Keras嵌入层中的mask_zero如何工作?

cra*_*cat 12 python machine-learning keras word-embedding

我以为mask_zero=True当输入值为0时将输出0,因此以下各层可能会跳过计算或其他操作。

如何mask_zero运作?

例:

data_in = np.array([
  [1, 2, 0, 0]
])
data_in.shape
>>> (1, 4)

# model
x = Input(shape=(4,))
e = Embedding(5, 5, mask_zero=True)(x)

m = Model(inputs=x, outputs=e)
p = m.predict(data_in)
print(p.shape)
print(p)
Run Code Online (Sandbox Code Playgroud)

实际输出为:(数字是随机的)

(1, 4, 5)
[[[ 0.02499047  0.04617121  0.01586803  0.0338897   0.009652  ]
  [ 0.04782704 -0.04035913 -0.0341589   0.03020919 -0.01157228]
  [ 0.00451764 -0.01433611  0.02606953  0.00328832  0.02650392]
  [ 0.00451764 -0.01433611  0.02606953  0.00328832  0.02650392]]]
Run Code Online (Sandbox Code Playgroud)

但是,我认为输出将是:

[[[ 0.02499047  0.04617121  0.01586803  0.0338897   0.009652  ]
  [ 0.04782704 -0.04035913 -0.0341589   0.03020919 -0.01157228]
  [ 0 0 0 0 0]
  [ 0 0 0 0 0]]]
Run Code Online (Sandbox Code Playgroud)

tod*_*day 18

实际上,设置mask_zero=True嵌入层不会导致返回零向量。而是,嵌入层的行为不会改变,它将返回索引为零的嵌入向量。您可以通过检查“嵌入”层权重来确认这一点(即在您提到的示例中为m.layers[0].get_weights())。取而代之的是,它将影响诸如RNN层之类的后续层的行为。

如果检查Embedding层的源代码,则会看到一个称为的方法compute_mask

def compute_mask(self, inputs, mask=None):
    if not self.mask_zero:
        return None
    output_mask = K.not_equal(inputs, 0)
    return output_mask
Run Code Online (Sandbox Code Playgroud)

此输出掩码将作为mask参数传递到支持掩码的以下层。这已经在__call__基本层的方法中实现了Layer

# Handle mask propagation.
previous_mask = _collect_previous_mask(inputs)
user_kwargs = copy.copy(kwargs)
if not is_all_none(previous_mask):
    # The previous layer generated a mask.
    if has_arg(self.call, 'mask'):
        if 'mask' not in kwargs:
            # If mask is explicitly passed to __call__,
            # we should override the default mask.
            kwargs['mask'] = previous_mask
Run Code Online (Sandbox Code Playgroud)

这使得下面的层可以忽略(即在其计算中不考虑)此输入步骤。这是一个最小的示例:

data_in = np.array([
  [1, 0, 2, 0]
])

x = Input(shape=(4,))
e = Embedding(5, 5, mask_zero=True)(x)
rnn = LSTM(3, return_sequences=True)(e)

m = Model(inputs=x, outputs=rnn)
m.predict(data_in)

array([[[-0.00084503, -0.00413611,  0.00049972],
        [-0.00084503, -0.00413611,  0.00049972],
        [-0.00144554, -0.00115775, -0.00293898],
        [-0.00144554, -0.00115775, -0.00293898]]], dtype=float32)
Run Code Online (Sandbox Code Playgroud)

如您所见,第二和第四时间步长的LSTM层的输出分别与第一和第三时间步长的输出相同。这意味着这些时间步伐已被掩盖。

更新:由于在内部增加了损失函数以支持使用weighted_masked_objective以下方法进行掩蔽,因此在计算损失时也会考虑使用掩码:

def weighted_masked_objective(fn):
    """Adds support for masking and sample-weighting to an objective function.
    It transforms an objective function `fn(y_true, y_pred)`
    into a sample-weighted, cost-masked objective function
    `fn(y_true, y_pred, weights, mask)`.
    # Arguments
        fn: The objective function to wrap,
            with signature `fn(y_true, y_pred)`.
    # Returns
        A function with signature `fn(y_true, y_pred, weights, mask)`.
    """
Run Code Online (Sandbox Code Playgroud)

编译模型时

weighted_losses = [weighted_masked_objective(fn) for fn in loss_functions]
Run Code Online (Sandbox Code Playgroud)

您可以使用以下示例对此进行验证:

data_in = np.array([[1, 2, 0, 0]])
data_out = np.arange(12).reshape(1,4,3)

x = Input(shape=(4,))
e = Embedding(5, 5, mask_zero=True)(x)
d = Dense(3)(e)

m = Model(inputs=x, outputs=d)
m.compile(loss='mse', optimizer='adam')
preds = m.predict(data_in)
loss = m.evaluate(data_in, data_out, verbose=0)
print(preds)
print('Computed Loss:', loss)

[[[ 0.009682    0.02505393 -0.00632722]
  [ 0.01756451  0.05928303  0.0153951 ]
  [-0.00146054 -0.02064196 -0.04356086]
  [-0.00146054 -0.02064196 -0.04356086]]]
Computed Loss: 9.041069030761719

# verify that only the first two outputs 
# have been considered in the computation of loss
print(np.square(preds[0,0:2] - data_out[0,0:2]).mean())

9.041070036475277
Run Code Online (Sandbox Code Playgroud)


小智 8

通知模型数据的某些部分实际上是 Padding 且应被忽略的过程称为Masking

input masksKeras模型中有3种引入方式:

  1. 添加一层keras.layers.Masking
  2. 配置一个keras.layers.Embedding图层mask_zero=True
  3. 当调用支持该参数的层(例如 RNN 层)时,手动传递掩码参数。

Input Masks下面给出的是介绍使用的代码keras.layers.Embedding

import numpy as np

import tensorflow as tf

from tensorflow.keras import layers

raw_inputs = [[83, 91, 1, 645, 1253, 927],[73, 8, 3215, 55, 927],[711, 632, 71]]
padded_inputs = tf.keras.preprocessing.sequence.pad_sequences(raw_inputs,
                                                              padding='post')

print(padded_inputs)

embedding = layers.Embedding(input_dim=5000, output_dim=16, mask_zero=True)
masked_output = embedding(padded_inputs)

print(masked_output._keras_mask)
Run Code Online (Sandbox Code Playgroud)

上述代码的输出如下所示:

[[  83   91    1  645 1253  927]
 [  73    8 3215   55  927    0]
 [ 711  632   71    0    0    0]]

tf.Tensor(
[[ True  True  True  True  True  True]
 [ True  True  True  True  True False]
 [ True  True  True False False False]], shape=(3, 6), dtype=bool)
Run Code Online (Sandbox Code Playgroud)

有关更多信息,请参阅此Tensorflow 教程