 200, token2idx['' ])######### YOUR CODE HERE #############
batch_size = 32######### YOUR CODE HERE #############
n_epochs = 4######### YOUR CODE HERE #############
learning_rate = 0.005######### YOUR CODE HERE #############
learning_rate_decay = 1.414######### YOUR CODE HERE #############
dropout_keep_probability = 0.5######### YOUR CODE HERE #############If you got an error “Tensor conversion requested dtype float64 for Tensor with dtype float32” in this point, check if there are variables without dtype initialised. Set the value of dtype equals to tf.float32 for such variables.
Finally, we are ready to run the training!
sess = tf.Session()
sess.run(tf.global_variables_initializer())
print('Start training... \n')
for epoch in range(n_epochs):
# For each epoch evaluate the model on train and validation data
print('-' * 20 + ' Epoch {} '.format(epoch+1) + 'of {} '.format(n_epochs) + '-' * 20)
print('Train data evaluation:')
eval_conll(model, sess, train_tokens, train_tags, short_report=True)
print('Validation data evaluation:')
eval_conll(model, sess, validation_tokens, validation_tags, short_report=True)
# Train the model
for x_batch, y_batch, lengths in batches_generator(batch_size, train_tokens, train_tags):
model.train_on_batch(sess, x_batch, y_batch, lengths, learning_rate, dropout_keep_probability)
# Decaying the learning rate
learning_rate = learning_rate / learning_rate_decay
print('...training finished.')Start training...
-------------------- Epoch 1 of 4 --------------------
Train data evaluation:
processed 105778 tokens with 4489 phrases; found: 77141 phrases; correct: 154.
precision: 0.20%; recall: 3.43%; F1: 0.38
Validation data evaluation:
processed 12836 tokens with 537 phrases; found: 9302 phrases; correct: 21.
precision: 0.23%; recall: 3.91%; F1: 0.43
-------------------- Epoch 2 of 4 --------------------
Train data evaluation:
processed 105778 tokens with 4489 phrases; found: 1149 phrases; correct: 340.
precision: 29.59%; recall: 7.57%; F1: 12.06
Validation data evaluation:
processed 12836 tokens with 537 phrases; found: 117 phrases; correct: 30.
precision: 25.64%; recall: 5.59%; F1: 9.17
-------------------- Epoch 3 of 4 --------------------
Train data evaluation:
processed 105778 tokens with 4489 phrases; found: 4461 phrases; correct: 1736.
precision: 38.92%; recall: 38.67%; F1: 38.79
Validation data evaluation:
processed 12836 tokens with 537 phrases; found: 318 phrases; correct: 125.
precision: 39.31%; recall: 23.28%; F1: 29.24
-------------------- Epoch 4 of 4 --------------------
Train data evaluation:
processed 105778 tokens with 4489 phrases; found: 4656 phrases; correct: 2810.
precision: 60.35%; recall: 62.60%; F1: 61.45
Validation data evaluation:
processed 12836 tokens with 537 phrases; found: 363 phrases; correct: 166.
precision: 45.73%; recall: 30.91%; F1: 36.89
...training finished.
Now let us see full quality reports for the final model on train, validation, and test sets. To give you a hint whether you have implemented everything correctly, you might expect F-score about 40% on the validation set.
The output of the cell below (as well as the output of all the other cells) should be present in the notebook for peer2peer review!
print('-' * 20 + ' Train set quality: ' + '-' * 20)
train_results = eval_conll(model, sess, train_tokens, train_tags, short_report=False)
print('-' * 20 + ' Validation set quality: ' + '-' * 20)
validation_results = eval_conll(model,sess,validation_tokens, validation_tags, short_report =False) ######### YOUR CODE HERE #############
print('-' * 20 + ' Test set quality: ' + '-' * 20)
test_results = eval_conll(model,sess,test_tokens, test_tags,short_report=False)######### YOUR CODE HERE #############-------------------- Train set quality: --------------------
processed 105778 tokens with 4489 phrases; found: 4703 phrases; correct: 3477.
precision: 73.93%; recall: 77.46%; F1: 75.65
company: precision: 85.95%; recall: 89.42%; F1: 87.65; predicted: 669
facility: precision: 72.46%; recall: 70.38%; F1: 71.41; predicted: 305
geo-loc: precision: 79.46%; recall: 94.38%; F1: 86.28; predicted: 1183
movie: precision: 38.46%; recall: 7.35%; F1: 12.35; predicted: 13
musicartist: precision: 42.62%; recall: 33.62%; F1: 37.59; predicted: 183
other: precision: 73.72%; recall: 79.66%; F1: 76.57; predicted: 818
person: precision: 73.14%; recall: 93.12%; F1: 81.93; predicted: 1128
product: precision: 53.39%; recall: 59.43%; F1: 56.25; predicted: 354
sportsteam: precision: 82.00%; recall: 18.89%; F1: 30.71; predicted: 50
tvshow: precision: 0.00%; recall: 0.00%; F1: 0.00; predicted: 0
-------------------- Validation set quality: --------------------
processed 12836 tokens with 537 phrases; found: 405 phrases; correct: 183.
precision: 45.19%; recall: 34.08%; F1: 38.85
company: precision: 65.48%; recall: 52.88%; F1: 58.51; predicted: 84
facility: precision: 47.37%; recall: 26.47%; F1: 33.96; predicted: 19
geo-loc: precision: 61.22%; recall: 53.10%; F1: 56.87; predicted: 98
movie: precision: 0.00%; recall: 0.00%; F1: 0.00; predicted: 0
musicartist: precision: 0.00%; recall: 0.00%; F1: 0.00; predicted: 4
other: precision: 30.00%; recall: 29.63%; F1: 29.81; predicted: 80
person: precision: 36.78%; recall: 28.57%; F1: 32.16; predicted: 87
product: precision: 6.45%; recall: 5.88%; F1: 6.15; predicted: 31
sportsteam: precision: 50.00%; recall: 5.00%; F1: 9.09; predicted: 2
tvshow: precision: 0.00%; recall: 0.00%; F1: 0.00; predicted: 0
-------------------- Test set quality: --------------------
processed 13258 tokens with 604 phrases; found: 454 phrases; correct: 225.
precision: 49.56%; recall: 37.25%; F1: 42.53
company: precision: 66.07%; recall: 44.05%; F1: 52.86; predicted: 56
facility: precision: 48.28%; recall: 29.79%; F1: 36.84; predicted: 29
geo-loc: precision: 75.00%; recall: 56.36%; F1: 64.36; predicted: 124
movie: precision: 0.00%; recall: 0.00%; F1: 0.00; predicted: 0
musicartist: precision: 40.00%; recall: 7.41%; F1: 12.50; predicted: 5
other: precision: 33.03%; recall: 34.95%; F1: 33.96; predicted: 109
person: precision: 44.44%; recall: 38.46%; F1: 41.24; predicted: 90
product: precision: 5.13%; recall: 7.14%; F1: 5.97; predicted: 39
sportsteam: precision: 50.00%; recall: 3.23%; F1: 6.06; predicted: 2
tvshow: precision: 0.00%; recall: 0.00%; F1: 0.00; predicted: 0
Conclusions
Could we say that our model is state of the art and the results are acceptable for the task? Definately, we can say so. Nowadays, Bi-LSTM is one of the state of the art approaches for solving NER problem and it outperforms other classical methods. Despite the fact that we used small training corpora (in comparison with usual sizes of corpora in Deep Learning), our results are quite good. In addition, in this task there are many possible named entities and for some of them we have only several dozens of trainig examples, which is definately small. However, the implemented model outperforms classical CRFs for this task. Even better results could be obtained by some combinations of several types of methods, e.g. see this paper if you are interested.