1. 出现Elemwise{xxx,no_inplace}.0
这是因为没有定义theano.function所致，参考下面错误示范：
y = np.random.normal(size=(2,2))
z1 = T.round(y, mode='half_to_even')
print(z1)
# 输 出
Elemwise{round_half_to_even,no_inplace}.0 正确操作如下：
x = T.dmatrix('x')
y = np.random.normal(size=(2,2))
z = T.round(x, mode='half_to_even')
f = function([x], z)
print(f(y))
# 输 出
[[ 0., 1.],
[-1., -1.]]2. tensor.round(a, mode='half_to_even')函数学习
其实就是对a近似，离哪个整数值近就近似成那个值，有2种模式(我的theano版本默认为half_to_even)，具体区别见下面代码:
# -*- coding:utf-8 -*-
# Theano learning
# 文档位置：http://deeplearning.net/software/theano/library/tensor/basic.html?highlight=round#theano.tensor.round
import theano
import numpy as np
import theano.tensor as T
from theano import function
# version 1.0.4
if __name__ == '__main__':
"""
theano.tensor.round(a, mode='half_to_even')
mode='half_to_even'(default) or 'half_away_from_zero'
"""
x = T.dmatrix('x')
y = np.random.normal(size=(2,2))
z1 = T.round(x, mode='half_to_even')
z2 = T.round(x, mode='half_away_from_zero')
f1 = function([x], z1)
f2 = function([x], z2)
print(y)
print(f1(y))
print(f2(y))
# 输 出
# 试了好几个输入，可以看出2种模式近似几乎没区别，只有在近似到0时会出现-0这样
[[-0.06255656 -0.44239336]
[-2.25829238 -0.23422771]]
[[ 0. 0.]
[-2. 0.]]
[[-0. -0.]
[-2. -0.]]