Tensorflow中的InvalidArgumentError

Question

I`m trying to create neural network using Tensorflow tools.

sizeOfRow = len(data[0])
x = tensorFlow.placeholder("float", shape=[None, sizeOfRow])
y = tensorFlow.placeholder("float")

def neuralNetworkTrain(x):
  prediction = neuralNetworkModel(x)
  # using softmax function, normalize values to range(0,1)
  cost = tensorFlow.reduce_mean(tensorFlow.nn.softmax_cross_entropy_with_logits(prediction, y))

this is a part from the net I have got error:

InvalidArgumentError (see above for traceback): logits and labels must be same size: logits_size=[500,2] labels_size=[1,500]
 [[Node: SoftmaxCrossEntropyWithLogits = SoftmaxCrossEntropyWithLogits[T=DT_FLOAT, _device="/job:localhost/replica:0/task:0/cpu:0"](Reshape, Reshape_1)]]

someone know what`s wrong?

edit: also I have got from this code:

for temp in range(int(len(data) / batchSize)):
    ex, ey = takeNextBatch(i) # takes 500 examples
    i += 1
    # TO-DO : fix bug here
    temp, cos = sess.run([optimizer, cost], feed_dict= {x:ex, y:ey}) 

this error TypeError: unhashable type: 'list'

Answer 1

Well, the error is quite self-describing.

logits and labels must be same size: logits_size=[500,2] labels_size=[1,500]

So, first, your labels should be transposed to have size 500, 1 and second, the softmax_cross_entropy_with_logits expects labels to be presented in a form of a probability distribution (eg [[0.1, 0.9], [1.0, 0.0]] ).

If you know your classes are exclusive (which is probably the case), you should switch to using sparse_softmax_cross_entropy_with_logits .