Question

我想将我的keras模型转换为theano函数，以便我可以计算输入的渐变。我认为这对于可视化网络来说可能很酷。我想使用这些渐变来增强原始图像中基于神经网络认为的特征。我不明白以下代码我做错了什么。

model = Sequential()
model.add(InputLayer((3, H, W)))
model.add(GaussianNoise(0.03))

model.add(Flatten())
model.add(Dense(512, activation = 'relu', name = 'dense'))
model.add(Dropout(0.2))
model.add(Dense(20, activation = 'relu'))
model.add(Dense(C, activation = 'softmax', W_regularizer = l2()))
...
f = theano.function([model.input], model.output)

我收到以下异常。

theano.gof.fg.MissingInputError: A variable that is an input to the graph was neither provided as an input to the function nor given a value. A chain of variables leading from this input to an output is [keras_learning_phase, DimShuffle{x,x}.0, Elemwise{switch,no_inplace}.0, dot.0, Elemwise{add,no_inplace}.0, Elemwise{add,no_inplace}.0, Elemwise{mul,no_inplace}.0, dot.0, Elemwise{add,no_inplace}.0, Softmax.0]. This chain may not be unique
Backtrace when the variable is created:
  File "<frozen importlib._bootstrap>", line 222, in _call_with_frames_removed
  File "/usr/local/lib/python3.5/dist-packages/keras/backend/__init__.py", line 51, in <module>
    from .theano_backend import *
  File "<frozen importlib._bootstrap>", line 969, in _find_and_load
  File "<frozen importlib._bootstrap>", line 958, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 673, in _load_unlocked
  File "<frozen importlib._bootstrap_external>", line 662, in exec_module
  File "<frozen importlib._bootstrap>", line 222, in _call_with_frames_removed
  File "/usr/local/lib/python3.5/dist-packages/keras/backend/theano_backend.py", line 13, in <module>
    _LEARNING_PHASE = T.scalar(dtype='uint8', name='keras_learning_phase')  # 0 = test, 1 = train

Answer 1

在FAQ之后，尝试：

from keras import backend as K
get_last_layer_output = K.function([model.layers[0].input],
                                   [model.layers[-1].output])

对于最新版本的Keras（1.0），请使用

from keras import backend as K
get_last_layer_output = K.function([model.layers[0].input],
                                   [model.layers[-1].get_output(train=False)])

Answer 2

对于“旧”keras（例如0.3.x）：

我不使用此版本，但像this one这样的示例应该可以使用。

对于“新”keras（1.0 +）：

由于您使用Dropout图层，因此您需要添加另一个输入K.learning_phase()并为其指定值0（0表示测试，1表示培训。）

代码：

from keras import backend as K
K.function([model.layers[0].input, K.learning_phase()], [model.layers[-1].output])

参考：keras FAQ

如何将整个keras模型转换为theano函数

2 个答案: