Question

我正在尝试可视化用千层面构建的神经网络的输出。特别是我修改了关于mnist示例的代码：https://github.com/Lasagne/Lasagne/blob/master/examples/mnist.py

在第299行我插入了以下代码行：

        input_var=inputs
        prediction=lasagne.layers.get_output(network,input_var)
        print(prediction.eval())
        sys.exit('debug')

如果我们选择模型＆＃39; mlp＆＃39;在第234行：

def main(model='mlp', num_epochs=500):

同时，选择模型＆＃39; cnn＆＃39;通过改变第234行如下：

def main(model='cnn', num_epochs=500):

行

print(prediction.eval())

给出错误：

 Traceback (most recent call last):
  File "/dos/mnist_lasagne_original.py", line 364, in <module>
main(**kwargs)
  File "/dos/mnist_lasagne_original.py", line 299, in main
print(prediction.eval())
  File "/usr/local/lib/python2.7/dist-packages/theano/gof/graph.py", line 523, in eval
rval = self._fn_cache[inputs](*args)
   File "/usr/local/lib/python2.7/dist-packages/theano/compile/function_module.py", line 871, in __call__
storage_map=getattr(self.fn, 'storage_map', None))
  File "/usr/local/lib/python2.7/dist-packages/theano/gof/link.py", line 314, in raise_with_op
reraise(exc_type, exc_value, exc_trace)
  File "/usr/local/lib/python2.7/dist-packages/theano/compile/function_module.py", line 859, in __call__
outputs = self.fn()
ValueError: CorrMM received weight with wrong type.
Apply node that caused the error: CorrMM{valid, (1, 1)}(TensorConstant{[[[[ 0.  0..0.  0.]]]]}, Subtensor{::, ::, ::int64, ::int64}.0)
Toposort index: 8
Inputs types: [TensorType(float32, (False, True, False, False)), TensorType(float64, 4D)]
Inputs shapes: [(500, 1, 28, 28), (32, 1, 5, 5)]
Inputs strides: [(3136, 3136, 112, 4), (200, 200, -40, -8)]
Inputs values: ['not shown', 'not shown']
Outputs clients: [[Elemwise{Composite{(i0 * (Abs((i1 + i2)) + i1 + i2))}}(TensorConstant{(1, 1, 1, 1) of 0.5}, CorrMM{valid, (1, 1)}.0, InplaceDimShuffle{x,0,x,x}.0)]]

Backtrace when the node is created(use Theano flag traceback.limit=N to make it longer):
   File "/dos/mnist_lasagne_original.py", line 364, in <module>
main(**kwargs)
   File "/dos/mnist_lasagne_original.py", line 298, in main
prediction=lasagne.layers.get_output(network,input_var)
  File "/home/paul/src/lasagne/lasagne/layers/helper.py", line 185, in get_output
all_outputs[layer] = layer.get_output_for(layer_inputs, **kwargs)
  File "/home/paul/src/lasagne/lasagne/layers/conv.py", line 257, in get_output_for
conved = self.convolve(input, **kwargs)
  File "/home/paul/src/lasagne/lasagne/layers/conv.py", line 535, in convolve
filter_flip=self.flip_filters)

我有很多谷歌，我无法弄清楚这个问题的根源。我有兴趣可视化神经网络的输出，以了解它是如何工作的。任何帮助将不胜感激。

Answer 1

在这里阅读：http://lasagne.readthedocs.io/en/latest/user/layers.html#propagating-data-through-layers

我找到了以下解决方案（那些调试行插入到原始代码的第299行）：

        x = theano.tensor.tensor4('x')
        y = lasagne.layers.get_output(network, x)
        f = theano.function([x], y)
        output=f(inputs)
        print(output)
        sys.exit('debug')

theano - 评估神经网络的输出

1 个答案: