Question

我最近在Keras构建了一个小的网格搜索功能（基本上只是一个通过几个不同的神经网络架构的循环）并遇到了一个相当奇怪的问题。

让我们举一个例子，我正在使用以下列表作为我的网格（它不是真正的网格，因为只有一个超参数，但我仍然称之为）：

layerListGrid = [] 
layerListGrid.append([5,5,5])
layerListGrid.append([10,10,10])
layerListGrid.append([100,100,100,100,100])

换句话说，我想连续三个神经网络，每次都有如上所述的密集层。

要执行此操作，请执行以下操作

#start grid loop
for layerList in layerListGrid:

    #define model structure
    init = keras.initializers.TruncatedNormal(mean=initializerMean, stddev=initializerStdDev, seed=None)    

    model = Sequential()
    firstEntry=True
    for layer in layerList:
        if firstEntry:
            model.add(Dense(layer, activation=activation, input_shape=(inputSize,), kernel_initializer=init)) #first hidden layer
        else:
            model.add(Dense(layer, activation=activation, kernel_initializer=init))
        if dropoutRate>0.0:
            model.add(Dropout(dropoutRate))
        firstEntry=False

    model.add(Dense(1, activation=finalActivation, kernel_initializer=init)) #outputLayer
    model.summary()

    #compile model
    sgd = keras.optimizers.SGD(lr=learningRate, momentum=momentum, decay=decay, nesterov=False)
    model.compile(loss=loss,
                  optimizer=sgd,
                  metrics=['accuracy'])

    #define callbacks
    startTime = datetime.datetime.now()
    startTimeText = startTime.strftime("%Y-%m-%d_%H-%M-%S")
    tensorboard = keras.callbacks.TensorBoard(log_dir='./TFLogs/'+startTimeText, histogram_freq=1, write_graph=True, write_grads=True, write_images=False)

    #fit model
    history = model.fit(trainX, trainY,
                        batch_size=batchSize,
                        epochs=epochs,
                        verbose=1,
                        validation_data=validationData,
                        validation_split=validation_split,
                        callbacks=[tensorboard]
                        )
    print('Fitting done')
    score = model.evaluate(testX, testY, verbose=0)
    print('Test loss:', score[0])
    print('Test accuracy:', score[1])

现在，第一次运行循环（即第一次“layerList”），代码正确执行。然后第二个循环训练它的第一个时期，然后在结束时终止（根据下面的错误日志，当张量板回调触发时）。然后它抛出一个错误，抱怨“dense_1”的输入格式不正确。 “dense_1”是Keras给出的第一个循环执行的第一个密集层的唯一名称！在第二个循环中，第一个密集层实际上称为“dense_4”。似乎某种内存指针在循环之外持续存在，但我无法弄清楚它是什么。有没有人遇到这样的事情，或者有人建议我如何解决这个问题？

下面是堆栈跟踪

File "<stdin>", line 1, in <module>
File "/home/username/Repo//dfc/Analytics/tf_modeling/20170919_DFC_TF_dev.py", line 174, in <module>
  callbacks=[tensorboard]
File "/home/username/.local/lib/python2.7/site-packages/keras/models.py", line 867, in fit
  initial_epoch=initial_epoch)
File "/home/username/.local/lib/python2.7/site-packages/keras/engine/training.py", line 1598, in fit
  validation_steps=validation_steps)
File "/home/username/.local/lib/python2.7/site-packages/keras/engine/training.py", line 1203, in _fit_loop
  callbacks.on_epoch_end(epoch, epoch_logs)
File "/home/username/.local/lib/python2.7/site-packages/keras/callbacks.py", line 77, in on_epoch_end
  callback.on_epoch_end(epoch, logs)
File "/home/username/.local/lib/python2.7/site-packages/keras/callbacks.py", line 777, in on_epoch_end
  result = self.sess.run([self.merged], feed_dict=feed_dict)
File "/usr/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 895, in run
  run_metadata_ptr)
File "/usr/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 1124, in _run
  feed_dict_tensor, options, run_metadata)
File "/usr/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 1321, in _do_run
  options, run_metadata)
File "/usr/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 1340, in _do_call
  raise type(e)(node_def, op, message)
tensorflow.python.framework.errors_impl.InvalidArgumentError: You must feed a value for placeholder tensor 'dense_1_input' with dtype float and shape [?,89]
 [[Node: dense_1_input = Placeholder[dtype=DT_FLOAT, shape=[?,89], _device="/job:localhost/replica:0/task:0/cpu:0"]()]]

Keras循环执行仍然在前一次迭代中寻找图层

0 个答案: