In [11]:

    

# shared variable
import numpy as np
import theano
import theano.tensor as T

state = theano.shared(np.array(0, dtype=np.float64), 'state')
inc = T.scalar('inc', dtype=state.dtype)
accumulator = theano.function([inc], state, updates=[(state, state + inc)])

#get variable value
print state.get_value()
accumulator(1)
print state.get_value()
accumulator(10)
print state.get_value()


    

    




0.0
1.0
11.0

In [13]:

    

#set variable value
state.set_value(-1)
accumulator(2)
print state.get_value()


    

    




1.0

In [14]:

    

#temporarily 
tmp_func = state*2 + inc
a = T.scalar(dtype=state.dtype)
skip_shared = theano.function([inc, a], tmp_func, givens=[(state, a)])
print skip_shared(2,3)
print state.get_value()


    

    




8.0
1.0