How to reuse Anano function with different shared variables without rebuilding the chart?

Question

I have an Anano function that is called several times, each time with different shared variables. The way it is currently implemented, the Theano function becomes overridden each time it starts. I suppose this makes the whole program slow, because every time the Theano functions are detected, the graph is restored.

def sumprod_shared(T_shared_array1, T_shared_array2): f = theano.function([], (T_shared_array1 * T_shared_array2).sum(axis=0)) return f() for factor in range(10): m1 = theano.shared(factor * array([[1, 2, 4], [5, 6, 7]])) m2 = theano.shared(factor * array([[1, 2, 4], [5, 6, 7]])) print sumprod_shared(m1, m2) 

For non-general (normal) variables, I can define a function once, and then call it with different variables without redefinition.

 def sumprod_init(): T_matrix1 = T.lmatrix('T_matrix1') T_matrix2 = T.lmatrix('T_matrix2') return theano.function([T_matrix1, T_matrix2], (T_matrix1 * T_matrix2).sum(axis=0)) sumprod = sumprod_init() for factor in range(10): np_array1 = factor * array([[1, 2, 4], [5, 6, 7]]) np_array2 = factor * array([[1, 2, 4], [5, 6, 7]]) print sumprod(np_array1, np_array2) 

Is this also possible for shared variables?

Answer 1

You can use the givens keyword in anano.function to do this. Basically, you do the following.

 m1 = theano.shared(name='m1', value = np.zeros((3,2)) ) m2 = theano.shared(name='m2', value = np.zeros((3,2)) ) x1 = theano.tensor.dmatrix('x1') x2 = theano.tensor.dmatrix('x2') y = (x1*x2).sum(axis=0) f = theano.function([],y,givens=[(x1,m1),(x2,m2)],on_unused_input='ignore') 

then for cyclic values, you simply set the value of the shared variables to the value you want. You must set on_unused_input to "ignore" in order to use functions without arguments in anano, by the way. Like this:

 array1 = array([[1,2,3],[4,5,6]]) array2 = array([[2,4,6],[8,10,12]]) for i in range(10): m1.set_value(i*array1) m2.set_value(i*array2) print f() 

It should work, at least the way I worked on it.

Answer 2

It is currently not possible to reuse the Anano function with another shared variable.

But you have an alternative:

• Is this really a bottleneck? This is true in the example, but I suppose this is a simplified case. The only way to find out is with a profile.
• You compile 1 Anano function with the first common variable. You can then call get_value / set_value for these shared variables before calling the Anano function. This way you will not need to recompile the Theano function.