快捷導(dǎo)航

keras 自定義loss層+接受輸入實(shí)例

更新時(shí)間：2020年06月28日 14:52:35 作者：lgy_keira

這篇文章主要介紹了keras 自定義loss層+接受輸入實(shí)例，具有很好的參考價(jià)值，希望對大家有所幫助。一起跟隨小編過來看看吧

loss函數(shù)如何接受輸入值

keras封裝的比較厲害，官網(wǎng)給的例子寫的云里霧里，

在stackoverflow找到了答案

You can wrap the loss function as a inner function and pass your input tensor to it (as commonly done when passing additional arguments to the loss function).

def custom_loss_wrapper(input_tensor):
 def custom_loss(y_true, y_pred):
  return K.binary_crossentropy(y_true, y_pred) + K.mean(input_tensor)
 return custom_loss

input_tensor = Input(shape=(10,))
hidden = Dense(100, activation='relu')(input_tensor)
out = Dense(1, activation='sigmoid')(hidden)
model = Model(input_tensor, out)
model.compile(loss=custom_loss_wrapper(input_tensor), optimizer='adam')

You can verify that input_tensor and the loss value will change as different X is passed to the model.

X = np.random.rand(1000, 10)
y = np.random.randint(2, size=1000)
model.test_on_batch(X, y) # => 1.1974642
X *= 1000
model.test_on_batch(X, y) # => 511.15466

fit_generator

fit_generator ultimately calls train_on_batch which allows for x to be a dictionary.

Also, it could be a list, in which casex is expected to map 1:1 to the inputs defined in Model(input=[in1, …], …)

### generator
yield [inputX_1,inputX_2],y
### model
model = Model(inputs=[inputX_1,inputX_2],outputs=...)

補(bǔ)充知識(shí)：keras中自定義 loss損失函數(shù)和修改不同樣本的loss權(quán)重(樣本權(quán)重、類別權(quán)重)

首先辨析一下概念：

1. loss是整體網(wǎng)絡(luò)進(jìn)行優(yōu)化的目標(biāo)，是需要參與到優(yōu)化運(yùn)算，更新權(quán)值W的過程的

2. metric只是作為評(píng)價(jià)網(wǎng)絡(luò)表現(xiàn)的一種“指標(biāo)”，比如accuracy，是為了直觀地了解算法的效果，充當(dāng)view的作用，并不參與到優(yōu)化過程

一、keras自定義損失函數(shù)

在keras中實(shí)現(xiàn)自定義loss，可以有兩種方式，一種自定義 loss function, 例如：

# 方式一
def vae_loss(x, x_decoded_mean):
 xent_loss = objectives.binary_crossentropy(x, x_decoded_mean)
 kl_loss = - 0.5 * K.mean(1 + z_log_sigma - K.square(z_mean) - K.exp(z_log_sigma), axis=-1)
 return xent_loss + kl_loss
 
vae.compile(optimizer='rmsprop', loss=vae_loss)

或者通過自定義一個(gè)keras的層（layer）來達(dá)到目的，作為model的最后一層，最后令model.compile中的loss=None：

# 方式二
# Custom loss layer
class CustomVariationalLayer(Layer):
 
 def __init__(self, **kwargs):
  self.is_placeholder = True
  super(CustomVariationalLayer, self).__init__(**kwargs)
 def vae_loss(self, x, x_decoded_mean_squash):
 
  x = K.flatten(x)
  x_decoded_mean_squash = K.flatten(x_decoded_mean_squash)
  xent_loss = img_rows * img_cols * metrics.binary_crossentropy(x, x_decoded_mean_squash)
  kl_loss = - 0.5 * K.mean(1 + z_log_var - K.square(z_mean) - K.exp(z_log_var), axis=-1)
  return K.mean(xent_loss + kl_loss)
 
 def call(self, inputs):
 
  x = inputs[0]
  x_decoded_mean_squash = inputs[1]
  loss = self.vae_loss(x, x_decoded_mean_squash)
  self.add_loss(loss, inputs=inputs)
  # We don't use this output.
  return x
 
y = CustomVariationalLayer()([x, x_decoded_mean_squash])
vae = Model(x, y)
vae.compile(optimizer='rmsprop', loss=None)

在keras中自定義metric非常簡單，需要用y_pred和y_true作為自定義metric函數(shù)的輸入?yún)?shù) 點(diǎn)擊查看metric的設(shè)置

注意事項(xiàng)：

1. keras中定義loss，返回的是batch_size長度的tensor，而不是像tensorflow中那樣是一個(gè)scalar

2. 為了能夠?qū)⒆远x的loss保存到model，以及可以之后能夠順利load model，需要把自定義的loss拷貝到keras.losses.py 源代碼文件下，否則運(yùn)行時(shí)找不到相關(guān)信息，keras會(huì)報(bào)錯(cuò)

有時(shí)需要不同的sample的loss施加不同的權(quán)重，這時(shí)需要用到sample_weight，例如

discriminator.train_on_batch(imgs, [valid, labels], class_weight=class_weights)

二、keras中的樣本權(quán)重

# Import
import numpy as np
from sklearn.utils import class_weight
 
# Example model
model = Sequential()
model.add(Dense(32, activation='relu', input_dim=100))
model.add(Dense(1, activation='sigmoid'))
 
# Use binary crossentropy loss
model.compile(optimizer='rmsprop',
    loss='binary_crossentropy',
    metrics=['accuracy'])
 
# Calculate the weights for each class so that we can balance the data
weights = class_weight.compute_class_weight('balanced',
           np.unique(y_train),
           y_train)
 
# Add the class weights to the training           
model.fit(x_train, y_train, epochs=10, batch_size=32, class_weight=weights)