Created by: walloollaw

when the label for fluid.layers.cross_entropy is not in the expected range,
for example when my real class num is 100, but i give it a 1000, this layer can be executed and output a large value

my test code:

import sys
import numpy as np

def get_data(shape=[1, 100]):
    sz = reduce(lambda x, y: x*y, shape)
    d = [1] * sz
    data = np.array(d, dtype='float32').reshape(shape)
    return data / sz


def test():
    import paddle.fluid as fluid

    data = get_data()

    x = fluid.layers.data(name='x', shape=data.shape[1:], dtype='float32')
    label = fluid.layers.data(name='label', shape=[1], dtype='int64')
    out = fluid.layers.cross_entropy(input=x, label=label)

    place = fluid.CPUPlace()
    exe = fluid.Executor(place)
    fetch_list = [out]

    exe.run(fluid.default_startup_program())
    program = fluid.default_main_program()

    y = exe.run(
            program=program,
            feed={'x': data, 'label': np.array(1000).reshape([1,-1])},
            fetch_list=fetch_list)

    print('results:')
    print y

test()

results:

results:
[array([[1.e+20]], dtype=float32)]