Theano（6）：Theano条件语句，IfElse vs Switch

http://deeplearning.net/software/theano/tutorial/conditions.html

IfElse vs Switch

• Both ops build a condition over symbolic variables.
• IfElse takes a boolean condition and two variables as inputs.
• Switch takes a tensor as condition and two variables as inputs. switch is an elementwise operation and is thus more general thanifelse.
• Whereas switch evaluates both output variables, ifelse is lazy and only evaluates one variable with respect to the condition.

ifelse的使用：

>>> from theano.ifelse import ifelse
>>> a,b=T.dscalars('a','b')
>>> x=T.dscalar('x')
>>> y=T.dscalar('y')
>>> z=ifelse(T.lt(a,b),x+y,x-y)
>>> ifelseFun=theano.function([a,b,x,y],z,mode=theano.Mode(linker='vm'))
>>> AltB=ifelseFun(1,2,3,4)
>>> print AltB
7.0
>>> BltA=ifelseFun(2,1,3,4)
>>> print BltA
-1.0

ifelse和switch比较：

from theano import tensor as T
from theano.ifelse import ifelse
import theano, time, numpy

a,b = T.scalars('a', 'b')
x,y = T.matrices('x', 'y')

z_switch = T.switch(T.lt(a, b), T.mean(x), T.mean(y))
z_lazy = ifelse(T.lt(a, b), T.mean(x), T.mean(y))

f_switch = theano.function([a, b, x, y], z_switch,
                           mode=theano.Mode(linker='vm'))
f_lazyifelse = theano.function([a, b, x, y], z_lazy,
                               mode=theano.Mode(linker='vm'))

val1 = 0.
val2 = 1.
big_mat1 = numpy.ones((10000, 1000))
big_mat2 = numpy.ones((10000, 1000))

n_times = 10

tic = time.clock()
for i in range(n_times):
    f_switch(val1, val2, big_mat1, big_mat2)
print('time spent evaluating both values %f sec' % (time.clock() - tic))

tic = time.clock()
for i in range(n_times):
    f_lazyifelse(val1, val2, big_mat1, big_mat2)
print('time spent evaluating one value %f sec' % (time.clock() - tic))

IfElse  op （只需要计算一个变量）花费的时间大约是  Switch  op （两个都需要）的一半。

注意：Unless linker='vm' or linker='cvm' are used, ifelse will compute both variables and take the same computation time as switch. Although the linker is not currently set by default to cvm, it will be in the near future.