無所不能的Embedding4 - Doc2vec第二彈[skip-thought & tf-Seq2Seq源碼解析]
前一章Doc2Vec裏提到，其實Doc2Vec只是經過加入Doc_id捕捉了文本的主題信息，並無真正考慮語序以及上下文語義，n-gram只能在局部解決這一問題，那麼還有別的解決方案麼？依舊是通用文本向量，skip-thought嘗試應用encoder-decoder來學習包含上下文信息和語序的句子向量。魔改後的實現能夠看這裏( ´▽｀) github-DSXiangLi-Embedding-skip_thoughthtml
Skip-Thought模型分析
Skip-Thought顧名思義是沿用了skip-gram的路子,不熟悉的童鞋看這裏 無所不能的Embedding1 - Word2vec模型詳解&代碼實現python
skip-gram是用中間詞來預測周圍單詞，skip-Thought是用中間句子來預測前一個句子和後一個句子，模型思路就是這麼簡單粗暴，具體實現就涉及到句子的信息要如何提取，以及loss function的選擇。做者選擇了encoder-decoder來提取句子信息，用翻譯模型經常使用的log-perplrexity做爲loss。git
這裏想提一句不一樣模型，在不一樣的樣本上，訓練出的文本向量所包含的信息是不一樣的。例如word2vec的假設就是context（windo_size內周圍詞）類似的單詞更類似（向量空間距離更近）。skip-thought做者對於文本向量的假設是：能更好reconstruct先後句子的信息，就是當前句子的所含信息，換言以前後句子類似的句子，文本向量的空間距離更近。github
第一次讀到這裏感受哇make perfect sense！可越琢磨越覺着這個task有些迷幻，word2vec skip-gram能夠這麼搞，是由於給定中間詞window_size內的單詞選擇是相對有限的。你給我個句子就讓我精準預測先後句子的每個詞，這能收斂？you what?! 不着急後面彷佛有反轉~app
Encoder部分負責提取中間句子的信息生成定長向量output_state，Decoder則基於ouput_state進行迭代生成前（後）句子。Encoder-Decoder支持任意記憶單元，這裏做者選擇了GRU-GRU。框架
簡單回顧下GRU Cell，GRU有兩個Gate，從兩個角度衡量歷史sequence信息和當前token的相關程度，\(\Gamma_r\)控制多少歷史信息參與state的從新計算是reset gate，\(\Gamma_u\)控制多少歷史信息直接進入當前state是update gate，這裏安利一篇博客 Illustrated Guide to LSTM’s and GRU’s: A step by step explanationdom
Encoder部分通過GRU把長度爲T的sequence信息壓縮到hidden_size的\(h^{<T>}\)，這裏\(h^{<T>}\)也是最終skip-thought爲每個句子生成的通用向量表達。ide
Decoder部分基於\(h^{<T>}\)向前預測下一個/上一個句子中的每個單詞。Decoder比Encoder略複雜，在於訓練階段和預測階段對於input的處理存在差別。函數
訓練階段使用了100%的Teacher Forcing，每一個cell的輸入除了上一個cell的hidden state，還有預測句子中前一個真實token對應的embedding，如圖工具
而在預測階段真實序列未知，所以會轉而使用前一個cell的output來預測前一個token，再用預測token的embedding做爲輸入，如圖
對於翻譯模型來講，在訓練階段使用TeacherForcing的好處是能夠加速模型收斂，避免向前迭代預測的偏差進一步放大。壞處天然是訓練和預測時decoder的表現存在差別（Exposure Bias)，以及預測時decode的output會受到訓練樣本的約束。這裏最經常使用的解決方案是Scheduled Sampling, 簡單來講就是在訓練階段有P的機率輸入用teacher forcing，1-P的機率用預測output。可是！skip-thought並無使用這個解決方案，爲啥嘞？反轉來了V(^_^)V
看到無採樣的teacherforcing這裏，前面的迷惑已然解答。其實skip-thought並不僅是使用中間句子來預測先後句子，而是基於中間句子的ouput_state，用先後句子中T-1前的單詞來預測第T個單詞（感受和missing imputation只有一步之遙)。encoder部分只須要在output_state中最大程度的提取句子信息，保證在不一樣的先後句子上output state均可以generalize。至於decoder的預測部分效果如何模型並不關心，由於skip-thought的預測輸出就是encoder部分的output state，因此天然是不須要使用Scheduled Sampling
skip-thought的Decoder還有兩點特殊：
- 前/後句子用兩個decoder來訓練，兩個decoder除了word embedding共享以外，參數獨立
- encoder state不僅做爲decoder的initial_state，而是直接傳入decoder的每個cell，起到相似residual/connditional的做用，避免encoder的信息傳着傳着給傳沒了。這裏感受不用attention而是直接傳入outpu_state也是爲了保證這個output_state能最大程度的學到用來reconstruct先後句子的信息
loss部分做者用了語言模型的log-perplexity把先後句子的loss加總獲得loss function
論文比較有意思的一個點還有vocabulary expansion，就是如何把word embedding擴展到訓練集以外。做者嘗試用linear-mapping的方式學習word2vec和skip-thought裏面word-embedding的映射關係，就是找到word2vec和skip-thought交集的word, 對他們的embedding作regression $ X_{word2vec} \sim W \cdot X_{skipthought} $，這樣對樣本外可是word2vec內的單詞直接用W映射就能獲得skip-thougt的詞向量
這裏直接用word2vec/glove的word embedding來初始化skip-thougt的詞向量是否是更好？在後面的模型實現裏我就是直接用word2vec來初始化了embedding, word2vec以外詞用random.uniform(-0.1,0.1)來初始化
最終在生成文本向量的時候，做者給出了幾種方案，遵循大力必定出奇蹟的原則天然方案3效果更好
- 2400-dim的uni-skip, 就是以上encoder生成的output state
- 兩個1200-dim的bi-skip向量拼接獲得2400-dim的向量，是在處理訓練樣本時一個用正常語序訓練，一個reverse語序訓練
- 把上面兩個2400-dim拼接獲得4800-dim
skip-thought 模型實現
這裏有點任性的對論文作了魔改。。。部分細節和論文已經天差地別，能夠拿來了解encoder-decoder的實現但不保證徹底reproduce skip-thought的結果。。。如下只保留代碼核心部分，完整代碼在 github-DSXiangLi-Embedding-skip_thought。 這裏用了tensorflow seq2seq的框架，不熟悉的童鞋能夠先看後面seq2seq的代碼解析~
dataset
論文中是\((s_{i-1}, s_i, s_{i+1})\)做爲一組樣本，其中\(s_i\)是encoder source，\(s_{i-1}\)和\(s_{i+1}\)是decoder target，這裏我直接處理成\((s_i,s_{i-1})\),\((s_i,s_{i+1})\)兩組樣本。
其中encoder source不須要多作處理，可是decoder source在Train和Eval時須要在sequence先後加入start和end_token標記序列的開始和結束，在Predict時須要加入start_token標記開始。最後經過word_table把token映射到token_id，再Padding到相同長度就齊活。
這裏在Dataset的部分加入了獲取word2vec embedding的部分, word2vec之外的單詞默認random.uniform(-0.1,0.1)
class SkipThoughtDataset(BaseDataset):
def __init__(self, data_file, dict_file, epochs, batch_size, buffer_size, min_count, max_count,
special_token, max_len):
...
def parse_example(self, line, prepend, append):
features = {}
tokens = tf.string_split([tf.string_strip(line)]).values
if prepend:
tokens = tf.concat([[self.special_token.SEQ_START], tokens], 0)
if append:
tokens = tf.concat([tokens, [self.special_token.SEQ_END]], 0)
features['tokens'] = tokens
features['seq_len'] = tf.size(tokens)
return features
...
def make_source_dataset(self, file_path, data_type, is_predict, word_table_func):
prepend, append = self.prepend_append_logic(data_type, is_predict)
dataset = tf.data.TextLineDataset(file_path).\
map(lambda x: self.parse_example(x, prepend, append), num_parallel_calls=tf.data.experimental.AUTOTUNE).\
map(lambda x: word_table_func(x), num_parallel_calls=tf.data.experimental.AUTOTUNE)
return dataset
def build_dataset(self, is_predict=0):
def input_fn():
word_table_func = self.word_table_lookup(self.build_wordtable())
_ = self.build_tokentable() # initialize here to ensure lookup table is in the same graph
encoder_source = self.make_source_dataset(self.data_file['encoder'], 'encoder', is_predict, word_table_func)
decoder_source = self.make_source_dataset(self.data_file['decoder'], 'decoder', is_predict, word_table_func)
dataset = tf.data.Dataset.zip((encoder_source, decoder_source)).\
filter(self.sample_filter_logic)
if not is_predict:
dataset = dataset.\
repeat(self.epochs)
dataset = dataset. \
padded_batch( batch_size=self.batch_size,
padded_shapes=self.padded_shape,
padding_values=self.padding_values,
drop_remainder=True ). \
prefetch( tf.data.experimental.AUTOTUNE )
else:
dataset = dataset.batch(1)
return dataset
return input_fn
def load_pretrain_embedding(self):
if self.embedding is None:
word_vector = gensim.downloader.load(PretrainModel)
embedding = []
for i in self._dictionary.keys():
try:
embedding.append( word_vector.get_vector( i ) )
except KeyError:
embedding.append( np.random.uniform(low=-0.1, high=0.1, size=300))
self.embedding = np.array(embedding, dtype=np.float32)
return self.embedding
Encoder-Decoder
Encoder的部分很常規，確認cell類型，而後通過dynamic_rnn迭代，輸出output和state
def gru_encoder(input_emb, input_len, params):
gru_cell = build_rnn_cell('gru', params)
# state: batch_size * hidden_size, output: batch_size * max_len * hidden_size
output, state = tf.nn.dynamic_rnn(
cell=gru_cell, # one rnn units
inputs=input_emb, # batch_size * max_len * feature_size
sequence_length=input_len, # batch_size * seq_len
initial_state=None,
dtype=params['dtype'],
time_major=False # whether reshape max_length to first dim
)
return ENCODER_OUTPUT(output=output, state=state)
Decoder的部分能夠分紅helper, decoder, 以及最終dynamic_decode的部分。比較容易踩坑的有幾個點
- dynamic_decode部分的max_iteration在訓練時能夠不設定，train_helper會基於seq_len判斷finish，在Eval時由於用GreedyEmbeddingHelper因此須要手動傳入pad_len，在predict部分只需給定最大max_iter保證預測必定會中止就好
- Decoder的output layer必需要有, 由於須要作hidden_size -> vocab_size(softmax)的轉換，用於預測每一個cell的token輸出
- 前面dataset的decoder_source咱們在先後都加了開始/中止token，訓練時須要移除最後一個token(對於trainHelper能夠改inputs也能夠改sequence_length), 這樣在計算loss時能夠和移除第一個token的target對齊。
這裏針對上面提到的把encoder的output_state直接傳入每一個decoder cell作了實現，直接把encoder state和embedding input作了拼接做爲輸入。
def get_helper(encoder_output, input_emb, input_len, batch_size, embedding, mode, params):
if mode == tf.estimator.ModeKeys.TRAIN:
if params['conditional']:
# conditional train helper with encoder output state as direct input
# Reshape encoder state as auxiliary input: 1* batch_size * hidden -> batch_size * max_len * hidden
decoder_length = tf.shape(input_emb)[1]
state_shape = tf.shape(encoder_output.state)
encoder_state = tf.tile(tf.reshape(encoder_output.state, [state_shape[1],
state_shape[0],
state_shape[2]]),
[1, decoder_length, 1])
input_emb = tf.concat([encoder_state, input_emb], axis=-1)
helper = seq2seq.TrainingHelper( inputs=input_emb, # batch_size * max_len-1 * emb_size
sequence_length=input_len-1, # exclude last token
time_major=False,
name='training_helper' )
else:
helper = seq2seq.GreedyEmbeddingHelper( embedding=embedding_func( embedding ),
start_tokens=tf.fill([batch_size], params['start_token']),
end_token=params['end_token'] )
return helper
def get_decoder(decoder_cell, encoder_output, input_emb, input_len, embedding, output_layer, mode, params):
batch_size = tf.shape(encoder_output.output)[0]
if params['beam_width'] >1 :
# If beam search multiple prediction are uesd at each time step
decoder = seq2seq.BeamSearchDecoder( cell=decoder_cell,
embedding=embedding_func( embedding ),
initial_state=encoder_output,
beam_width=params['beam_width'],
start_tokens=tf.fill([batch_size], params['start_token']),
end_token=params['end_token'],
output_layer=output_layer )
else:
helper = get_helper(encoder_output, input_emb, input_len, batch_size, embedding, mode, params)
decoder = seq2seq.BasicDecoder( cell=decoder_cell,
helper=helper,
initial_state=encoder_output.state,
output_layer=output_layer )
return decoder
def gru_decoder(encoder_output, input_emb, input_len, embedding, params, mode):
gru_cell = build_rnn_cell( 'gru', params )
if mode == tf.estimator.ModeKeys.TRAIN:
max_iteration = None
elif mode == tf.estimator.ModeKeys.EVAL:
max_iteration = tf.reduce_max(input_len) # decode max sequence length(=padded_length)in EVAL
else:
max_iteration = params['max_decode_iter'] # decode pre-defined max_decode iter in predict
output_layer=tf.layers.Dense(units=params['vocab_size']) # used for infer helper sample or train loss calculation
decoder = get_decoder(gru_cell, encoder_output, input_emb, input_len, embedding, output_layer, mode, params)
output, state, seq_len = seq2seq.dynamic_decode(decoder=decoder,
output_time_major=False,
impute_finished=True,
maximum_iterations=max_iteration)
return DECODER_OUTPUT(output=output, state = state, seq_len=seq_len)
loss
loss這了本身實現的一版sequence_loss，把計算loss和按不一樣維度聚合拆成了兩塊。感受tf.sequence_loss只針對train,對eval的部分並不友好，由於trainHelper能夠保證source和target的長度一致，可是infer時調用GreedyEmbeddingHelper是沒法保證輸出長度的(不知道是否是我哪裏理解錯了，若是是請大神指正(o^^o)), 因此對eval部分也作了特殊處理。
def sequence_loss(logits, target, mask, mode):
with tf.variable_scope('Sequence_loss_matrix'):
n_class = tf.shape(logits)[2]
decode_len = tf.shape(logits)[1] # used for infer only, max_len is determined by decoder
logits = tf.reshape(logits, [-1, n_class])
if mode == tf.estimator.ModeKeys.TRAIN:
# In train, target
target = tf.reshape(target[:, 1:], [-1]) # (batch * (padded_len-1)) * 1
elif mode == tf.estimator.ModeKeys.EVAL:
# In eval, target has paded_len, logits have decode_len
target = tf.reshape(target[:, : decode_len], [-1]) # batch * (decode_len) *1
else:
raise Exception('sequence loss is only used in train or eval, not in pure prediction')
loss_mat = tf.nn.sparse_softmax_cross_entropy_with_logits(labels = target, logits = logits)
loss_mat = tf.multiply(loss_mat, tf.reshape(mask, [-1])) # apply padded mask on output loss
return loss_mat
def agg_sequence_loss(loss_mat, mask, axis):
with tf.variable_scope('Loss_{}'.format(axis)):
if axis == 'scaler':
loss = tf.reduce_sum(loss_mat)
n_sample = tf.reduce_sum(mask)
loss = loss/n_sample
else:
loss_mat = tf.reshape(loss_mat, tf.shape(mask)) # (batch_size * max_len) * 1-> batch_size * max_len
if axis == 'batch':
loss = tf.reduce_sum(loss_mat, axis=1) # batch
n_sample = tf.reduce_sum(mask, axis=1) # batch
loss = tf.math.divide_no_nan(loss, n_sample) # batch
elif axis == 'time':
loss = tf.reduce_sum(loss_mat, axis=0) # max_len
n_sample = tf.reduce_sum(mask, axis=0) # max_len
loss = tf.math.divide_no_nan(loss, n_sample) # max_len
else:
raise Exception('Only scaler/batch/time are supported in axis param')
return loss
model
encoder, decoder, loss都ready，拼一塊就齊活了, 這裏embedding咱們用了前面加載的word2vec來進行初始化。
class QuickThought(object):
def __init__(self, params):
self.params = params
self.init()
def init(self):
with tf.variable_scope('embedding', reuse=tf.AUTO_REUSE):
self.embedding = tf.get_variable(dtype = self.params['dtype'],
initializer=tf.constant(self.params['pretrain_embedding']),
name='word_embedding' )
add_layer_summary(self.embedding.name, self.embedding)
def build_model(self, features, labels, mode):
encoder_output = self._encode(features)
decoder_output = self._decode(encoder_output, labels, mode )
loss_output = self.compute_loss( decoder_output, labels, mode )
...
def _encode(self, features):
with tf.variable_scope('encoding'):
encoder = ENCODER_FAMILY[self.params['encoder_type']]
seq_emb_input = tf.nn.embedding_lookup(self.embedding, features['tokens']) # batch_size * max_len * emb_size
encoder_output = encoder(seq_emb_input, features['seq_len'], self.params) # batch_size
return encoder_output
def _decode(self, encoder_output, labels, mode):
with tf.variable_scope('decoding'):
decoder = DECODER_FAMILY[self.params['decoder_type']]
if mode == tf.estimator.ModeKeys.TRAIN:
seq_emb_output = tf.nn.embedding_lookup(self.embedding, labels['tokens']) # batch_size * max_len * emb_size
input_len = labels['seq_len']
elif mode == tf.estimator.ModeKeys.EVAL:
seq_emb_output = None
input_len = labels['seq_len']
else:
seq_emb_output = None
input_len = None
decoder_output = decoder(encoder_output, seq_emb_output, input_len,\
self.embedding, self.params, mode)
return decoder_output
def compute_loss(self, decoder_output, labels, mode):
with tf.variable_scope('compute_loss'):
mask = sequence_mask(decoder_output, labels, self.params, mode)
loss_mat = sequence_loss(logits=decoder_output.output.rnn_output,
target=labels['tokens'],
mask=mask,
mode=mode)
loss = []
for axis in ['scaler', 'batch', 'time']:
loss.append(agg_sequence_loss(loss_mat, mask, axis))
return SEQ_LOSS_OUTPUT(loss_id=loss_mat,
loss_scaler=loss[0],
loss_per_batch=loss[1],
loss_per_time=loss[2])
tf.seq2seq 代碼解析
稀裏糊塗開始用seq2seq，結果盯着shape mismatch的報錯險些看到地老天荒，索性咱老老實實看一遍tf的實現, 如下代碼只保留了核心部分，完整的官方代碼在這裏喲 tf.seq2seq.contrib
Encoding
Encoding部分就是一個dynamic_rnn，先看下輸入
- cell：任意類型記憶單元rnn,gru, lstm
- inputs：rnn輸入，通常是[batch_size, max_len, emb_siz] 也就是padding的序列token,通過embedding映射以後做爲輸入
- sequence_length: 真實序列長度(不包含padding)，用於判斷序列是遍歷完
- initial_state: encoder最初state，None則默認是zero_state
- dtype: output數據類型，建議全局設置統一數據類型，否則會有各類mismatch，不要問我是怎麼知道的>.<
- parallel_iteration：內存換速度，沒有上下文依賴的op進行並行計算
- time_major：若是你的輸入數據是[max_len, batch_size，emb_siz]則爲True，通常爲False在dynamic_rnn內部再作reshape。
dynamic_rnn主函數其實只作了輸入/輸出數據的處理部分，包括
- reshape_input：對應上面time_major=False, 把輸入數據從[batch_size, max_len, emb_siz]轉換爲[max_len, batch_size，emb_siz]
- inital_state: 默認是batch size的zero_state
- reshape_output: output輸出是[max_len, batch_size，hidden_siz]轉換爲[batch_size, max_len, hidden_size]
def dynamic_rnn(cell, inputs, sequence_length=None, initial_state=None,
dtype=None, parallel_iterations=None, swap_memory=False,
time_major=False, scope=None):
flat_input = nest.flatten(inputs)
if not time_major:
flat_input = [ops.convert_to_tensor(input_) for input_ in flat_input]
flat_input = tuple(_transpose_batch_time(input_) for input_ in flat_input)
batch_size = _best_effort_input_batch_size(flat_input)
state = cell.zero_state(batch_size, dtype)
inputs = nest.pack_sequence_as(structure=inputs, flat_sequence=flat_input)
(outputs, final_state) = _dynamic_rnn_loop(
cell,
inputs,
state,
parallel_iterations=parallel_iterations,
swap_memory=swap_memory,
sequence_length=sequence_length,
dtype=dtype)
if not time_major:
# (T,B,D) => (B,T,D)
outputs = nest.map_structure(_transpose_batch_time, outputs)
return (outputs, final_state)
核心計算部分都在_dynamic_rnn_loop，是一個while_loop, 因此須要定義三要素[loop_var, body, condition]
- loop_var：(time, output_ta, state)
- time:遍歷到第幾個token
- output_ta: 每一個cell的輸出，padding部分是zero-output
- state: 最後一個cell的輸出，對於padding的序列，只計算到最後一個真實token，以後的state是直接copy through
這裏output_ta的shape是(batch, max_len, hidden_units), 對於rnn和GRU，state就是最後一個output, 那shape天然是(1, batch, hidden_units), 但LSTM是有兩個hidden state的，一個用於向前傳遞信息一個用於輸出，因此這裏state的shape會是（2, batch, hidden_units)
- body
loop的核心計算部分是lambda: cell(input_t, state)，也就是相應記憶單元的計算。當sequence_length給定時，_rnn_step的額外操做實際上是對已經遍歷完的序列直接copy through（zero_output, last_state)
def _time_step(time, output_ta_t, state):
input_t = tuple(ta.read(time) for ta in input_ta)
input_t = nest.pack_sequence_as(structure=inputs, flat_sequence=input_t)
call_cell = lambda: cell(input_t, state)
if sequence_length is not None:
(output, new_state) = _rnn_step(
time=time,
sequence_length=sequence_length,
min_sequence_length=min_sequence_length,
max_sequence_length=max_sequence_length,
zero_output=zero_output,
state=state,
call_cell=call_cell,
state_size=state_size,
skip_conditionals=True)
else:
(output, new_state) = call_cell()
# Pack state if using state tuples
output = nest.flatten(output)
output_ta_t = tuple(ta.write(time, out) for ta, out in zip(output_ta_t, output))
return (time + 1, output_ta_t, new_state)
- condition
中止loop的條件loop_bound=min(max_sequence_length, max(1,time_steps) , 其中time_step是輸入的max_len維度，也就是padding length, max_sequence_length是輸入batch的最大真實長度，若是是batch_padding這兩個取值應該是同樣的
time_steps = input_shape[0]
if sequence_length is not None:
min_sequence_length = math_ops.reduce_min(sequence_length)
max_sequence_length = math_ops.reduce_max(sequence_length)
else:
max_sequence_length = time_steps
loop_bound = math_ops.minimum(time_steps, math_ops.maximum(1, max_sequence_length))
_, output_final_ta, final_state = control_flow_ops.while_loop(
cond=lambda time, *_: time < loop_bound,
body=_time_step,
loop_vars=(time, output_ta, state),
parallel_iterations=parallel_iterations,
maximum_iterations=time_steps,
swap_memory=swap_memory)
Decoding
Decoding主要有三個組件，Decoder，Helper和dynamic_decode。還有比較特殊獨立出來的BeamSearch和Attention，這兩個後面用到再說
BasicDecoder
BasicDecoder主要接口有2個
- initialize生成decode階段的最初input
- step實現每一步decode的計算,以後被dynamic_decode的while_loop調用
其中initialize拼接了helper的初始化返回再加上initial_state，也就是encoder最後一步的output_state，helper返回的部分咱們放在後面說。
def initialize(self, name=None):
return self._helper.initialize() + (self._initial_state,)
step部分作了以下操做
- 輸入上一步的output, state計算下一步的output，這是Decoder的核心計算
- 若是定義了output_layer，對output作transform，爲啥須要output_layer嘞? 這個看到Helper你就明白了
- sample, next_inputs: 都是調用Helper的接口
- 輸出: BasicDecoderOutput(rnn_output, sample_id), next_state, next_inputs, finished
class BasicDecoderOutput(
collections.namedtuple("BasicDecoderOutput", ("rnn_output", "sample_id"))):
pass
class BasicDecoder(decoder.Decoder):
"""Basic sampling decoder."""
def __init__(self, cell, helper, initial_state, output_layer=None):
def step(self, time, inputs, state, name=None):
with ops.name_scope(name, "BasicDecoderStep", (time, inputs, state)):
cell_outputs, cell_state = self._cell(inputs, state)
if self._output_layer is not None:
cell_outputs = self._output_layer(cell_outputs)
sample_ids = self._helper.sample(
time=time, outputs=cell_outputs, state=cell_state)
(finished, next_inputs, next_state) = self._helper.next_inputs(
time=time,
outputs=cell_outputs,
state=cell_state,
sample_ids=sample_ids)
outputs = BasicDecoderOutput(cell_outputs, sample_ids)
return (outputs, next_state, next_inputs, finished)
這裏發現BasicDecoder的實現只包括了承上的部分，啓下的部分都放在了Helper裏面，下面咱們具體看下Helper的next_input和Sample接口乾了啥
Helper
咱們主要看兩個helper一個用於訓練，一個用於預測，主要實現3個接口
- initialize：生成decode階段的最初input
- sample：生成decode下一步的input id
- next_inputs：生成decode下一步的input
TrainHelper用於訓練，sample接口實際並無用，next_input把sample_id定義爲unused_kwargs.
- initialize返回 (finished, next_inputs)
- finished: 判斷當前batch每一個sequence是否已經遍歷完, sequence_length是不包含padded的實際sequencec長度
- 除非batch裏全部seq_length的長度都是0，不然直接讀取每一個sequence的第一個token做爲decoder的初始輸入
decoder輸入sequence會在預處理時加入start_token標記seq的開始，對應上圖的\(<go>\)標記,同時加入start_token也爲了造成source和target的錯位，作到輸入T-1個字符預測T個字符。例如source是[\(<go>\), I, love, you]，target是[I, love, you, \(<eos>\)]
- next_inputs輸出(finished, next_inputs, state)
- finished: 判斷當前batch每一個sequence是否已經遍歷完, sequence_length是不包含padded的實際sequencec長度
- next_inputs: 訓練時使用Teaching Force，傳入下一個decoder cell的就是前一個位置的實際token embedding，因此這裏next_input直接讀取input sequence的下一個值，若是finished都是True就返回0【其實返回啥都無所謂由於在loss那裏padded的部分會被mask掉】
- state: 這裏是打醬油的，直接pass-throuh
class TrainingHelper(Helper):
def __init__(self, inputs, sequence_length, time_major=False, name=None):
...
def initialize(self, name=None):
with ops.name_scope(name, "TrainingHelperInitialize"):
finished = math_ops.equal(0, self._sequence_length)
all_finished = math_ops.reduce_all(finished)
next_inputs = control_flow_ops.cond(
all_finished, lambda: self._zero_inputs,
lambda: nest.map_structure(lambda inp: inp.read(0), self._input_tas))
return (finished, next_inputs)
def next_inputs(self, time, outputs, state, name=None, **unused_kwargs):
"""next_inputs_fn for TrainingHelper."""
with ops.name_scope(name, "TrainingHelperNextInputs",
[time, outputs, state]):
next_time = time + 1
finished = (next_time >= self._sequence_length)
all_finished = math_ops.reduce_all(finished)
def read_from_ta(inp):
return inp.read(next_time)
next_inputs = control_flow_ops.cond(
all_finished, lambda: self._zero_inputs,
lambda: nest.map_structure(read_from_ta, self._input_tas))
return (finished, next_inputs, state)
GreedyHelper用於預測
-
initialize返回 (finished, next_inputs)
- finished: 都是False，由於infer階段不用判斷input_sequence長度
- next_inputs: 返回start_token對應的embedding，和訓練保持一致
-
sample返回sample_id
負責根據每一個decoder cell的output計算出現機率最大的token，做爲下一個decoder cell的輸入，這裏也是上面提到須要output_layer的緣由，由於須要hidden_size -> vocab_size的變換，才能進一步計算softmax
- next_input
- finished: 若是sequence預測爲end_token則該sequence預測完成，判斷batch裏全部sequence是否預測完成
- next_inputs: 對sample_id作embedding_lookup做爲下一步的輸入，若是finished都是True就返回start_token
- state: 繼續打醬油
class GreedyEmbeddingHelper(Helper):
def __init__(self, embedding, start_tokens, end_token):
self._start_tokens = ops.convert_to_tensor(
start_tokens, dtype=dtypes.int32, name="start_tokens")
self._end_token = ops.convert_to_tensor(
end_token, dtype=dtypes.int32, name="end_token")
self._start_inputs = self._embedding_fn(self._start_tokens)
。。。
def sample(self, time, outputs, state, name=None):
sample_ids = math_ops.cast(
math_ops.argmax(outputs, axis=-1), dtypes.int32)
return sample_ids
def initialize(self, name=None):
finished = array_ops.tile([False], [self._batch_size])
return (finished, self._start_inputs)
def next_inputs(self, time, outputs, state, sample_ids, name=None):
finished = math_ops.equal(sample_ids, self._end_token)
all_finished = math_ops.reduce_all(finished)
next_inputs = control_flow_ops.cond(
all_finished,
lambda: self._start_inputs,
lambda: self._embedding_fn(sample_ids))
return (finished, next_inputs, state)
dynamic_decode
承上啓下的工具都齊活了，要實現對sequence的預測，只剩下一步就是loop，因而有了dynamic_decode，它其實就幹了個while_loop的活，因而仍是loop三兄弟[loop_vars, condition, body]
-
loop_vars=[initial_time, initial_outputs_ta, initial_state, initial_inputs, initial_finished, initial_sequence_lengths]
- initial_finished, initial_inputs, initial_state是上面decoder的initialize返回
- initial_time, initial_sequennce=0
- initial_output_ta是每一個elemennt都是batch * decoder.output_size的不定長TensorArray, 這裏output_size=(rnn_output_size,sample_id_shape),預測返回1個token的sample_id_shape都是scaler, 有output_layer時rnn_output_size=output_layer_size, default= hidden_size
-
condition: 判斷是否全部finished都爲True，都遍歷完則中止loop
-
body: loop的核心計算邏輯
-
step：調用Decoder進行每一步的decode計算
-
finished: 這裏finished主要由三個邏輯判斷（tracks_own_finished我沒用過先忽略了哈哈）其他兩個是：
- helper的next_inputs回傳的finished：trainHelper判斷輸入sequence是否遍歷完，GreedyEmbeddingHelper判斷預測token是否爲end_token。
- max_iteration：只用於預測，爲了不預測token一直不是end_token致使預測無限循環下去，設置一個最大預測長度，訓練時max_iteraion應該爲空
-
sequence_length: 記錄實際預測sequence長度，沒有finished的sequence+1
-
impute_finished: 若是sequence已遍歷完, 後面的output補0，後面的state再也不計算直接pass through當前state
-
def body(time, outputs_ta, state, inputs, finished, sequence_lengths):
(next_outputs, decoder_state, next_inputs,
decoder_finished) = decoder.step(time, inputs, state)
if maximum_iterations is not None:
next_finished = math_ops.logical_or(
next_finished, time + 1 >= maximum_iterations)
next_sequence_lengths = array_ops.where(
math_ops.logical_and(math_ops.logical_not(finished), next_finished),
array_ops.fill(array_ops.shape(sequence_lengths), time + 1),
sequence_lengths)
# Zero out output values past finish
if impute_finished:
emit = nest.map_structure(
lambda out, zero: array_ops.where(finished, zero, out),
next_outputs,
zero_outputs)
else:
emit = next_outputs
# Copy through states past finish
def _maybe_copy_state(new, cur):
# TensorArrays and scalar states get passed through.
if isinstance(cur, tensor_array_ops.TensorArray):
pass_through = True
else:
new.set_shape(cur.shape)
pass_through = (new.shape.ndims == 0)
return new if pass_through else array_ops.where(finished, cur, new)
if impute_finished:
next_state = nest.map_structure(
_maybe_copy_state, decoder_state, state)
else:
next_state = decoder_state
outputs_ta = nest.map_structure(lambda ta, out: ta.write(time, out),
outputs_ta, emit)
return (time + 1, outputs_ta, next_state, next_inputs, next_finished,
next_sequence_lengths)
歡迎留言吐槽以及評論喲～
Reference
skip-thought
- Ryan Kiros, yukun Zhu. 2015. SKip-Thought Vectors
- Lajanugen logeswaran, Honglak Lee. 2018. An Efficient Framework for Learning Sentence Representations.
- https://zhuanlan.zhihu.com/p/64342563
- https://towardsdatascience.com/document-embedding-techniques-fed3e7a6a25d
- https://towardsdatascience.com/the-best-document-similarity-algorithm-in-2020-a-beginners-guide-a01b9ef8cf05
- https://blog.floydhub.com/when-the-best-nlp-model-is-not-the-best-choice/
- https://towardsdatascience.com/document-embedding-techniques-fed3e7a6a25d
- https://zhuanlan.zhihu.com/p/72575806
- Scheduled Sampling for Sequence Prediction with Recurrent Neural Networks