手寫數字識別問題：尺寸錯誤

Question

我正在嘗試訓練我的神經網絡模型來識別手寫數字，但是我被卡住了。

我面臨的問題是，當我嘗試訓練我的神經網絡時，我試圖將我的訓練數據傳遞給函數，但是當我這樣做時，出現了維數錯誤的錯誤。

我的代碼：

import theano
import lasagne as lse
import theano.tensor as T

def build_nn(input_var=None):
    l_input=lse.layers.InputLayer(shape=(None,1,28,28),input_var=input_var)
    ldrop=lse.layers.DropoutLayer(l_input,p=0.2)
    l_hid1=lse.layers.DenseLayer(ldrop,num_units=800,
                                  nonlinearity=lse.nonlinearities.rectify,
                                  W=lse.init.GlorotUniform())
    l_hid1_drop=lse.layers.DropoutLayer(l_hid1,p=0.5)

    l_hid2=lse.layers.DenseLayer(l_hid1_drop,num_units=800,
                                  nonlinearity=lse.nonlinearities.rectify,
                                  W=lse.init.GlorotUniform())
    l_hid2_drop=lse.layers.DropoutLayer(l_hid2,p=0.5)

    l_output=lse.layers.DenseLayer(l_hid2_drop,num_units=10,nonlinearity=lse.nonlinearities.softmax)

    return l_output

input_var=T.tensor4('inputs')
target_var=T.lvector('targets')

network=build_nn(input_var)
prediction=lse.layers.get_output(network)

loss=lse.objectives.categorical_crossentropy(prediction,target_var)

loss=loss.mean()

params=lse.layers.get_all_params(network,trainable=True)

update=lse.updates.nesterov_momentum(loss,params,learning_rate=1,momentum=0.9)
tain_fn=theano.function([input_var,target_var],loss,updates=update)

num_training_step=1000

for steps in range(num_training_step):
    train_err=tain_fn(x_train,y_train)
    print('Step '+str(steps))

錯誤：

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-33-2827076f729d> in <module>
      2 
      3 for steps in range(num_training_step):
----> 4     train_err=tain_fn(x_train,y_train)
      5     print('Step '+str(steps))

~\Anaconda3\lib\site-packages\theano\compile\function_module.py in __call__(self, *args, **kwargs)
    811                         s.storage[0] = s.type.filter(
    812                             arg, strict=s.strict,
--> 813                             allow_downcast=s.allow_downcast)
    814 
    815                     except Exception as e:

~\Anaconda3\lib\site-packages\theano\tensor\type.py in filter(self, data, strict, allow_downcast)
    176             raise TypeError("Wrong number of dimensions: expected %s,"
    177                             " got %s with shape %s." % (self.ndim, data.ndim,
--> 178                                                         data.shape))
    179         if not data.flags.aligned:
    180             try:

TypeError: Bad input argument to theano function with name "<ipython-input-32-e01e77ca594c>:14" at index 0 (0-based).  
Backtrace when that variable is created:

  File "C:\Users\hp\Anaconda3\lib\site-packages\ipykernel\zmqshell.py", line 536, in run_cell
    return super(ZMQInteractiveShell, self).run_cell(*args, **kwargs)
  File "C:\Users\hp\Anaconda3\lib\site-packages\IPython\core\interactiveshell.py", line 2819, in run_cell
    raw_cell, store_history, silent, shell_futures)
  File "C:\Users\hp\Anaconda3\lib\site-packages\IPython\core\interactiveshell.py", line 2845, in _run_cell
    return runner(coro)
  File "C:\Users\hp\Anaconda3\lib\site-packages\IPython\core\async_helpers.py", line 67, in _pseudo_sync_runner
    coro.send(None)
  File "C:\Users\hp\Anaconda3\lib\site-packages\IPython\core\interactiveshell.py", line 3020, in run_cell_async
    interactivity=interactivity, compiler=compiler, result=result)
  File "C:\Users\hp\Anaconda3\lib\site-packages\IPython\core\interactiveshell.py", line 3185, in run_ast_nodes
    if (yield from self.run_code(code, result)):
  File "C:\Users\hp\Anaconda3\lib\site-packages\IPython\core\interactiveshell.py", line 3267, in run_code
    exec(code_obj, self.user_global_ns, self.user_ns)
  File "<ipython-input-32-e01e77ca594c>", line 1, in <module>
    input_var=T.tensor4('inputs')
Wrong number of dimensions: expected 4, got 3 with shape (60000, 28, 28).

Answer 1

根據您的代碼，您期望將形狀為（X，1，28，28）的輸入張量

l_input=lse.layers.InputLayer(shape=(None,1,28,28),input_var=input_var)

其中1是通道數。 如果您打算僅使用一個通道（灰度），則可以將輸入重構為shape =（None，28,28），或者重構數據以使其具有額外的維度以適合輸入圖層。

Answer 2

您的代碼有兩個問題：

1. 您正在使用l_input作為（batch_size，n_channels，高度，寬度） ， x_train作為（batch_size，高度，寬度） 。 改變l_input至（的batch_size，高度，寬度）。作為@Platon攝影師：Matt建議。
2. 修復1后，您將收到另一個錯誤： DenseLayer期望輸入形狀（批處理，n_dimensions）。 為此，您需要重塑x_train並使其具有形狀（批處理，高度*寬度）：

l_input = lse.layers.InputLayer(shape=(None, 28, 28),input_var=input_var)
l_reshape = lse.layers.ReshapeLayer(l_input, ((None, 28*28))
ldrop=lse.layers.DropoutLayer(l_reshape, p=0.2)
# etc...